HISTORY

Prehistory
----------

We designed the Mercury execution algorithm in October 1993. We started
working on a Mercury compiler in December 1993. Semantic analysis
started working around May 1994. We started generating code around
August 1994; we started work on optimizations very soon after. The
compiler successfully compiled itself on 24 February 1995.


Mercury 0.1, April 8 1995
-------------------------
Initial beta-test release. 
Very resource-hungry, not very well documented.


Mercury 0.2, April 18 1995
--------------------------
Much better error messages for determinism errors, much
faster compilation, much lower memory requirements for bootstrapping.
The C compilation is much faster and requires much less memory since we
now by default compile to one C function per procedure, but we also
improved the speed of the Mercury compiler itself by nearly 40% by
implementing some library predicates much more efficiently.


Mercury 0.2.5, 1 June 1995
--------------------------
Much better syntax error messages.
Better documentation, now including a library reference manual.
Added a GNU autoconf configuration script.
Ported to IRIX 5.
Added `multidet'.  
Enabled the use of tag bits in combination with conservative garbage
collection (this improved the speed of the compiler by about 20%).
Compile directly to C rather than via the old .mod files, which were
post-processed by a Perl script (improves compilation speed, reduces
disk space requirements, removes dependency on Perl).
Lots of bug fixes.


Mercury 0.3, 18 July 1995
-------------------------
The first public release.
Better type error messages.
Better determinism error messages.
Only recompiles <module>_init.c after `mmake depend', rather than after
anything changes.
Ported to ULTRIX (N.B. - ULTRIX's /bin/sh sucks).
Avoid saving variables on the stack before negated contexts.
Don't embed label names as strings in the executable.
A few other efficiency improvements.
Lots of bug fixes.
Made the rule for `mmake clean' less reckless (don't do `rm -f *.c').
Rationalized the options to `mc'.  Implemented a couple of new ones.
Added a symbol demangler to improve linker error messages.
Made very significant improvements to the documentation.
Added a "Prolog to Mercury transition guide".


Mercury 0.4, 14 September 1995
------------------------------

* Higher-order predicates and lambda expressions are now implemented.
  (This means that `call/{1,2,3,4}' and `solutions/2' are now usable;
  unfortunately call/{5,6,...} are still not yet implemented.)
* Unique modes are now partially implemented (but don't use them
  for anything except I/O, as the implementation is not yet complete).
* Partially instantiated modes are now closer to being fully
  implemented.
* The installation process is more standard (basically
  just `configure; make; make install').
* Autoconfiguration is a bit more robust.
* `msc' and `mnc' now produce somewhat smaller object files.
* Error and warning messages are a little better in some cases.
* Fixed a few code generation bugs.
* Ported to DEC Alpha/OSF and BSDI BSD/386.
* We've improved the efficiency of the 386 port by about 70%.
  (50% because asm_fast.gc mode now works on the 386, the rest
  due to better optimization).
* We generate better code for calls to `\='.
* We generate better code for `compare/3'.
* A few other new optimizations.
* The profiler now works (see the documentation in the Mercury
  User's Guide).
* Some new library predicates, including `string__format'
  (like C's sprintf).
* `set__join/2' has been renamed as `set__power_union/2'.
* `list__sort/2' has been renamed as `list__sort_and_remove_dups/2'.
* There is a new `list__sort/2' which does not remove duplicates.


Mercury 0.5, 15 February 1996
-----------------------------

* We now support committed choice nondeterminism in a logical and declarative
  fashion, using the new determinism categories `cc_nondet' and `cc_multi'.
  Like `nondet' and `multi' respectively, they specify that a predicate
  may have more than one solution, but they convey the additional
  assertion that the predicate will only be used in contexts in which
  only one solution is needed.  The compiler will check that all uses
  of the predicate satisfy this requirement.  Having proven the
  assertion to be correct, the compiler can then generate much more
  efficient code for the predicate.  By pushing pruning inwards, the
  compiler can often avoid creating choice points at all.

* We now check for backtracking over unique modes.
  (This may potentially break some programs using unique modes in ways
  that the compiler can't prove are safe.  In such cases, replacing
  `multi' with `cc_multi' should solve the problem.  If you have any
  trouble with this, we'll be happy to help you.)
  We have also added "mostly unique" modes, which provide support
  for backtrackable destructive update.
  See the Mercury Language Reference Manual.

* We now provide genuine arrays with destructive update.
  See the library module `uniq_array'.  (Warning: this has not had
  much testing.  The interface is not yet stable.)

* We now support interfacing to C code.
  See the documentation in the Mercury Language Reference Manual.

* There is now an `inline' pragma which you can use as a hint to the
  compiler to inline a particular predicate.

* We've ported the system to ULTRIX (thanks to Gertjan van Noord
  and especially Eero Pajarre).

* We now support shared libraries for IRIX 5.

* We now allow the use of compilers other than gcc -
  see the user's guide for details.
  We don't recommend the use of compilers other than gcc,
  since the inability to use gcc's special features will
  most likely lead to much less efficient code.

* To complement our source distribution, we now also provide binary
  distributions for a variety of platforms.  Installation should be
  quick and easy.

* Various other minor improvements:
  - In portable C mode, we now generate better code for loops.
  - We've made a few other minor improvements in the generated code.
  - Unary plus and minus are now implemented.
  - Updated the documentation to reflect changes in unique modes.
  - Corrected a lot of typos in the documentation.
  - Fixed quite a few bugs.

* Parts of the library module `std_util' have been moved into separate
  modules `assoc_list' and `bool'; if you have existing code which
  used those parts of `std_util', you may need to add `import_module'
  declarations to import `assoc_list' and/or `bool'.


Mercury 0.6, 2 August 1996
--------------------------

* We now provide preliminary support for type and mode inference.
  
  The `:- pred' and `:- mode' declarations are now optional for
  predicates that are local to a module; the compiler will infer
  the types and modes automatically.

  This support is not yet complete, and so type and mode inference are
  not enabled by default.  They can be enabled using the `--infer-types'
  and `--infer-modes' options.  The limitations of the current
  implementation are: (1) type inference loops for certain very rare
  type-incorrect programs; (2) mode inference doesn't support
  reordering; (3) there are some modes which the current mode inference
  algorithm is not capable of inferring.

  Note that omitting type and mode declarations can significantly
  reduce the quality of the compiler's error messages; if you have
  trouble understanding an error message, then it is a good idea to add
  explicit type and mode declarations.

* We now support functional syntax.

  Functions can be declared with `:- func' declarations,
  and defined using equations or conditional equations,
  in a similar manner to the way predicates are declared
  with `:- pred' declarations and defined using facts and rules.
  Terms can contain function calls and if-then-elses.
  For example:

	:- func sum(list(int)) = int.	
	sum([]) = 0.				% unconditional equations
	sum([X|Xs]) = X + sum(Xs).		% using function calls

	:- func max(int, int) = int.
	max(X, Y) = Max :-			% conditional equation
		(X >= Y -> Max = X ; Max = Y).

	:- func min(int, int) = int.
	min(X, Y) = (if X >= Y then X else Y).	% unconditional equation
						% using if-then-else expression

  By default, functions are assumed to have a single det mode with
  all arguments input and the result output, but you can override
  this default by supplying explicit mode declarations; for example,
  this allows you to use functions in a backwards mode to compute the
  function inverse. 

  Zero-arity functions can be used to define constants.

	:- func pi = float.
	pi = 3.14159265359.

  We also support higher-order functions and function lambda expressions.
  See the Mercury Language Reference Manual for details.
  
  The support for functions is not quite complete; there is one minor
  limitation and one major one.  The minor limitation is that the C
  interface does not yet support Mercury functions.  (Fortunately there
  is an easy work-around - since the C interface does support predicates,
  you can just make your Mercury function call a Mercury predicate.)
  The major limitation is that there is currently no support for
  debugging programs using functional syntax; obviously using a Prolog
  system won't work, since Prolog doesn't support functions.

* We now support overloading of predicates (and functions) with the
  same name and arity defined in different modules. 
  
  Previously, an explicit module qualifier was required to resolve the
  ambiguity; now, the typechecker will use the types to resolve the
  ambiguity, if possible.  (Note that you cannot define more than one
  predicate with the same name and arity in the same module.  Allowing
  that would add considerable complexity to the syntax, and would be
  of arguable utility, so we do not anticipate lifting that restriction.)

* We now support defining types with the same name in different modules.

  Previously this resulted in multiply defined symbol errors in certain
  circumstances, e.g. when profiling.

* We have removed the limitations on the number and order of arguments in
  higher-order predicate calls.

* Support for floating point is much improved.
  
  The above-mentioned support for functional syntax and overloading
  have enabled major improvements in the interface used for floating
  point.  You can now use the usual arithmetic and comparison
  operators, rather than the cumbersome builtin_float_plus (etc.)
  predicates that were used previously.  We have also improved code
  generation for floating point operations, by avoiding many
  unnecessary boxing/unboxing operations, and by making the
  `--static-ground-terms' optimization apply also to floating point
  constants.  These two optimizations improved performance on the
  realistic "pseudoknot" benchmark by nearly a factor of four.
  (There's still plenty of room for performance improvements, however.)

* We now support reverse mode arithmetic.

  This means that a goal such as `A is B + C' can be used not only to compute
  A from B and C, but also to compute B from A and C, or to compute C from
  A and B.

* We have added some new compiler options to simplify the choice of which
  optimization options to enable.

  - You can now use `-O<n>' to select an optimization level between -1 and 6.
    The default level is `-O2'; if you want to minimize compilation
    time, compile with `-O0', which disables all optimizations except those
    that actually improve overall compilation speed.

  - You can now use `--optimize-space' to select optimization for space,
    instead of optimization for time (which is the default).

* We have continued to improve the compiler's optimization.

  As well as the above-mentioned improvements to float-point code generation:

  - We now specialize calls to higher-order predicates within the same module
    in the case when the higher-order arguments have known values.
    This optimization is enabled by the `--optimize-higher-order' option.

  - We now specialize calls to predicates within the same module which have
    unused arguments.  This often happens for polymorphic predicates, since
    the compiler introduces type_info arguments which are often unused.
    This optimization is enabled by the `--optimize-unused-args' option.

  - The `--inlining' option now governs the settings of three separate options.
    One, `--inline-simple', reproduces the previous behavior of `--inlining':
    it inlines procedures whose definition is simple (e.g. a conjunction of
    builtins). Another, `--inline-single-use', tells the compiler to inline
    all procedures that are called exactly once. The compiler can also
    inline procedures called more than once as long as they are not too
    big; the argument of the option `--inline-threshold' sets a limit
    on the size of the procedure definition (roughly in terms of the number
    of logical connectives it has) multiplied by the number of calls to
    the predicate.

  - There's a new option `--optimize-dead-proc' option to eliminate unused
    procedures within a module.  This is useful even if the original source
    code didn't contain any unused procedures, since inlining and code
    specialization can make a procedure unused even if there original source
    had calls to that procedure.

  - There's a new option `--split-c-files' which causes the compiler to put
    each C function in its own C file, so that the linker can link in only
    what is used, and a new Mmake target `foo.split' for
    building a version of `foo' compiled with `--split-c-files'.
    (`--split-c-files' has much the same effect as `--optimize-dead-proc',
    except that it works globally, not just within a module.)
    On platforms for which we don't support shared libraries, installing
    a split version of the Mercury library fixes the problem of dumb Unix
    linkers dragging in nearly the entire library even if most of it is
    unused, and so reduces the size of a hello world program from >400k
    to around 120k.

  - The code generator generates special code for dense switches in
    which all the output variables are constant terms.  Instead of
    generating dense jump tables, it puts the answers in static constant
    arrays, and retrieves them using array indexing, avoiding
    an expensive jump and reducing code size.

  - The code generator now emits better code for constructing deeply nested
    terms, and avoids some unnecessary register shuffling.

  - The value numbering optimization now processes code sequences
    containing several heap allocations.

  - The `--pred-value-number' option now works.  If it is given, and
    value numbering notices that a value is available from a location
    that is faster to access than the one from which another code
    sequence retrieves the value, this other code sequence will be
    reoptimized taking this into account.

* We now support a C-to-Mercury interface, ie. we allow C code to call
  Mercury predicates, using a new `pragma export' declaration.
  (Unfortunately, this is not yet documented.  There is however a
  fairly detailed and reasonably well-documented example in the
  samples/c_interface/c_calls_mercury directory.)

* We have added a little bit of inline assembler code for the Alpha so
  that we can support the use of gcc non-local gotos and asm labels.
  This improved both code size and execution speed for the Alpha port
  by about 20-25%.

* We have ported the Mercury implementation to RS/6000 systems running AIX.
  (Thanks to Andreas Kuehlmann.)

* We have made a few changes to the Mercury standard library.
  The changes are listed here, but see the Mercury Library Reference Manual
  for details such as documentation on the new predicates.

  - The getopt library is now more flexible.  It can handle command
    line switches (such as `-O2') that affect the values of many
    options, and it can handle long options that specify a value after
    an equal sign, in the GNU style (e.g. `--optimization-level=2').

  - We have added a few new predicates using higher order facilities:
	list__map/3, list__filter/3, list__filter/4, list__foldl/4,
	list__foldr/4, list__sort/4 and list__merge/4 in list.m and
	maybe_pred/3 in std_util.m. 

  - There are a couple of new all-solutions predicates in std_util.m:
    solutions_set/2 and unsorted_solutions/2. 

  - We have added some additional predicates for handling association lists:
    assoc_list__search/3 and assoc_list__remove/4.

  - There are a few new predicates for converting floats to ints:
    float__ceiling_to_int/2, float__floor_to_int/2
    float__round_to_int/2, and float__truncate_to_int/2.

  - There are quite a few changes to the graph module.  The interface has been
    made more consistent with the rest of the library.  The predicate
    graph__lookup_node/3 has been replaced by two predicates:
    graph__search_node/3 which is nondet and returns all the nodes with
    matching data, and graph__find_matching_nodes/3 which is det and
    returns all the nodes with matching data.

  - We have renamed the varset__lookup predicates to varset__search in order to
    conform to our general naming convention.

  - We have removed the (undocumented) library predicate list__map_maybe/3.

  - The set module is now implemented using sorted lists rather than
    unsorted lists.  (The interface is still the same.)  The old
    unsorted lists representation is still available, in the
    set_unordlist module.


Mercury 0.7, 15 August 1997
---------------------------

* The Mercury language now supports higher-order syntax.

  You can now write `P(X)' as an alternative to `call(P, X)'
  or `F(X)' as an alternative for `apply(F, X)'.

* Module qualifiers are now optional.

  You can use just plain `write_string' rather than `io__write_string'.

* There is a new `:- use_module' directive.

  This is the same as `:- import_module', except all uses of the imported
  items must be explicitly module qualified.

  More changes to the module system are expected in the future,
  possibly including changing the module qualifier operator to `.'.
  Currently either `:' or `__' can be used as module qualifiers,
  but we advise you to stick with using only `__' rather than `:',
  to avoid problems in the future if we do change the module
  qualifier to `.'.

* We've improved the C interface.

  The C interface now handles Mercury functions properly --
  previously it only handled predicates, not functions. 
  Also, exporting semidet predicates or functions to C now works.
  We've improved the documentation, and we've included some examples
  of how to use the C interface to interface with C++.
  We also now support `main' being defined in C rather than in Mercury.

  See samples/c_interface for examples of all of the above.

* We now support cross-module optimizations.

  The `--intermodule-optimization' option enables cross-module inlining
  and cross-module specialization of higher-order predicates.
  Also `--intermod-unused-args' enables cross-module elimination of
  unused input arguments.

* We've continued to improve the quality of the code we generate.

  We now use a more efficient argument-passing convention, and the code
  we generate for polymorphic predicates uses a more efficient "type-info"
  representation than previous versions. 

  (Note that this means code generated by Mercury 0.7 is not compatible
  with code generated by earlier versions, so you will need to
  recompile any existing Mercury object files or libraries when you
  install the new version.)

  We handle floating point code a bit better.  We don't box floating
  point values on 64-bit architectures anymore, and on 32-bit
  architectures we do a better job of avoiding unnecessary box/unbox
  operations.  We also make some use of floating point registers for
  holding temporary values.

  We've made several improvements to the code generator that result in
  better code in common situations.

  There's also a new optimization option, `--inline-alloc', which can
  speed up code that does a lot of memory allocation by inlining the
  GC_malloc() function.  (This option is also enabled by `-O6'.)

* We now support ELF shared libraries on Linux.

  See README.Linux for details.

  Note that using shared libraries is not yet the default,
  so if you want to take advantage of this, you must explicitly
  enable it as described in README.Linux.

* We have added support for very large tables of facts.

  See the documentation for `pragma fact_table' in the
  "implementation-dependent pragmas" section of the Mercury
  Language Reference Manual.

* We have fixed quite a few bugs.

  Mode inference now works a little bit better. 
  
  We now allow a function of arity N to coexist with a predicate of
  arity N+1.

  The Mercury `char' type is now 8-bit clean (previously, "for
  compatibility with NU-Prolog" we only supported 7-bit characters).

* The `mc' script has been renamed `mmc'.

  This was done to avoid name clashes with the Midnight Commander
  and the Modula Compiler.

* We've added `man' pages.

  The documentation now includes Unix-style `man' pages for
  most of the development tools, including mmake, mmc, mgnuc, ml,
  and mprof.  These supplement the existing documentation in the
  Mercury User's Guide.

  Most of the information in the man pages is also available using
  the standard `--help' option.

* We've improved the compiler's diagnostics a bit.

  Some of the compiler's error messages are a bit more informative, and
  it catches some errors that previously it missed (such as specifying
  modes in some but not all of the arguments of a `:- pred' declaration).

* We have made quite a few changes to the Mercury standard library.

  The changes are listed here, but see the Mercury Library Reference Manual
  for details such as documentation on the new predicates.

  - The std_util.m module now contains functions and predicates for
    traversing and constructing terms of arbitrary type, and for
    accessing types at runtime.

    	+ For traversing terms: 
    		Functions argument/3, det_argument/3, functor/3,
		and predicate deconstruct/4.  These are similar to
		Prolog's arg/3, functor/3, and '=..'.

	+ For constructing terms:
		Functions num_functors/1, construct/3 and
		predicate get_functor/5.

	+ For accessing and constructing types:
		Functions type_of/1, type_ctor/1, type_args/1,
		type_ctor_name/1, type_ctor_arity/1, make_type/2,
		and predicates type_ctor_and_args/3 and
		type_ctor_name_and_arity/3.

    There are also some new functions for accessing values of the
    universal type `univ', namely univ/2 and univ_type/1.

  - There is a new module called `prolog' which contains some predicates that
    may be useful for compatibility with Prolog: arg/3, functor/3,
    `=:=', `=\=', `==', `\==', `@<', `@>', `@=<', `@>='.  We plan to
    eventually move the definitions of cut (`!') and `is' here too.
    
  - We've finally implemented generic input-output predicates,
    namely io__print/3, io__write/3, and io__read/3, using the
    functions and predicates described above.  These can read or write
    data of any type.  We've also added io__nl/3 to print a newline.
    Together with the change to make module qualifiers optional, these
    changes make performing output quite a bit simpler; it's nice to be
    able to write `print("Foo = "), print(Foo), nl'.

  - We've also added generic predicates io__write_binary/3 and
    io__read_binary/3, for doing binary I/O on values of any type.
    (The current implementations actually just call io__read/3 and
    io__write/3 respectively, though, so they're not maximally efficient.)

  - The predicates term_to_type/2 and type_to_term/2, which convert
    values of any type to or from type `term', are now implemented.

  - We have a new module called benchmarking.m to make benchmarking easier.
    The predicate report_stats, which used to be in std_util, is now
    in this module.

  - The interface to the relation.m module has been changed extensively.
    Elements must now be explicitly added to the domain of the relation,
    using relation__add_element/4, and relation operations such as
    relation__add are now performed on relation_keys.  There are also
    four new operations which convert elements to relation_keys and
    vice versa:
	relation__search_element/3, relation__lookup_element/3,
	relation__search_key/3, and relation__lookup_key/3

  - We changed the order of the arguments to set_bbbtree__subset,
    for consistency with the order in set__subset and elsewhere.
    We also changed the implementation of set__subset and
    set_ordlist__subset to match the behaviour specified in the
    documentation.

  - We made some extensive additions to bag.m to include the standard set
    operations (union, intersection, subtraction), and some other predicates
    for manipulating bags.  We also changed bag__contains/2 (swapped the 
    arguments), and bag__remove (now semidet) to be consistent with set.m 
    and map.m. 

  - There are two new predicates io__tmpnam and io__remove_file,
    with semantics similar to the ANSI C functions tmpnam() and remove().

  - There are new predicates int__max_int, int__min_int, int__bits_per_int,
    char__min_char_value, and char__max_char_value, with semantics similar
    to INT_MAX, INT_MIN, (CHAR_BIT * sizeof(int)), CHAR_MIN, and CHAR_MAX
    in ANSI C (respectively).

  - We've added list__merge_and_remove_dups/4 and list__sort_and_remove_dups/4
    to complete the set of list__merge and list__sort operations.

  - We've added io__write_list/5 and io__write_list/6; these predicates write
    lists using a user-specified procedure to write the elements and separating
    the elements with a user-specified separator string.

  - We've added io__read_file/{3,4} and io__read_binary_file/{3,4} which read
    whole files (until error or eof).

  - We've added a double accumulator version of list__foldl/4 called
    list__foldl2/6, which is a convenient generalisation for accumulators
    that also do I/O.  Also, we've added list__map_foldl/5, which is an
    amalgam of list__map/3 and list__foldl/4.

  - We've added a new constructor `maybe_string' to getopt's option_data
    type, for parsing optional string-valued command-line arguments. 
    See library/getopt.m for details.  Also added to getopt are some
    missing option-lookup predicates: getopt__lookup_accumulating_option/3
    and getopt__lookup_maybe_string_option/3.

  - We've added string__foldl to the library.  It has the same semantics as
    (string__to_char_list(String, Chars), list__foldl(Pred, Chars, Acc0, Acc))
    but is implemented more efficiently.

  - We've cleaned up the handling of integer division and modulus/remainder.
    Previously the semantics of `//' and `mod' were not really well defined.
    The semantics of `//' and `mod' have now been clarified and there are
    new functions `div' and `rem'.  `//' truncates towards zero, and `rem'
    is remainder with respect to `//', whereas `div' truncates towards minus
    infinity, and `mod' is remainder with respect to `div'.

  - The old array.m module has been renamed bt_array.m (short for
    "backtrackable array", or "binary tree array"), and uniq_array.m
    has been renamed array.m.  The interfaces of both modules have been
    extended to make them closer to each other.

    The following predicates have been added to array.m (formerly
    uniq_array.m):

	+ array__shrink/3: this is similar to array__resize/4 except
	  that it's designed for cases when you only want to make an
	  array smaller, so you don't have to supply a filler element.

	+ array__min/2, array__bounds/3: find the lower bound or both
	  bounds (respectively) of an array.  (In this implementation,
	  the lower bound is always 0.)

    The following predicates have been added to bt_array.m (formerly
    array.m):

	+ bt_array__min/2, bt_array__max/2, bt_array__size/2: find
	  the lower bound, upper bound and size of a bt_array
	  respectively.
	
	+ bt_array__in_bounds/2: check if an index is within the
	  bounds of a bt_array.
	
	+ bt_array__semidet_set/4: the semidet version of bt_array__set/4.

        + bt_array__from_list/3: a replacement for bt_array__from_list/2,
	  which has been removed.  The extra argument is the lower bound
	  for the new bt_array.

	+ bt_array__shrink/4: analogous to array__shrink/3.

	+ bt_array__resize/5: a replacement for bt_array__resize/4.  There
	  was a design flaw in the previous interface, in that if the
	  array increased in bounds, the extra slots were filled with one
	  particular element from the old bt_array.  The extra argument is
	  the element to use to fill these slots instead.

* There is a new `extras' directory in the distribution that contains
  some additional libraries.  These provide support for the following
  application areas:

  - graphics using Tk and OpenGL

  - arithmetic on complex and imaginary numbers

  - processing HTML forms using the CGI interface.


Mercury 0.7.2, 13 October 1997
------------------------------

We have split the distribution into two parts, a `core' part and an
`extras' part.  We still recommend that people get both parts.

Changes to the Mercury language:
********************************

* We have added support for constraint handling.

  To support constraint handling, we've made the mode system a bit
  more flexible.  There is a new inst `any' for variables whose value
  is unknown but which may have constraints on them. 

  The support for `any' insts is not 100% complete; in particular, we
  do not support passing values of inst `free' where values of inst
  `any' are expected, so sometimes you have to explicitly call a predicate
  to initialize a free variable to inst `any'.  Also the Mercury language
  reference manual does not yet contain any documentation on `any' insts.

  The `extras' distribution includes packages for doing constraint
  solving on (a) floating point numbers and (b) terms containing
  Prolog-style variables.  See below.

* The C interface now includes generalized trailing support.

  The compiler has a new set of grades `*.tr' (e.g. `asm_fast.gc.tr')
  which provide support for trailing.  They could be used by predicates or
  functions defined using the C interface to perform such things as
  constraint solving, backtrackable destructive update, or even automatic
  unwinding of database transactions on backtracking.  See the
  documentation in the "Trailing" section of the Mercury language
  reference manual (it's at the end of the "C interface" section,
  which is in the chapter on "Pragmas").

* It is now possible to stop the compiler from optimizing "impure"
  Mercury code inappropriately.

  This is accomplished by declaring impure predicates to be impure,
  allowing the compiler to treat them cautiously.  The compiler tracks
  impurity, and requires all impure predicates, and calls to them, to
  be declared.  For more information, see "Impurity" section of the
  "Pragmas" chapter of the Mercury Language Reference Manual.

* We now support user-defined equality predicates.

  See the Mercury Language Reference Manual for details.

  However, until we have support for type classes (coming soon :-),
  you will probably run into trouble if you try to use compare/3,
  write/1, functor/2, etc., on a type with user-defined equality.
  Hence we recommend that this feature should not yet be used.
  Because of this, we haven't bothered to document the
  rationale or use for user-defined equality predicates
  (though rest assured that when we do have type classes,
  this feature will indeed be useful).

* We have introduced new syntax to allow higher-order predicate expressions
  to use DCG notation.

  For details, see the "Data-terms" section of the "Syntax" chapter
  and/or the "Creating higher-order terms" section of the "Higher-order"
  chapter in the Mercury Language Reference Manual.


Changes to the Mercury standard library:
****************************************

* We have rewritten the `store' module to use unique modes.

  The `store' module provides a declarative interface to mutable variables
  with destructive update.

* The library predicate unsorted_aggregate/4 in std_util.m
  now interleaves calls of the aggregation predicate with
  the generation of solutions, rather than first finding all
  solutions and then aggregating them.  This allows you
  to print out solutions as they are found, for example.

* We have added a few new predicates, namely list__takewhile/4,
  bag__to_list/2, and varset__new_named_var/4.

* We have changed the interface to lexer__get_token_list to use a more
  efficient representation of lists of tokens. The functionality is
  unchanged.

* We have fixed a problem where io__call_system/4 was not returning
  the exit code of the invoked command on some operating systems.

* We have fixed a bug in relation__rtc/4.

* We have added the predicate queue__delete_all/3.

* Map (and tree234) have 2 new predicates: map__foldl which is
  analogous to list__foldl, and map__map_values which is analogous
  to list__map.

* We have added integer.m, which implements arbitrary precision integers,
  and rational.m, which implements arbitrary precision rational numbers.


New library packages in the `extras' distribution:
**************************************************

* We have added a CLP(R) interface.

  The new library package `cfloat_lib', in the extras/clpr directory,
  is a Mercury interface to the CLP(R) constraint solver.  The module
  `cfloat' defines a type `cfloat' for constrained floating point numbers,
  together with the usual arithmetic operators (+, -, *, /, <, >, =<, >=)
  as well as some non-linear constraint functions (abs, min, max,
  sin, cos, arcsin, and arccos).  The module `dump' provides I/O predicates
  for printing out constraints.

  Note that since `cfloat' is a different type than `float', you
  have to use the `==' operator provided in this package rather
  than `=' if you want to unify a cfloat with a float.
  
  We don't yet support any equivalent to SICStus Prolog's
  call_residue/3 or the three-argument version of CLP(R)'s dump predicate.

  But apart from that, it all works nicely.  And even though we support
  equivalents to such nasty non-logical meta-programming constructs
  as CLPR's `dump' primitive, we still manage to preserve referential
  transparency -- the interface provided is a completely pure declarative
  interface.

* We have added some support for Prolog-style variables and coroutining.

  The module extras/trailed_updated/var.m provides a type `var(T)'
  which is a Prolog-style variable that holds a value of type T.
  These variables can have the new inst `any' described above.
  There's also an implementation of freeze/2, for coroutining
  (dynamic scheduling).  The extras/trailed_update/samples subdirectory
  has an example of the use of freeze/2 to solve the N-queens problem.

* We have added library modules for backtrackable destructive update.

  See the new modules `tr_array' and `tr_store' in the extras/trailed_update.
  These are versions of `array' and `store' that use trailed backtrackable
  destructive update.  The extras/trailed_update/samples subdirectory
  has an example of the use of tr_store to provide a reasonably efficient
  meta-interpreter.

* We have added an interface to ODBC databases in extras/odbc. 

  Thanks to the people from Mission Critical, in particular Renaud Paquay,
  for providing the original version. 


Changes to the Mercury compiler:
********************************

* We have added support for termination analysis.

  For details, see the "Termination analysis" subsection of the
  "Implementation-dependent pragmas" section of the "Pragmas" chapter
  of the Mercury Language Reference Manual.

  This implementation is experimental, but our tests show that it is
  capable of proving the termination of most predicates and functions
  in real programs.

  The current implementation of termination analysis depends on the
  third-party package lp_solve. This is package is available from
  <ftp://ftp.es.ele.tue.nl/pub/lpsolve>; it is also included in the
  lp_solve subdirectory of the Mercury source distribution. Note
  that the copyright of lp_solve includes substantial restrictions.

  Details of the analysis are available in "Termination Analysis for
  Mercury" by Chris Speirs, Zoltan Somogyi and Harald Sondergaard. In P.
  Van Hentenryck, editor, "Static Analysis: Proceedings of the Fourth
  International Symposium", Lecture Notes in Computer Science. Springer,
  1997.  A longer version is available for download from
  <http://www.cs.mu.oz.au/publications/tr_db/mu_97_09.ps.gz>.

* We have made it easier to use different compilation models ("grades").

  The Mercury compiler (mmc), the Mercury front-end to GNU C (mgnuc),
  and the Mercury linker (ml) now figure out which grade to use based
  on the options specified.  This means that for example to enable
  profiling, you can just compile and link with `--profiling', rather
  than having to specify grade `asm_fast.gc.prof'.

  Attempts to mix object files compiled with different grades should now
  result in errors at link time rather than undefined behaviour.

* We have improved the C interface.

  We now handle the case when `main' is defined in C rather than in Mercury
  slightly more cleanly -- there are functions mercury_init()
  and mercury_terminate() for initializing and terminating the
  Mercury runtime.  See runtime/init.h for documentation of the functions,
  and see samples/c_interface/simpler_c_calls_mercury for an example of
  their use.

* The compiler does a better job of constant-propagation optimization.

* We have fixed a few minor bugs.


Mercury 0.7.3, 1 November 1997
------------------------------

This release is primarily a bug-fix release.  The problems fixed
include installation problems on Windows, some problems with the
profiler, and compatibility with GNU Make versions >= 3.76.
But as well as bug fixes, there are a few minor improvements:

* The profiler now allows you to merge profiling data from multiple runs.

  There's a new script `mprof_merge_runs' to support this.
  See the "Profiling" section of the Mercury User's Guide,
  or the man page for `mprof_merge_runs'.

* Termination analysis no longer uses the `lp_solve' package,
  so we have removed it from the distribution.
  
  This avoids some portability problems and some copyright issues
  (the `lp_solve' package had a quite restrictive license).

* We've fixed one of the limitations: unique mode declarations
  no longer have to precede non-unique mode declarations.


Mercury 0.7.4, 1 November 1997
------------------------------

This release just corrected a couple of bugs in the binary
distribution for 0.7.3.


Mercury 0.8, 18 November 1998
-----------------------------

Changes to the Mercury language:
********************************

* The type system now includes support for Haskell-style type classes.

  Type classes let you specify an interface and then provide multiple
  different implementations of that interface.  They're similar to
  abstract base classes in C++ or "interfaces" in Java.
  See the "Type classes" chapter of the Mercury Language Reference Manual
  for details.

  Unlike Haskell 1.4, Mercury supports multi-parameter type classes,
  but we do not (yet) support constructor classes, and nor do we
  support default methods.

* Mode inference can now infer "mostly-unique" modes as well as
  "unique" modes.

* You can now declare both committed-choice ("cc") and backtracking (non-cc)
  modes for the same predicate.
  
  Determinism analysis will pick the appropriate one to use for each
  call based on the context.

* The module system now includes support for sub-modules.

  The aim of this extension is twofold.  One aim is to provide more
  fine-grained encapsulation control: nested sub-modules within a
  single source file provide a convenient method for encapsulating
  units smaller than a source file.  The other aim is to provide better
  support for structuring large software packages that consist of many
  source files.  Sub-modules can be defined in separate files, with
  separate compilation, which means that you can also use this feature
  to combine a group of existing Mercury modules into a single logical
  package, with proper namespace control and encapsulation.

  See the "Modules" chapter of the Mercury language reference manual
  for details.

* We have made more improvements to the C interface.

  The C interface now includes support for defining procedures
  that can have multiple solutions (i.e. those whose determinism
  is `nondet' or `multi') in C.

  Also there's a new declaration, `pragma import', which is a bit
  like the existing `pragma c_code' declaration except that
  instead of giving a C code fragment, you just give the name
  of a C function.  `pragma import' is like the inverse of the
  existing `pragma export' declaration.

* We have added support for automatic tabling (memoization).

  See the "Tabled evaluation" subsection of the "Implementation-dependent
  pragmas" section of the "Pragmas" chapter of the Mercury language
  reference manual.

* We have added (tentative) support for exception handling.

  The interface to exception handling is actually via an `exception.m'
  library module rather than a new language construct.  
  For now, this module is located in the `extras/exceptions'
  directory, which is part of the `mercury-extras' distribution,
  but our intent is to eventually migrate this into the Mercury
  standard library if experience with its use proves positive.

  The exception handling interface uses committed choice nondeterminism
  to avoid some semantic problems with previous exception handling
  proposals.

  See the documentation in the interface of `exception.m' for details.

Changes to the Mercury standard library:
****************************************

* There is also a new builtin function promise_only_solution/1,
  for calling `cc_multi' or `cc_nondet' code from `det' or `semidet'
  procedures.  See the "builtin" chapter of the Mercury Library
  Reference Manual for details.

* The getopt module now supports a new type of option data, namely
  `maybe_int(maybe(int))', to allow optional arguments with integer values.
  There is also a new corresponding lookup predicate,
  getopt__lookup_maybe_int_option/3. 

  See the "getopt" chapter of the Mercury Library Reference Manual for details.

* Support for memory profiling: new predicates report_full_memory_stats/0
  in benchmarking.m and io__report_full_memory_stats/2 in io.m.

  See the "benchmarking" chapter of the Mercury Library Reference Manual
  for details.

* The types `term', `var', `var_supply' and `varset' are now polymorphic.
  This allows one to distinguish between terms, etc. denoting different kinds
  of things by giving them different types. The new coercion predicates
  listed below allow one to coerce terms, etc between types.

  The monomorphic versions of these have been retained as equivalences
  to the polymorphic ones with the type variable instantiated to a dummy
  type `generic'.

* Miscellaneous new predicates.

  The Mercury standard library now includes the following new predicates:

	  bag__det_remove_list/3
	  bag__least_upper_bound/3
	  bag__remove_list/3
	  det_univ_to_type/2
	  eqvclass__same_eqvclass_list/2
	  io__read_line_as_string/{3,4}
	  list__take_upto/3
	  map__det_insert_from_assoc_list/3
	  map__det_intersect/5
	  map__det_union/5
	  map__intersect/4
	  map__sorted_keys/2
	  map__to_sorted_assoc_list/2
	  map__union/4
	  relation__add_values/4
	  relation__compose/3
	  relation__from_assoc_list/2
	  set__count/2
	  set_ordlist__count/2
	  store__new_cyclic_mutvar/4
	  term__coerce/2
	  term__coerce_var/2
	  term__coerce_var_supply/2
	  varset__coerce/2
	  varset__select/3

  In addition, there are four new system constants added to the float
  library module, float__radix, float__mantissa_digits, float__min_exponent
  and float__max_exponent.  There are also predicate equivalents for these.

  Also the old relation__to_assoc_list/2 predicate has been renamed as
  relation__to_key_assoc_list/2; there is a new relation__to_assoc_list/2
  predicate with a different type for the second argument.

  See the Mercury Library Reference Manual for details.

* A few library procedures that have implicit side effects and are thus 
  intended for debugging use only have been declared `impure'.
  You will need to write `impure' before every call to these procedures
  and typically you will need to add a `pragma promise_pure' declaration
  for the callers.

  The predicates affected are report_stats/0 and report_full_memory_stats/0
  in library/benchmarking.m; unsafe_dump/2, unsafe_dump_float/1, and
  unsafe_dump_tableaus/0 in extras/clpr/dump.m; and debug_freeze/3
  and debug_freeze/4 in extras/trailed_update/var.m.

* The interface to the predicate term__compare/4 was found to be error-prone,
  and so we are phasing it out; it is declared with `pragma obsolete'
  in this version, so any use of it will now result in a warning, and
  the predicate will be removed entirely in some future version.

Changes to the Mercury implementation:
**************************************

* We've added a new source-to-source transformation - deforestation.

  Deforestation transforms conjunctions to avoid the construction
  of intermediate data structures and to avoid multiple traversals
  over data structures. Deforestation is enabled at optimization level
  `-O3' or higher, or by using the `--deforestation' option.

* The compiler can now perform type specialization.

  Type specialization removes the overhead of polymorphic code, including
  code which uses type classes. The disadvantage is increased code size.
  Currently we do not perform inter-module type specialization.
  Type specialization is enabled by using the `--type-specialization' option.

* We've added support for "transitive" inter-module analysis.

  With the previous support for inter-module optimization, when
  analysing a module, the compiler could make use of information
  about the modules that it imports directly, but not about
  modules that are imported indirectly.  "Transitive" inter-module
  analysis gives the compiler information about indirectly
  imported modules.

  However, currently this is only used for termination analysis;
  optimizations such as inlining still use only ordinary inter-module
  analysis, not transitive inter-module analysis.

* Array bounds checking can now be disabled.

  To disable array bounds checking, you must compile with
  `--intermodule-optimization' enabled and you must also
  define the C macro ML_OMIT_ARRAY_BOUNDS_CHECKS (e.g. by using
  `MGNUCFLAGS=-DML_OMIT_ARRAY_BOUNDS_CHECKS' in your Mmakefile). 

* Domain checking for higher mathematical operations can now be disabled.

  To disable domain  checking, you must compile with
  `--intermodule-optimization' enabled and you must also
  define the C macro ML_OMIT_MATH_DOMAIN_CHECKS (e.g. by using
  `MGNUCFLAGS=-DML_OMIT_MATH_DOMAIN_CHECKS' in your Mmakefile). 

  See the Mercury Library Reference Manual for details.

* We've added some primitive debugging support.

  The runtime system now includes a "four-port" style debugger
  (actually with eight ports).
  To use this debugger, you need to build your program with debugging
  enabled, which is normally done using the `--debug' (or `-g') option,
  and then run it using the `mdb' command, e.g. `mdb a.out'.
  Type `h' at the `mdb>' prompt for a list of the available commands,
  or see the "Debugging" chapter of the Mercury User's Guide for further
  details.

* The support for debugging using Prolog now includes support for
  detailed control over how terms are printed out during debugging.

  See the "Using Prolog" section of the Mercury User's Guide for details.
  However, for most purposes we now recommend using the native Mercury
  debugger rather than debugging using Prolog.

* The Mercury profiler has a number of new features.

  The profiler now supports profiling just user time, or profiling
  real (elapsed) time, rather than profiling user + system time.
  We've also added support for memory profiling.

  See the "Profiling" chapter of the Mercury User's Guide for details.

* Profiling should now work on MS Windows.

  To enable profiling on MS Windows, you need to have Sergey
  Okhapkin's latest version of gnu-win32 that includes his patch to add
  support for setitimer().  Sergey's "CoolView" version of cygwin.dll
  is available via <http://miracle.geol.msu.ru/sos/>; his patch will
  probably also be included in the next (b19) release of gnu-win32.
  Note that on Windows, you must use the Mercury runtime system's `-Tr'
  (profile real time) option; profiling just user time or user + system
  time is still not supported on Windows, because to the best of our
  knowledge Windows doesn't provide the necessary system calls.

* Intermediate files can be placed in subdirectories.

  If you use the `--use-subdirs' option to `mmake' or `mmc',
  then they will create the various intermediate files used
  by the Mercury implementation in a subdirectory called `Mercury'
  rather than in the current directory.  (For `mmake', if there
  is already a `Mercury' subdirectory, then this option is the default.)
  This keeps your source directories much less cluttered.

* Mmake has a new variable GRADEFLAGS for specifying options that
  affect the grade (compilation model).

  This means that for example to enable profiling, you can build with
  `GRADEFLAGS = --profiling' in your Mmakefile, rather than having to
  use the more obscure incantation `GRADE = asm_fast.gc.prof'.

* Mmake now supports per-file settings of MCFLAGS and other *FLAGS variables.

  For example, if you want to disable singleton variable warnings just
  for a single module `horrible_code.m', you can just include the line

	  MCFLAGS-horrible_code = --no-warn-singleton-variables

  in your Mmakefile.

* Mmake now warns about variables which are defined but not used.

  To disable this warning, use the `-w-' or `--no-warn-undef-variables'
  option.

* The components of the argument to the `--grade' option and of the `GRADE'
  Mmake variable may now be given in any order. The compiler also has a
  new option `--output-grade-string' which prints the canonical grade string
  for the set of options with which the compiler was invoked.

* Mmake now runs a bit faster, particularly on Windows.

* We've made a few small improvements to the efficiency of the generated code.

* The system has been ported to Linux/PPC.

* The system has been ported to work with version b19 of cygwin32
  (this port has not been tested well, though). 
  See README.MS-Windows for details.

* We've updated to version 4.13alpha2 of the Boehm garbage collector.

* We've made the MERCURY_OPTIONS environment variable somewhat easier to use.

* Mtags is now able to produce tags for type class declarations.  It is
  also able to produce tags files in the extended format supported by
  recent versions of Vim and Elvis.  Do `mtags --help' for more
  information.

* Numerous bug fixes.


Mercury 0.8.1, 16 December 1998
-------------------------------

This release just corrected some bugs in the binary
distribution for 0.8.


Mercury 0.9, 18 December 1999
-----------------------------

HIGHLIGHTS
==========

Changes to the Mercury language:
* The Mercury type system now supports existentially quantified types.
* We now allow abstract instance declarations.
* We now support a simple form of user-defined infix operators.

Changes to the Mercury standard library:
* Exception handling support is now part of the standard library.
* There are two new standard library modules `time' and `gc'.
* We've added function versions of many of the predicates in the
  Mercury standard library.

New library packages in the Mercury extras distribution:
* We've added support for optional lazy evaluation.
* The extras distribution now includes support for dynamic linking.
* We've added some bindings to POSIX.3 functionality.

Changes to the Mercury implementation:
* Mmake, the Mercury make tool, now includes better support for
  installing libraries.
* The Mercury debugger (mdb) is much improved.
  It now includes support for interactive queries, command-line editing
  and command-line history, display of source line numbers, and
  setting breakpoints on source line numbers.
  The GNU Emacs interface provides a source-linked debugger.
* We've removed the support for using a Prolog debugger on Mercury programs.
* We've added support for user-guided type specialization.
* Numerous bug fixes.

DETAILED LISTING
================

Changes to the Mercury language:
********************************

* The Mercury type system now supports existentially quantified types.

  Existential types let you create heterogeneous collections (e.g. lists
  containing objects of different types).  In combination with type
  classes, they allow you to write code in an OOP-like style.
  See the "Existential types" chapter of the Mercury Language Reference
  Manual for details.

  Our current implementation still has a couple of important limitations;
  see the "Known bugs and limitations" section of the "Existential types"
  chapter of the Mercury Language Reference Manual.

* We now allow abstract instance declarations.

  You can declare in the interface of a module that a type is an
  instance of a particular type class, and provide the definition
  of that instance in the implementation section of that module.

* We now support a simple form of user-defined infix operators.
  
  Terms in the form of x `fun` y are transformed into fun(x, y).  `fun`
  is parsed as an infix operator with the highest possible precedence
  and left associativity.

* We've made a small change to the rule for quantification of lambda
  expressions.  

  The new rule is that all variables in the arguments of lambda
  expressions are treated as locally quantified to that lambda expression.
  For function lambda expressions, variables in the result term
  use the normal quantification rules.  See the "Data-terms" section
  of the "Syntax" chapter of the Mercury Language Reference Manual
  for details.
  
  Previously, the exact quantification rule for lambda expressions was
  not documented, but the compiler would locally quantify variables in
  function return values, and it would only locally quantify variables
  occurring at the top level of an argument term, not those occurring in
  subterms.  Both of these were rather surprising for functional
  programmers.

  It is possible that this change may break some existing code using
  predicate lambda expressions with compound terms as arguments, but we
  expect this to be extremely rare.  If it does occur, the compiler
  will issue a warning about variables having overlapping scopes, and
  the work-around is simple: use a fresh variable for the lambda
  predicate argument and unify it with the compound term in the body of
  the lambda expression.

* The old-style syntax for predicate lambda expressions,
  `lambda([<Args>] is <Det>, <Goal>)', is now officially deprecated.

  Please use the new syntax-style `(pred([<Args>]) is <Det> :- <Goal>)'
  instead.  The compiler still supports the old-style syntax, but
  we plan to eventually drop this support in some future release.

Changes to the Mercury standard library:
****************************************

* Exception handling support is now part of the standard library.

  The module `exception', which was previously part of the "extras"
  distribution, has been moved into the standard library.
  The predicate error/1 now throws an exception rather than just
  terminating execution.

  However, many of the operations in the standard library still handle
  errors by aborting execution rather than by throwing exceptions.

* There's a new standard library module `time'.

  The `time' module provides an interface to the ANSI/ISO C <time.h>
  functions, and to the POSIX times() function.  Thanks to Tomas By
  for contributing the original version of this module.

* There's a new standard library module `gc', for controlling the
  garbage collector.

  Currently it contains only one predicate, `garbage_collect',
  which forces a garbage collection.  We may add more later.

* We've added some new predicates to the Mercury standard library:
	array__map/3,
	bag__count_value/3,
	std_util__do_while/4.

* We've added function versions of many of the predicates in the
  Mercury standard library.

  One small drawback of this change is that it is not completely
  backwards compatible; in certain circumstances, there is a potential
  ambiguity between a function call and a partially applied predicate,
  and for some occurrences of this the compiler may not be able to
  resolve the ambiguity unless the user provides additional type
  declarations (or the like).  But such cases should be quite rare,
  and when they do occur the fix is easy, so we thought the clear
  advantages of using a functional syntax were well worth this minor
  glitch in backwards compatibility.

* The following predicates have been replaced by functions with
  the same names, and will be removed in a future release.

  The predicate versions were intended for use in programs which needed
  to work in both Prolog and Mercury, but executing Mercury programs using
  Prolog is no longer supported.

	float__ceiling_to_int/2,
	float__floor_to_int/2,
	float__round_to_int/2,
	float__truncate_to_int/2,
	float__abs/2,
	float__max/3,
	float__min/3,
	float__pow/3,
	float__hash/2,
	float__max/1,
	float__min/1,
	float__epsilon/1,
	float__radix/1,
	float__mantissa_digits/1,
	float__min_exponent/1,
	float__max_exponent/1.

* The implementations of `int:>>/2' and `int:<</2' have been changed to define
  the results for negative shift counts and shift counts greater than the
  word size.

  For efficiency, we also provide the functions `int:unchecked_left_shift/2'
  and `int:unchecked_right_shift/2' that, like the previous implementations
  of `int:>>/2' and `int:<</2', do not check for these cases.

* `int:^/2' and `integer:^/2' have been replaced by `int__xor/2' and
  `integer__xor/2', and will be removed in a future release.
  The operator `^' will be used by record syntax.

New library packages in the Mercury extras distribution:
********************************************************

* We've added support for optional lazy evaluation.

  The extras distribution now includes a new module `lazy',
  which provides support for optional lazy evaluation
  via a type `lazy(T)', with `delay' and `force' operations.
  There's also a `lazy_list' module which uses the `lazy' module.
  See the files in extras/lazy_evaluation for details.

* The extras distribution now includes support for dynamic linking.

  The interface is based on the C functions dlopen(), dlsym(), and co.,
  which are supported by most modern Unix systems.
  See the files in extras/dynamic_linking for details.

* We've added some bindings to POSIX.3 functionality.

  At this stage it's quite incomplete.
  See the files in extras/posix for details.

Changes to the Mercury implementation:
**************************************

* Mmake, the Mercury make tool, now includes better support for
  installing libraries.

  It's now much easier to build and install libraries in several
  different grades (e.g. for debugging, time profiling, and memory
  profiling) or for more than one architecture.

  See the "Supporting multiple grades and architectures" section
  of the "Libraries" chapter of the Mercury User's Guide.

* We've fixed a bug in switch detection.

  This change may break some code written for Mercury 0.8. Some
  disjunctions which Mercury 0.8 found to have determinism `det'
  now have determinism `nondet'.

  Mercury 0.8 (but not Mercury 0.7) allowed switches where a unification
  to test the switched-on variable against a function symbol occurred after
  the first call in the disjunct. Doing this may remove infinite loops,
  violating the strict sequential semantics (see the "Semantics" chapter
  of the Mercury Language Reference Manual).

  To fix switches for which determinism errors are now reported, simply
  reorder the goals in each disjunct so that only unifications occur
  before the test of the switched-on variable.

* The Mercury debugger (mdb) now includes support for interactive queries.

  See the "Interactive query commands" subsection of the "Debugger commands"
  section of the "Debugging" chapter of the Mercury User's Guide for details.

* The Mercury debugger (mdb) now optionally supports command-line editing
  and command-line history.

  This support uses the GNU Readline library.  For the source distribution,
  the Mercury configure script will detect whether readline has been
  installed and will only enable the command-line editing and history
  support if readline has been installed.  For the binary distribution,
  if the binary distribution was built with readline, then you will
  need to install GNU readline in order to use the debugger.

  For information on where to obtain GNU Readline, see the INSTALL file.

* The Mercury debugger (mdb) now displays source line numbers and allows
  setting breakpoints on source line numbers.

  The GNU Emacs interface takes advantage of this to provide a
  source-linked debugger.

* We've removed the support for using a Prolog debugger on Mercury programs.

  Now that we have a working Mercury debugger, there's no longer any need to
  use a Prolog debugger for debugging Mercury code.

  Normally we would warn at least one or two releases in advance, if
  any feature is to be removed.  However, in this case

  	- it was an implementation feature rather than a language feature;
	- the cost of maintaining the feature was quite high;
	- the feature was already broken is various ways [one being that it
	  doesn't work with the latest versions of SICStus Prolog, due to
	  those versions removing support for a SICStus Prolog feature
	  (save/1), apparently without any advance notice]; and
	- a simple work-around is available if anything breaks as a result
	  of the feature being removed. 

  In the unlikely event that anyone happened to have any makefiles or
  scripts that depended on the support for using Prolog, they can
  install the latest Mercury distribution and still continue to use the
  Prolog support from Mercury 0.8, just by including the `bin'
  directories for both versions in their PATH, with the more recent one
  first, of course.

* We've added support for user-guided type specialization.

  See the "Type specialization" section of the "Pragmas" chapter of the
  Mercury Language Reference Manual for details.

* Numerous bug fixes.


Mercury 0.9.1, 26 January 2000
------------------------------

This release is primarily a bug-fix release.  
It fixes some bugs with the binary distribution of 0.9,
stops the compiler accepting some incorrect inst declarations,
fixes a bug in exception handling and a problem with the source
distribution where `configure' did the wrong thing on some architectures
if you ran it twice.

In addition, Morphine has been added to the extras distribution.
Morphine is a trace analysis system, which allows Mercury programs to be
debugged and dynamically analyzed using a Prolog interface.  You need
the ECLiPSe Prolog system to use Morphine.  See the README file in the
Morphine directory for more details.


MLDS back-end history
---------------------

We started working on a new back-end for the Mercury compiler in July
1999.  This new back-end, called the MLDS back-end, generates much
higher level C code than original back-end.  The first prototype
compiled "hello world" in September 1999.  The compiler successfully
compiled itself using the MLDS back-end on 12 May 2000.


Mercury 0.10, 25 February 2001
------------------------------

HIGHLIGHTS

Changes to the Mercury language:
* We've added support for explicit type qualification.
* We've added support for tuples.
* We've added support for record syntax.
* Type class methods can now be defined by listing the clauses
  directly in the instance declaration.
* The syntax for defining insts and modes has been changed.
  The old syntax is still accepted but is deprecated.

Changes to the Mercury standard library:
* We've added several new standard library modules:
  - `pprint', for pretty printing.
  - `counter', for managing counters.
  - `enum', a typeclass for types which can be converted to and from integers.
  - `sparse_bitset', an abstract data type for storing sparse sets of integers
     or enumerations.
  - `bitmap', an abstract data type for storing sets of integers.
  - `hash_table', an generic hash table implementation
* The `store' module now makes use of existential types.

Changes to the Mercury implementation:
* We've implemented a new back-end for the Mercury compiler.
  This features improved compilation speed, offers better portability,
  and sometimes generates substantially better code.
  (The original back-end is still included.)
* There's a version of the new back-end which generates code
  for Microsoft's new .NET system.
* There's a version of the new back-end which compiles directly
  to assembler, using the GCC back-end.
* Various improvements to `mtags'.

Additional packages in the mercury-extras distribution:
* Moose: a parser generator for Mercury.
* concurrency: support for multi-threading/concurrency.
* stream: an implementation of generic I/O streams, using type classes.
* xml: a library for parsing XML.

DETAILED LISTING

Changes to the Mercury language:

* We've added support for explicit type qualification.

  An expression of the form "Term `with_type` Type",
  e.g. "X `with_type` list(int)", can be used in place of
  the specified Term to constrain the type of that term.
  This is sometimes useful for resolving type ambiguities,
  which can occur as a result of overloading or polymorphism.

  See the "Explicit type qualification" and "Variable scoping"
  sections of the language reference manual for details.

* We've added support for tuple types, similar to those in most
  other functional languages. Tuples use the syntax `{A, B, ...}'.
  See the "Builtin types" section of the "Types" chapter of the
  Mercury Language Reference Manual for details.

* We've added support for record syntax, so that fields of
  constructors can be conveniently extracted and updated
  without writing lots of trivial access predicates.
  See the "Field access functions" section of the "Types" chapter
  of the Mercury Language Reference Manual for details.

  Note that the syntax has changed slightly since the version
  that appeared in the release of the day in early January 2000.
  `Value =^ field' is now the syntax for DCG field selection,
  rather than `Value := ^ field'. Field update functions are
  named 'field :=' rather than 'field:='. We also allow field
  access functions to take extra arguments.

* The behaviour of the Mercury parser (parser__read_term) applied
  to terms with functor `{}/N' has been changed. The parser from
  Mercury 0.9 parsed "{1, 2, 3}" as `{}(','(1, ','(2, 3)))'.
  It is now parsed as `{}(1, 2, 3)'.

* The operator `^' is now used for record syntax, and cannot be
  used for user-defined functions or constructors.

* You can now declare functions by giving a determinism but without
  supplying the modes.  The default function modes will be assumed.
  This is particularly useful for partial functions.
  For example:
  	GetEvens = list__filter_map(
		(func(X) = X is semidet :- X mod 2 = 0)).

* We've generalized the higher-order term syntax a little:
  in `Foo(Args)', we now allow Foo to be any term, not just
  a variable.

* The syntax for defining insts and modes has been changed to be
  more uniform.  For example, the old syntax

  	:- inst myfree = free.
  	:- mode out :: myfree -> ground.

  would now be written

  	:- inst myfree == free.
  	:- mode out == myfree >> ground.

  The old syntax is still accepted but is deprecated.  Support for it may
  eventually be dropped.

* Type class methods can now be defined by listing the clauses
  directly in the instance declaration.  You no longer need to define a
  separate predicate or function for each type class method definition.

Changes to the standard library:

* We've added some new library predicates: assoc_list__keys_and_values,
  list__map2, list__map3, map__foldl2, tree234__foldl2, relation__traverse,
  std_util__aggregate2, and builtin__promise_only_solution_io.

* We've added function versions of std_util__solutions,
  std_util__solutions_set, std_util__aggregate, map__search,
  map__insert and map__update.

* We've added functions to allow record syntax to be used
  with some of the types in the standard library:
	array__elem/2, 'array__elem :='/3,
	bt_array__elem/2, 'bt_array__elem :='/3,
	map__elem/2, 'map__elem :='/3,
	map__det_elem/2, 'map__det_elem :='/3.

* We've added a pretty printing module, `pprint', to the standard library.

* We've added a new function to the Mercury standard library:
	std_util__construct_tuple/1.

* Functions `int:^/2' and `integer:^/2' have been removed.
  Use `int__xor/2' and `integer__xor/2' instead.
  The operator `^' is now used for record syntax.

* We've added reverse modes for `int__xor'.

* There is a new predicate `random__permutation', for
  computing a random permutation of a list.

* There is a new library module `counter' for managing counters.

* We've added a new library module `sparse_bitset', which implements
  an abstract data type for storing sets of integers or enumerations.

* There is a new library module `enum' which contains a typeclass
  describing types which can be converted to and from integers.

* Four new parametric instantiations `maybe/1', `maybe_error/1',
  `pair/2' and `pair/1' have been added to the `std_util' library
  module.  These make it more convenient to work with non-ground
  terms of the corresponding type.

* The `store' module now makes use of existential types.

  The `store__init/1' predicate and the `store__some_store_type' type
  are now deprecated; the new existentially typed predicate
  `store__new/1' should be used instead.

* We've reimplemented the `string__format/3' procedure.

  The new implementation is a lot more efficient and fixes several
  bugs in the old implementation.  The new implementation also
  eliminates some subtle differences in behaviour between
  string__format and the standard ANSI/ISO C printf() function:

	- For the old string__format, the default precision was 15
	  (i.e. the number of significant figures in an IEEE double
	  precision float), but for ISO C's printf(), the default
	  precision is 6.

	- For the old string__format, for the e, E, f, F, g and G conversions,
	  the "precision" field in the format always specified the
	  number of significant figures, but for ISO C's printf(), the
	  precision only specifies as the number of significant
	  figures for the g and G conversions, whereas for the e, E,
	  f, and F conversions the precision specifies the number of
	  digits after the decimal point.

	- For the old string__format, floating point numbers were
	  truncated to the specified precision, but for ISO C's
	  printf(), they are rounded rather than being truncated.

* We've added a new function, math__solve_quadratic/3.

* We've changed the semantics of deconstruct/4, in light of the introduction
  of existentially quantified types. Previously, if deconstruct/4 was given
  a value of type `univ' it automagically unwrapped it and gave back the
  functor, arity and arguments of the unwrapped value. This behaviour was
  not documented, but made sense because there was no way to unwrap a
  univ without knowing (or guessing) its type. Now that univ is defined
  as a normal (existentially quantified) type, this behaviour is unnecessary,
  and a wart besides, so has been removed. If you have a univ and you want
  to get the unwrapped value's functor, arity and arguments, then you can
  call "univ_value(Univ)" to extract the value before calling deconstruct.
  (Doing that also works in Mercury 0.9 and Mercury 0.10.)

* We've added func versions of the remaining preds in int.m that
  did not already have them.

* We've added a new `bitmap' library module.

* We've added std_util__dynamic_cast/2 for type-safe runtime dynamic
  type casting for ground types.

* We've extended the array module with array__sort/1, array__foldl/3 and
  array__foldr/3.

* We've added a new `hash_table' library module.

Changes to the Mercury implementation:

* We've implemented a new back-end for the Mercury compiler.

  The new back-end, which is enabled by using the `--high-level-code'
  (or `-H') option or the `hlc.gc' grade, generates much higher-level
  C code that does not require the use of GNU C extensions such as
  global register variables or non-local gotos.  It is also simpler
  and more portable than the old back-end.

  The main drawback of the new back-end is that for tail calls it only
  optimizes direct tail recursion; loops written using tail calls
  between two or more mutually recursive procedures are not guaranteed
  to use constant stack space.

  Preliminary benchmarking suggests that compilation speed is probably
  about 20% better with the new back-end, and the generated executables
  are likely to be smaller (though this will depend on the platform,
  optimization options, etc.).  Speed of the generated code varies:
  sometimes it is better than the old back-end, sometimes it is worse.
  There are a few optimizations that we have not yet implemented for
  the new back-end that might make a significant difference for some
  applications.  But there are also some optimizations which we have
  implemented for the new back-end that have not been implemented for
  the old back-end.  We encourage those for whom performance is
  important to try their application with both the old and new
  back-ends and compare for themselves.

  The new back-end is not yet quite as mature or complete as the old back-end.
  It does not yet support the following standard Mercury features:
  	- abstractly exported equivalence types defined as `float'
	- calling compare/3, or the `in = in' mode of unification,
	  for certain standard library types (std_util__type_desc/0,
	  and std_util__type_ctor_desc/0).
	- calling copy/2 on higher-order terms
  It also does not support the following implemention-specific
  features that the old back-end supports:
	- demangling of symbol names in the profiler
	- fact tables for procedures with determinism `nondet' or `multi'
  	- the Mercury debugger (mdb)
  	- the Morphine trace analysis system
	- the Aditi deductive database interface
  	- the `--split-c-files' option
  	- the `--introduce-accumulators' option
	- dynamic linking (via the dl__mercury_sym procedure in
	  extras/dynamic/dl.m in the mercury-extras distribution)
	  for procedures with arguments of type `float' or `char'

* There's a new back-end that targets .NET.

  Thanks to Microsoft's generous and ongoing support, both financial
  and otherwise, we've been able to port Mercury to Microsoft's new
  .NET system.  There's another new back-end for the Mercury compiler,
  based on the `--high-level-code' back-end, that compiles to IL, the
  Microsoft .NET Intermediate Language.

  This back-end is enabled using the new `--target il' option
  (or just `--il' for short), or the `ilc' grade.

  Compiler support for this new back-end is mostly complete,
  but large parts of the standard library are still not yet
  implemented for this new port.

  This is still work in progress.

  For more details, see the README.DotNet file, and
  <http://www.cs.mu.oz.au/research/mercury/dotnet.html>.

* Native code compiler.

  There's a new back-end for the Mercury compiler that compiles
  directly to assembler, rather than going via C.  This
  back-end is enabled using the new `--target asm' option.

  This new back-end is implemented by linking the Mercury compiler
  with the (relatively) language independent GNU Compiler Collection
  back-end.  In other words, there is now a Mercury front-end for GCC.

  Note that this should be considered as a beta release of the native
  code compiler.  Furthermore our current version of the native code
  compiler is based on an unreleased snapshot version of the GCC
  back-end.

  So far we have only tested it on i686-pc-linux-gnu (Intel x86-based
  PCs running Linux).  But in theory it should work fine on other
  platforms too.

  For details see <http://www.cs.mu.oz.au/mercury/download/gcc-backend.html>.

* The old back-end now generates faster code at low optimization levels.

* The compiler is now a little bit faster.

* The names of some of the `--enable-*' options to `configure' have changed.

  See the output of `configure --help' for details.

Changes to the development environment:

* The debugger has been improved in several respects:

  - It has some new forward movement commands:
    * `next' steps over a call, like gdb's `next' command;
    * `exception' skips forward until an exception is thrown.
  - It can perform retry across I/O.
  - It can now print parts of terms, and fields of terms can be
    specified by field name as well as by position number.
  - It has a more flexible mechanism for setting browsing parameters.
  - It now handles ambiguous procedure specifications in "break"
    commands gracefully.
  - The state of the debugger can now be saved and restored, using the
    `save' and `source' commands (respectively).

  For details, see the documentation of the `next', `exception',
  `break', `set', and `save' commands in the "Debugger commands" section
  of the "Debugging" chapter of the Mercury User's Guide.  (The changes
  to `retry' and `print' have unfortunately not been documented yet.)

* Several improvements have been made to `mtags' to make it easier to
  find predicate/function definitions and to improve support for
  enhanced features of Vim.  The command-line options to `mtags' have
  changed and Vim-style tags files are now output as the default (but
  will work with Vi as well).  Do `mtags --help' for more information.


Mercury 0.10.1, 3 April 2001
----------------------------

This is mainly a bug-fix release.

There are however some new packages in the mercury-extras distribution:
* lex: a lexical analyzer library for Mercury.
* curs: a simplified binding to the ncurses/panel libraries for terminal I/O.


Mercury 0.11, 24 December 2002
------------------------------

HIGHLIGHTS
==========

Changes to the Mercury language:
* Support for constrained polymorphic modes.
* Addition of state variable syntax.
* Improved support for higher-order functions.
* Predicate and function equivalence type and mode declarations.
* Support for defining predicates or functions
  using different clauses for different modes.
* Support for Haskell-like "@" expressions.
* Generalized foreign language interface.

Changes to the Mercury compiler:
* A new `--make' option, for simpler building of programs.
* A new `--smart-recompilation' option, for fine-grained dependency tracking.
* A new optional warning: `--warn-non-tail-recursion'.
* A new optimization: `--constraint-propagation'.
* A new optimization: `--loop-invariants'.
* Support for arbitrary mappings from module name to source file name. 

Portability improvements:
* Mac OS X is now supported "out-of-the-box".
* On Windows we now support generating non-Cygwin executables.
* Better conformance to ANSI/ISO C.

Changes to the compiler back-ends:
* The native code Linux/x86 back-end is now "release quality".
* The .NET CLR back-end is much improved.

Major improvements to the Mercury debugger, including:
* Support for source-linked debugging using vim (rather than emacs).
* Command-line completion.
* Ability to display values of higher-order terms.
* Declarative debugging.
* Support for transparent retries across I/O.

A new profiler, which we call the Mercury deep profiler or mdprof:
* Supports both time and memory profiling.
* Gathers information about individual call sites as well as procedures.
* Eliminates the assumption that all calls to a procedure have equal cost.
* Allows users to explore the gathered data interactively with a web browser.

Numerous minor improvements to the Mercury standard library.

A new testing tool in the extras distribution.
A new regex module for string matching and search-and-replace in the
extras distribution.

DETAILED LISTING
================

Changes to the Mercury language:

* We have added support for constrained polymorphic modes.  See the section on
  Constrained Polymorphic Modes in the Modes chapter of the Mercury Language
  Reference Manual.

* A more general alternative to DCG syntax has been added to the language
  to simplify the manipulation of threaded state.  See the section on State
  Variables in the Syntax chapter in the Mercury Language Reference Manual.

* If a higher-order function term has inst 'ground' it is now assumed to have
  the standard higher-order function inst 'func(in, .., in) = out is det'.
  This makes higher-order functional programming much easier, particularly when
  passing functions to polymorphic predicates.

  This change is not completely backwards compatible since, for safety,
  we must now disallow calls that would cause a variable that has a
  nonstandard function inst to become 'ground'.

* Predicate and function type and mode declarations can now be expressed
  in terms of higher-order predicate and function types and insts, rather
  than explicitly listing the argument types and modes.  This is useful
  where several predicates or functions must have the same type and
  mode signature.

  For example:
	:- type foldl_pred(T, U) == pred(T, U, U).
	:- inst foldl_pred == (pred(in, in, out) is det).
	:- pred p `with_type` foldl_pred(T, U) `with_inst` foldl_pred.
 
  For more information see the "Predicate and function type declarations"
  section of the "Types" chapter and the "Predicate and function mode
  declarations" section of the "Modes chapter" of the Mercury Language
  Reference Manual.

* The constructor for lists is now called '[|]' rather than '.'.
  `./2' will eventually become the module qualification operator.
  This change only affects programs which use `./2' explicitly.
  Programs which only use the `[H | T]' syntax will be unaffected.

* We've added a new kind of expression to the language.
  A unification expression, written `X @ Y', unifies X and Y
  and returns the result.

  Unification expressions are most useful when writing switches:
	p(X, X) :- X = f(_, _).
  can now be written as
	p(X @ f(_, _), X).

  See the "Data-terms" section of the "Syntax" chapter of the
  Mercury Language Reference Manual for more details.

* We've extended the language to allow you to specify different clauses
  for different modes of a predicate or function.  This is done by
  putting mode annotations in the head of each clause.
  For example, you can write

	:- mode p(in).
	:- mode p(out).
	p(X::in) :- ... /* clause for the `in' mode */
	p(X::out) :- ... /* clause for the `out' mode */

  For predicates or functions which have different clauses for different
  modes, you need to either (1) add a `pragma promise_pure' declaration
  for the predicate or function, and ensure that the declarative semantics
  remains the same in each mode, or (2) declare the predicate as impure.

* We now allow `:- pragma promise_semipure' declarations. For more
  information, see the "Impurity" chapter of the Mercury Language
  Reference Manual.

* We've added `:- pragma c_import_module' declarations, which are
  used to make the C declarations for predicates and functions with
  `:- pragma export' declarations in the imported module visible
  to any C code in the importing module. `mmake' uses
  `:- pragma c_import_module' declarations to make sure that the
  header file for the imported module is built before it is needed,
  which it can't do if the header file is explicitly #included.

* The foreign language interface has been generalized to support
  interfacing with languages other than C.

  In particular, the Mercury compiler's .NET back-end now supports
  interfacing with C#, IL, and "Managed C++" (C++ with Microsoft's
  extensions for the .NET CLR).  Mercury procedures can be defined
  using inline code fragments written in any of these languages.

  For details, see the new "Foreign language interface" chapter of
  the Mercury Language Reference Manual.

* We've removed the undocumented operators `export_adt', `export_cons',
  `export_module', `export_op', `export_pred', `export_sym', `export_type',
  `import_adt', `import_cons', `import_op', `import_pred', `import_sym',
  `import_type' `use_adt', `use_cons', `use_op', `use_pred', `use_sym'
  and `use_type'. These operators were reserved for module system
  extensions which are unlikely to be implemented.

Changes to the Mercury standard library:

* The Prolog-style term comparison operators @<, @=<, @>, @>= are now
  builtin.

* A new builtin function ordering/2 has been added.

* We've added a function to io.m to construct io__error codes from error
  messages: `io__make_io_error'.

* The assumptions that we make about user supplied comparison predicates and
  functions have been relaxed to allow more general orderings.  The new
  assumptions are documented in builtin.m.

* The builtin predicates !/0 and !/2 from Mercury's Prolog heritage have been
  removed (`!' is now a prefix operator used in the state variable syntax).

* We have added the type class `pprint__doc/1' and a new concatenation
  operator, `++/2', which should make it easier to construct doc values.
* Performance bugs in `pprint__to_doc' have now been fixed.  Even
  very large terms can now be converted to docs and pretty printed without
  causing a machine to thrash or run out of memory.

* `io__read_file' and `io__read_file_as_string' now have better error
  handling. The result types have changed, so code using these predicates
  will need minor modifications.
* We've added predicates `io__input_stream_foldl', `io__input_stream_foldl_io'
  and `io__input_stream_foldl2_io', which apply a predicate to each character
  of an input stream in turn.
* We've added predicates `io__binary_input_stream_foldl',
  `io__binary_input_stream_foldl_io' and `io__binary_input_stream_foldl2_io',
  which apply a predicate to each byte of a binary input stream in turn.
* We've added versions of `io__print', `io__write' and `io__write_univ'
  that allow the caller to specify how they should treat values of noncanonical
  types, e.g. types in which a single semantic value may have more than one
  syntactic expression.
* We've added four new predicates to allow programs to retrieve current
  streams: `io__current_input_stream', `io__current_output_stream',
  `io__current_binary_input_stream', and `io__current_binary_output_stream'.
* We've added a predicate to io.m to return the last modification time
  of a file: `io__file_modification_time'.
* We've added cc_multi modes to io__write_list/5 and io__write_list/6.
* You can now close stdin, stdout and stderr.

* We've added four functions to list.m for mapping functions over
  corresponding members of lists: list__map_corresponding/3,
  list__map_corresponding3/4, list__filter_map_corresponding/3
  and list__filter_map_corresponding3/4.
* We've added some other new functions to list.m, namely
  list__last_det/2, list__split_last/3 and list__split_last_det/3.
* We've added cc_multi modes to list__foldl/4 and list__foldr/4.
* We've added a predicate list__map_foldl2.
* As mentioned above, the constructor for lists has changed from './2'
  to `[|]/2'. This change affects the behaviour of the term manipulation
  predicates in the standard library when dealing with values of
  type `term__term/1' representing lists. The affected predicates are
  parser__read_term, parser__read_term_from_string, term__type_to_term,
  term__term_to_type, term_io__read_term and term_io__write_term.
  Also beware that std_util__functor and std_util__deconstruct now
  return `[|]' rather than `.' for lists, and calls to std_util__construct
  which construct lists may need to be updated.
* We've added the predicate list__is_empty/1 and list__is_not_empty/1.
* We've added the predicate list__remove_adjacent_dups/3.

* We've added a function version of error/1, called func_error/1, to require.m.

* ops.m now defines a typeclass which can be used to define operator
  precedence tables for use by parser.m and term_io.m. See
  samples/calculator2.m for an example program.

  The `ops__table' type has been renamed `ops__mercury_op_table'.
  `ops__init_op_table' has been renamed `ops__init_mercury_op_table'.
  `ops__max_priority' is now a function taking an operator table argument.

* The predicates and functions in int.m, float.m, math.m and array.m now
  generate exceptions rather than program aborts on domain errors and
  out-of-bounds array accesses. There are new functions
  `float__unchecked_quotient/2', `int__unchecked_quotient/2' and
  `int__unchecked_rem/2' for which no checking is performed and the
  behaviour if the right operand is zero is undefined.

* We've removed the reverse modes of the arithmetic functions in
  float.m and extras/complex_numbers.  (Because of rounding errors,
  the functions aren't actually reversible.)

* float__pow now works for negative exponents, and runs much faster
  for large exponents.

* We've removed the destructive update modes of string__set_char,
  string__set_char_det and string__unsafe_set_char. The compiler
  currently always stores constant strings in static data, even
  if they are passed to procedures with mode `di', so any attempt
  to update a constant string will cause a crash. Fixing this properly
  will be a lot of work, so for now we have just removed the modes.

* We've added string__suffix, string__words/1, string__foldr,
  string__foldl_substring and string__foldr_substring.

* The exception module has a new predicate `try_store', which is
  like `try_io', but which works with stores rather than io__states.

* We've fixed a bug in time.m. Type `tm' didn't store the day of the month,
  which meant that the functions which required that field (e.g. time__asctime,
  time__mktime) did not work.

  The order of the fields of type `time__tm' has been changed so that
  comparison of values of type `tm' whose `tm_dst' fields are identical
  is equivalent to comparison of the times those values represent.

* std_util.m now contains predicates and functions `map_maybe',
  `fold_maybe', `map_fold_maybe' and `map_fold2_maybe', which are
  analogues of `list__map', `list__foldl', `list__map_foldl' and
  `list__map_foldl2' operating on values of type `maybe' instead
  of `list'.

* We've added a predicate to io.m to return the last modification time
  of a file (io__file_modification_time).

* There is a variant of io__call_system, io__call_system_return_signal
  which on interrupt returns the number of the signal which interrupted
  the command rather than just an error message.

* We've added several new predicates for deconstructing terms to
  std_util.m. `named_argument' and `det_named_argument' are analogous
  to `argument' and `det_argument' respectively, but specify the desired
  argument by its name, not its position. We have also added committed choice
  version of all the predicates that deconstruct terms. These differ from the
  existing versions in that they do not abort when called upon to deconstruct
  non-canonical terms, such as values of types with user-defined equality.

* We've added a new predicate `intersect_list' in each of the modules
  implementing sets in the Mercury standard library.

* We've added a predicate version of `set__fold'.

* We've added function versions of `builtin__unsafe_promise_unique',
  `ops__init_op_table' and `ops__max_priority'.

* We've added a version of `getopt__process_options' which returns
  the option arguments as well as the nonoption arguments. Both versions
  allow multi as well nondet predicates to specify the default values
  of options.

* We've added a new predicate `getopt__process_options_track', which can be
  usefully invoked more than once to process separate sets of options.

* All these predicates in getopt.m now allow the allow negation of
  accumulating options. Negating an accumulating option empties
  the accumulated list of strings. They also accept a `maybe_string_special'
  option type.

* We've added some functions to the term_io module to return printable
  representations of term components as strings.

* We've made the outputs of the string concatenation primitives unique.

* New convenience/readability predicates `int__even/1' and `int__odd/1'.

* New predicate benchmark_det_io for benchmarking code that performs I/O.

* We've removed the long obsolete `int__builtin_*' and
  `float__builtin_float_*' predicates, which were synonyms
  for the arithmetic functions dating from when Mercury didn't
  have functions. 

* We've added int:'/'/2 as a synonym for int:'//'/2 and false/0 as a
  built-in synonym for fail/0 (both left-overs from Mercury's Prolog
  heritage.)

* dir:'/'/2 is now a synonym for `dir__make_path_name'.
  
* We've removed the long obsolete predicates `io__read_anything',
  `io__write_anything', and `io__print_anything', which were long ago
  renamed as `io__read', `io__write', and `io__print' respectively.

* We've added random__random/5, which produces a random integer in a
  given range, and random__randcount/3, which returns the number of
  distinct random numbers that can be generated.

Changes to the extras distribution:

* The lex subdirectory now contains a new module, regex, which provides
  for more traditional string-based ways of defining regular expressions
  and provides string matching and search-and-replace functionality.

* There's a new testing tool called "quickcheck", which is similar to
  Haskell's "QuickCheck".  See quickcheck/tutes/index.html.

* The interface to Moose has been changed in a non-backwards compatible
  way to support user-defined modes for the parser state and integrate
  better with lex.

Changes to the Mercury compiler:

* There is a new `--make' option which performs most of the functions
  of Mmake.  The advantages of `mmc --make' are that no `mmake depend'
  step is necessary and the dependencies are more accurate.  Parallel
  builds are not yet supported.  See the "Using Mmake" chapter of the
  "Mercury User's Guide" for details.

* The Mercury compiler can now perform smart recompilation, enabled by the
  `--smart-recompilation' option. With smart recompilation, when the
  interface of a module changes, only modules which use the changed
  declarations are recompiled. Smart recompilation does not yet work
  with `--intermodule-optimization'.

* The Mercury compiler can now handle arbitrary mappings from source files
  to module names.  If the program contains modules for which the source
  file name does not match the module name, before generating the
  dependencies the command `mmc -f SOURCES' must be run, where `SOURCES'
  is a list of the names of all of the source files.  If the names of the
  source files all match the contained module names, `mmc -f' need not be run.

* There is a new `--use-grade-subdirs' option which is similar to
  `--use-subdirs', but allows multiple grades to be built in a
  directory at the same time.  `--use-grade-subdirs' does not
  work with Mmake (it does work with `mmc --make').

* The compiler and scripts accept a `--mercury-stdlib-dir' option,
  which overrides the configured location of the Mercury standard
  library.  There is also an environment variable MERCURY_STDLIB_DIR
  which has the same effect.  The environment variables which were
  previously used to override the location of the standard library
  (MERCURY_ALL_C_INCL_DIRS, MERCURY_ALL_MC_C_INCL_DIRS, MERCURY_INT_DIR,
  MERCURY_C_LIB_DIR, MERCURY_MOD_LIB_MODS, MERCURY_TRACE_LIB_MODS) are
  now deprecated, and will be removed in a future release.
  MERCURY_C_INCL_DIR has already been removed.

* We've added a new compiler option `--warn-non-tail-recursion', which
  causes the compiler to issue a warning about any directly recursive
  call that is not a tail call.

* The automatically generated header files for modules containing
  `pragma export' declarations are now named `<module>.mh', not
  `<module>.h'. This avoids conflicts with system header files.

* We've fixed a long-standing bug in the handling of module imports.
  Previously, if `module1' imported `module2' which imported `module3' in
  its interface section, then any types, insts, modes and typeclasses defined
  in the interface of `module3' could be used in `module1' even
  if `module1' did not import `module3' directly.

  This change will break some existing programs, but that is easily fixed
  by adding any necessary `:- import_module' or `:- use_module' declarations.

* Options for the Mercury runtime can now be set at compile time using
  the new `--runtime-flags' option of ml and c2init.

* We've added a new optimization pass -- constraint propagation.

  Constraint propagation attempts to transform the code so
  that goals which can fail are executed as early as possible.
  It is enabled with the `--constraint-propagation' option
  (or `--local-constraint-propagation' for a more restricted
  version of the transformation).

* The Mercury compiler can now perform inter-module optimization using
  information from transitively imported modules. This is especially
  useful for back-ends which do not support abstract equivalence types
  properly (for example the .NET backend). To disable this behaviour and
  only optimize using information from directly imported modules, use the
  option `--no-read-opt-files-transitively'.

* For each `--Xflags' option there is now a `--Xflag' option which allows a
  single quoted argument to be passed to the invoked program.  This is useful
  where the argument is a directory name containing spaces.

* The `--convert-to-goedel' option has been removed.
  It never really worked anyway.

Portability improvements:

* Mac OS X is now supported "out-of-the-box".

  See README.MacOSX for details.

* On Windows we now support generating non-Cygwin executables.

  The Mercury compiler source distribution can be configured using
  `configure --with-cc="gcc -mno-cygwin"'.  This option ensures
  that the Mercury libraries are only linked with the standard
  Windows libraries, not the Cygwin Unix emulation library,
  so Mercury programs don't need Cygwin, and use DOS/Windows-style
  path names rather than Cygwin's Unix-style path names.

  Note that you still need Cygwin to install and use Mercury.
  The change is that the programs which you build using Mercury
  don't need Cygwin anymore.

* Better conformance to ANSI/ISO C.

  We now pass all the tests in the Mercury test suite
  when the compiler is built with the "lcc" C compiler,
  which is more strict about ANSI/ISO C conformance than GNU C.
  This should also make it easier to port to other C compilers.

Changes to the Mercury debugger:

* The debugger can now print goals just as Prolog debuggers do. At an exit
  port of e.g. append, the command "print goal" will print the current goal
  in a form such as "append([1], [2], [1, 2])".

* You can now navigate terms in the debugger by argument name as well as by
  argument number.

* The debugger can now print higher order values.

* The debugger can now print type_info structures. However, since such
  structures are normally of interest to implementors only, the debugger
  will print such values only if the user gives the command "print_optionals
  on".

* The debugger can now perform command line completion when compiled
  with GNU Readline support enabled.

* We've added a 'view' command to `mdb', which opens a `vim' window and
  in it displays the current source location, updated at each event.  This
  requires X11 and a version of `vim' with the `clientserver' feature
  enabled.

* The `--window' mdb option now creates a window for mdb, not
  the program.  The main advantage of the new behaviour is that
  redirection of the program's input and output works.  The old
  behaviour is still available with `mdb --program-in-window'.

* The debugger now includes support for declarative debugging.  The `dd'
  command starts diagnosis at any exit, fail or exception port in mdb.  See
  the Mercury User's Guide for more details.

* When a program is compiled in a debugging grade, the debugger can be
  asked, via the command `table_io start', to make I/O primitives (such as
  io__open_file, io__write_string etc) idempotent. This means that a given
  call to e.g. io__open_file will open the specified file only once,
  even if retry commands cause the call to be executed more than once.

A new profiler, which we call the Mercury deep profiler or mdprof:

* The old Mercury profiler is based on the technology of the standard Unix
  profiler gprof. This technology makes the assumption that all calls to a
  given C function (in Mercury, a given function or predicate in a given mode)
  have the same cost, whether the cost being measured is CPU time, memory cells
  allocated, memory words allocated etc. In C programs, this assumption is
  usually close enough to correct for the output of gprof to be useful. In
  Mercury, due to the presence of parametric polymorphism and the significantly
  higher frequency of higher order code, different call sites are far more
  likely to have distinct performance characteristics than in C, so the output
  of a gprof-style profiler is usually not accurate enough to be useful.

  The new profiler records, for each of its measurements, not just the current
  predicate/function and its caller, but the entire chain of ancestors. This
  "deep context" is what gives the profiler its name. Actually, to keep
  overheads down, we don't walk the stack at every measurement; we just
  associate the current context with each measurement, and update the current
  context when it changes. Given this fact, it costs very little extra to
  record measurements on every aspect of performance (counts of calls, exits,
  fails and redos, counts of memory cells and memory words allocated, and time
  spent). We thus have only one deep profiling grade component, .profdeep,
  as opposed to the old profiler which has several grade components
  for different subsets of these measurements.

* The deep context recorded by the deep profiler records the identities of
  the call sites as well as the identities of predicates and functions
  in the list of ancestors. If a predicate p contains two calls to predicate q,
  this allows the deep profiler to report that one call to q costs next to
  nothing while the other one is a major performance problem.

* The deep profiler gathers so much data that giving it to the user all at once
  would swamp the user with too much information. We therefore implemented the
  deep profiler as a CGI program. Users can use thus use a web browser to
  explore the information contained in profiling data files.

* The deep profiler currently does not handle programs that catch exceptions.

* Further information about the deep profiler is available in the paper
  "Deep profiling: engineering a profiler for a declarative programming
  language" by Thomas C. Conway and Zoltan Somogyi, available from our web
  site at <http://www.cs.mu.oz.au/mercury/information/papers.html#mu_01_24>.

Changes to the compiler back-ends:

* The native code Linux/x86 back-end is now "release quality".

  The native code back-end, which was first released in Mercury 0.10,
  compiles directly to assembler, rather than going via C.
  This back-end is enabled using the `--target asm' option.  It is
  implemented by linking the Mercury compiler with the (relatively)
  language independent GNU Compiler Collection back-end.  In other words,
  it is a Mercury front-end for GCC.

  This release is the first to be based on an officially released
  version of GCC (it is based on GCC 3.2).  In this release, the native
  code back-end now passes all of the applicable tests in the Mercury test
  suite, including bootstrapping the Mercury compiler.  Currently it is only
  supported on i*86-pc-linux-gnu (Intel x86-based PCs running Linux).

  For details see <http://www.cs.mu.oz.au/mercury/download/gcc-backend.html>.

* .NET CLR back-end much improved.

  The .NET CLR back-end, which generates MSIL code for Microsoft's new
  .NET Common Language Runtime, has been substantially improved.
  Mercury data structures are mapped to .NET CLR data types in a more
  natural and more efficient manner.  A lot more of the standard library
  is now supported.  Text files on Windows are now output with proper
  Windows CR-LF line endings.  Many bugs have been fixed.

  This back-end supports the whole of the Mercury language, but the
  Mercury standard library implementation for the .NET CLR is still
  not yet complete.  The .NET CLR back-end now passes about half of
  the tests in the Mercury test suite.

  This back-end is selected when you use the `--grade il' option.

  See <http://www.cs.mu.oz.au/mercury/dotnet.html> and/or
  <http://www.cs.mu.oz.au/mercury/information/dotnet/mercury_and_dotnet.html>.

* The cost in disk space of enabling debugging is now much smaller.

  When debugging is enabled, the compiler creates several data structures
  for use by the Mercury debugger mdb, and includes the definitions of these
  data structures in the generated C file. These definitions could be very
  large (several megabytes), because readable C code is an inherently verbose
  medium for expressing e.g. the values of pointers. By generating denser,
  less readable C code, we have greatly reduced the size of these definitions,
  and the sizes of the references to them in the executable code. The sizes
  of the C files generated by the compiler with debugging enabled are now
  typically between a quarter and a third of their previous sizes.

Mercury 0.12, 9 September 2005
------------------------------

HIGHLIGHTS
==========

Changes to the Mercury language:
* Infix `.' is now accepted as a module name separator.
* Field access syntax can now be used at the top-level in func and mode
  declarations and in the head of a clause for a user-defined field access
  function.
* We now support impure higher-order code.
* We now allow user-defined comparison predicates.
* User-defined equality and comparison predicates for a type are now
  required to be defined in the same module as the type.
* Experimental support for user-defined constrained types has been added.
* We now support the use of `..' as an infix operator.

Changes to the Mercury standard library:
* We've added several new modules: cord, for sequences with O(1) consing and
  concatenation, array2d, for two-dimensional arrays, and version_array,
  version_array2d, version_bitmap, version_hash_table, and version_store,
  implementing non-unique versions of these types supporting O(1) access for
  non-persistent use.  A new module term_to_xml has been added for converting
  arbitrary terms to XML documents. Two new modules, set_tree234 and
  set_ctree234, have been added to provide operations on sets with better
  worst-case behavior (but worse constant factors) than the current
  implementation. Ten new modules, svarray, sveqvclass, svmap, svmulti_map,
  svbimap, svset, svbag, svqueue, svrelation and svvarset now provide more
  convenient ways to update arrays, equivalence classes, maps, multi_maps,
  bimaps, sets, bags, queues, relations and varsets in code that uses state
  variables.
* New procedures have been added to many of the existing standard library
  modules.  Most notably, these include procedures for creating
  directories and symbolic links, for checking file types and file
  accessibility, for detecting floating point infinities and NaNs.
* The dir module now handles Microsoft Windows pathnames correctly.

Changes to the Mercury compiler:
* We have added optional support for a new type-accurate garbage collector
  as an alternative to using the Boehm et al conservative collector.
* Better support for incremental program development:
  there's two new compiler options, `--allow-stubs' and `--no-warn-stubs',
  to support execution of incomplete programs.
* There's a new warning option `--warn-dead-procs' for detecting unreachable
  code.
* It's now easier to use shared libraries on Linux/x86 systems
  with `mmc --make'.
* A new analysis: `--analyse-exceptions'.
  The compiler can use the results of this analysis to try and improve
  some optimizations.

Portability improvements:
* We have made the implementation compatible with gcc 3.4.
* Shared libraries now work on Mac OS X.

Changes to the Mercury debugger:
* Users can now arrange to have the goal and/or some variables printed
  every time execution arrives at a breakpoint.
* Users can now arrange to associate a condition with a breakpoint.
  Execution won't stop at the breakpoint if the condition is false.
* Users can now limit the output from stack traces.
* Users can now put breakpoints on unify and compare predicates.
* Users can now save runtime values to files.
* Users can now tell the declarative debugger to trust entire modules or 
  individual predicates or functions.
* The declarative debugger can track the origins of subterms.
* The declarative debugger can now use the divide-and-query search strategy.

Changes to the compiler back-ends:
* The .NET CLR back-end now bootstraps.
* Improvements to the Java back-end.
* The cost in disk space of enabling debugging is now much smaller.

Numerous minor improvements to the Mercury standard library.

Changes to the extras distribution:
* Added easyx, a new Xlib based graphics library suitable for implementing
  simple interactive graphical applications.
* Major improvements to the OpenGL binding.
* We've added a binding to GLUT (the GL utility toolkit).
* The OpenGL, GLUT and Tcl/Tk bindings have been ported to Mac OS X.

DETAILED LISTING
================

Changes to the Mercury language:

* The deprecated support for NU-Prolog style `when' declarations has been
  removed.	

* We have experimental support for user-defined constrained types, as
  documented in the reference manual section on "Solver types".  Variables
  of a constrained type can have constraints placed upon them before they
  are instantiated, allowing for various styles of constraint logic
  programming.

* We now allow user-defined comparison predicates, using the syntax
  :- type t ---> t where equality is unify_t, comparison is compare_t.

  See the "User-defined equality and comparison" chapter of the
  Mercury Language Reference Manual for details.

* User-defined equality and comparison predicates for a type are now
  required to be defined in the same module as the type.

* Infix `.' is now accepted as a module name separator.  Hence it is
  now possible to write io.write_string and list.member to mean the
  same thing as io__write_string and list__member, for instance.  This
  has required changing the associativity of `.' from xfy to yfx and
  from precedence 600 to 10.

* Field access notation can now be used at the top-level in func and
  mode declarations and in the head of a clause for a user-defined
  field access function.  That is, one can now write

  	:- func a ^ f(b) = c.
	:- mode a ^ f(b) = c is <detism>.
	A ^ f(B) = ...

* Mercury's support for impure code now also includes support for
  impure higher-order code.
  
  Specifically, impurity annotations can be used on higher-order types,
  lambda expressions, and higher-order calls; higher-order terms are
  permitted to call impure or semipure code provided that they are
  appropriately annotated as such.

  For details, see the "Higher-order impurity" section of the "Impurity"
  chapter of the Mercury Language Reference Manual.

* `..' is now accepted as an infix operator.  That means a list of
  successive integers can now be written as X .. Y.  For example:

	1 .. 5 = [1, 2, 3, 4, 5]


Changes to the Mercury standard library:

* We've add the function queue.from_list/1 as a synonym for
  queue.list_to_queue/1, function queue.to_list/1 (and predicate
  queue.to_list/2) as the converse of queue.from_list/1, queue.put_on_front/2
  (and predicate queue.put_on_front/3) to put items on to the front of a
  queue, queue.put_list_on_front/2 (and predicate queue.put_list_on_front/3)
  to put a list of items on to the front of a queue, and predicate
  queue.get_from_back/3 which removes the last element from a queue.

* We've added the function pqueue.from_assoc_list/1 as a synonym
  for pqueue.assoc_list_to_pqueue/1.

* We've added functions set.from_list/1 and set.from_sorted_list/1
  as synonyms for set.list_to_set/1 and set.sorted_list_to_set/1
  respectively.  Similar additions have also been made to the
  set_unordlist, set_ordlist and set_bbbtree modules.

* We've added some new higher-order predicates, rbtree.foldl2/6
  rbtree.foldl3 and rbtree.transform_value to the rbtree module.  The
  predicate rbtree.remove/3 has been deprecated.

* We've add some new predicates and functions to int.m.
  int.fold_up/4, int.fold_down/4, int.fold_up/5, int.fold_down/5,
  int.fold_up2/7 and int.fold_down2/7 support iteration over
  contiguous integer ranges.

* The predicate int.to_float/2 has now been deprecated.

* We've added a new library module, `array2d', for two-dimensional arrays.

* We've added a new module, cord, for sequences with O(1) consing and
  concatenation.  A cord is essentially a tree structure with data stored
  in the leaf nodes.  Joining two cords together to construct a new cord
  is therefore an O(1) operation.

* The dir module now handles Microsoft Windows pathnames correctly.

* dir__split_name and dir__basename are now semidet, not det.
  dir__split_name fails for root directories or pathnames not
  containing a directory separator.
  dir__basename fails for root directories.

* We've added some new predicates and functions to the dir module:
	basename_det/1,
	expand_braces/1,
	is_directory_separator/1,
	make_directory/4,
	make_single_directory/4,
	foldl2/6,
	parent_directory/0,
	path_name_is_absolute/1,
	path_name_is_root_directory/1,
	recursive_foldl2/7.

* We've added several new predicates to the io module:
	have_symlinks/0,
	make_symlink/4,
	follow_symlink/4,
	check_file_accessibility/5,
	file_type/4,
	input_stream_foldl2_io_maybe_stop/{6,7},
	binary_input_stream_foldl2_io_maybe_stop/{6,7}.

* We've added several new predicates and functions to the bimap module:
	det_insert,
	forward_search,
	reverse_search,
	from_corresponding_lists,
	map_keys,
	map_values,
	det_insert_from_assoc_list,
	det_insert_from_corresponding_lists,
	set_from_assoc_list,
	set_from_corresponding_lists,
	delete_key,
	delete_value,
	delete_keys,
	delete_values,
	overlay,
	apply_forward_map_to_list,
	apply_reverse_map_to_list,
	foldl

* We've added predicates relation__lookup_key_set_from/3 and
  relation__lookup_key_set_to/3.

* The type of the arguments giving the initial set of visited nodes
  to relation__dfs and relation__dfsrev has changed from set_bbbtree
  to sparse_bitset.

* Efficiency of the operations in the relation module has been
  greatly improved.

* Some predicates and functions have been added to the sparse_bitset module:
	to_set/1,
	from_set/1,
	member/2,
	foldl/4,
	foldr/4.

* exception.m now contains a predicate finally/6 which can be used to
  ensure that resources are released whether a called closure exits
  normally or throws an exception.

* exception.m now contains a predicate throw_if_near_stack_limits which
  can be used to prevent an application running out of stack space.

* We've changed the interface of exception.try_all/2 to separate
  exceptional results from normal results.

* We've added predicates multi_map.to_flat_assoc_list/2 and
  multi_map.from_flat_assoc_list/2. 

* Several new functions have been added to the string module, namely
  elem/2, unsafe_elem/2, chomp/1, lstrip/1, lstrip/2, rstrip/1, rstrip/2,
  strip/1, prefix_length/2, suffix_length/2, string/1, string/2, string/4
  and string.det_to_float/1. 

* We've added some new predicates, list__map2_foldl, list__map_foldl3,
  and list__foldl4 to list.m.

* We've added a predicate, list__cons/3 to list.m.  This is sometimes
  useful with higher-order code.  It can also be useful with state
  variables.  We've also added a function version. 

* We've added some new predicates, map__common_subset, map__foldl3,
  map__overlay_large_map and map__transform_value, to map.m.

* We've added a predicate, map_fold, to set.m.

* We've added a function, pred_to_bool, to bool.m.

* We've added the three predicates, `is_nan/1', `is_inf/1' and
  `is_nan_or_inf/1' to float.m.  These predicates are for use only on
  systems which support IEEE floating point arithmetic.

* We've added a function version of `hash_table__search/3'.

* We've added a predicate, copy_mutvar, to store.m.

* We've added a function, clk_tck, to time.m.

* builtin.m now contains types and insts `unify' and `compare' for use
  in defining user-defined equality and comparison predicates.

* builtin.m now defines insts `new' and `old' as synonyms for `free' and
  `any', respectively, since some of the HAL literature uses this terminology.
  Likewise it defines modes `no' for `new >> old' and `oo' for `old >> old'.

* We've fixed some problems with the use of `cc_nondet'.

  The incorrect cc_nondet modes of the following predicates have been removed:
	deconstruct.arg/4
	deconstruct.named_arg/4
	deconstruct.limited_deconstruct/6
	std_util.arg_cc/3
	std_util.argument_cc/3
	std_util.named_argument_cc/3
	std_util.limited_deconstruct_cc/5
  These have been replaced by cc_multi versions in which success or failure
  is indicated by returning a maybe type.

* We've added functions get_equivalent_elements, get_minimum_element and 
  remove_equivalent_elements to eqvclass.m.

* We've added semidet functions max_key and min_key to return the maximum and
  minimum keys in maps and 2-3-4 trees.

* We've added predicates member, remove_leq, remove_gt, foldl and filter
  to sparse_bitset.m.

* builtin.m now contains types and insts `unify' and `compare' for use
  in defining user-defined equality and comparison predicates.

* The following predicates, which were added in 0.11.0, have been deprecated:
	io.current_input_stream/3
	io.current_output_stream/3
	io.current_binary_input_stream/3
	io.current_binary_output_stream/3
  They were identical to the following long-existing predicates with
  similar names:
	io.input_stream/3
	io.output_stream/3
	io.binary_input_stream/3
	io.binary_output_stream/3

* The following functions have been added to the integer module:
	integer.zero/0
	integer.one/0
	integer.det_from_string/1
	integer.pow/2
  
  The predicate integer.pow/3 has been deprecated. 

* We've added some functions, rational.int/1, rational.from_integer/1,
  rational.from_integers/2 and rational.reciprocal/1 to rational.m
  The function rational.rational_from_integers/2 has been deprecated.

* A new module `term_to_xml' has been added to the standard library.  This
  module contains predicates to write arbitrary Mercury terms to an output
  stream as XML.  Automatic generation of DTDs for Mercury types is also
  supported.  Once a Mercury term is in XML it can be converted to many other
  formats such as HTML or XUL using an appropriate stylesheet.

Changes to the Mercury compiler:

* We have added optional support for a new type-accurate garbage collector
  as an alternative to using the Boehm et al conservative collector.

  The new collector is enabled by `--grade hlc.agc'.
  For details about how it works, see the paper
  "Accurate garbage collection in an uncooperative environment"
  which is available via our web page.

  Note that the new collector is a very naive copying collector, and still
  has a number of serious limitations which may make it undesirable for
  most applications.  It only works with `--high-level-code'.  The heap
  size is fixed at program startup; the collector does not attempt to
  resize the heap.  It does not do cheap heap reclamation on backtracking.
  There is no support for passing terms on the Mercury heap to C code.
  In most cases, the Boehm et all conservative collector will perform better.

* There's a new warning option `--warn-dead-procs' which can be used
  for detecting unreachable code.

  This is not yet enabled by default, because it can cause some spurious
  warnings in modules containing code which mixes Mercury clauses and
  `pragma foreign_proc' declarations for the same procedure.

* `mmc --make' now works correctly with Microsoft Visual C++.

* It's now easier to use shared libraries on Linux/x86 systems with
  `mmc --make'.  See the documentation for the `--mercury-linkage'
  and `--linkage' options and the `MERCURY_LINKAGE' Mmake variable
  in the Mercury User's Guide.

* The behaviour of the `--pre-link-command' and `--extra-init-command'
  options has changed.  They now take a command which will be passed
  the names of all source files in the program or library, with the
  name of the main module's source file passed first.
  See the "Build system options" section of the "Invocation" chapter
  of the Mercury User's Guide for details.

* It is now possible to reconfigure an existing Mercury installation
  to use a different C compiler.  See the "C compilers" chapter
  of the Mercury User's Guide for details.

* Inlining of builtins can now be disabled using the `--no-inline-builtins'
  option.  This is done by default when debugging, as without this option the
  execution of builtins is not traced.

* The Mercury compiler now uses `.' and not `:' as the module separator
  in all output.

* The environment variables which were previously used to override the
  location of the standard library (MERCURY_ALL_C_INCL_DIRS,
  MERCURY_ALL_MC_C_INCL_DIRS, MERCURY_INT_DIR, MERCURY_C_LIB_DIR,
  MERCURY_MOD_LIB_MODS, MERCURY_TRACE_LIB_MODS) have been removed.

* There is a new analysis: `--analyse-exceptions'.  This identifies
  predicates that will not throw an exception.  This information is
  made available to the optimizing passes of the compiler. 

Portability improvements:

* We have made the implementation compatible with gcc 3.4.
* Shared libraries now work on Mac OS X.

Changes to the Mercury debugger:

* Users can now limit the output from stack traces.

  The mdb command `stack' now takes an optional integer argument that
  specifies the maximum number of stack frames to be printed.

* Users can now put breakpoints on unify and compare predicates.

  The syntax for procedure specifications now has provision for specifying
  unify and compare predicates.

* Users can now save runtime values to files.

  We've added a new mdb command, `save_to_file', that saves a specified term
  to a specified file.

* The declarative debugger can now be told to trust entire modules or 
  individual predicates or functions using the `trust' mdb command.

* The declarative debugger can now also tell you where a value
  appearing in an atom came from (i.e. the call which constructed the value).

* The declarative debugger also now supports a divide-and-query search mode.
  You can tell the declarative debugger to use this search mode by invoking
  it with the command `dd -s divide_and_query'.

* The "pretty" and "raw_pretty" print formats have had their names swapped, so
  the "pretty" format is now prettier than the "raw_pretty" format.

Changes to the compiler back-ends:

* The .NET CLR back-end now bootstraps.

  We've fixed a lot of bugs, and implemented a lot more of the Mercury
  standard library.  As well as being able to bootstrap in grade `il',
  we also now pass more than 90% of the applicable tests in the
  Mercury test suite.  See README.DotNet for details.

* Improvements to the Java back-end.

  We've fixed a lot of bugs, and implemented a lot more of the Mercury
  standard library.  See README.Java for further details on the status
  of this backend.

Mercury 0.12.1, 21 November 2005
--------------------------------

This release is primarily a bug-fix release.
It fixes a problem with predicates that have existentially typed
arguments, makes sure that I/O tabling does not inadvertently
inline predicates that have a `:- pragma no_inline' declaration
attached to them, and makes various improvements to the MS-Windows
ports.

Mercury 0.12.2, 25 January 2006
-------------------------------

This release fixes some bugs with `mmc --make' and `--smart-recompilation'.


Mercury 0.13.0, 14 September 2006
---------------------------------

HIGHLIGHTS
==========

Changes to the Mercury language:
* The Mercury typeclass system now supports functional dependencies.
* A new language construct allows programmers to promise that any given
  goal is pure or semipure.
* Two new language constructs allow programmers to promise that all solutions
  of a given goal are equivalent with respect to the relevant equality
  theories.
* We now have support for optional module initialisation and finalisation.
* We now have support for module-local mutable variables.
* We now have support for recognizing switches in which multiple switch arms
  have shared code.
* A new pragma allows programmers to promise that in a predicate or function
  defined by mode-specific clauses, the mode-specific definitions have
  equivalent semantics.
* We now allow users to control how each argument of a `pragma memo' predicate
  is tabled.
* Support for the old-style lambda, mode and pragma syntax has been removed.
* ':' is now the type qualification operator, not a module qualifier.
* To ensure soundness, goals in negated contexts using non-local variables
  with dynamic modes (inst "any") must now be marked as impure.

Changes to the Mercury standard library:
* We have removed the predicates dealing with runtime type information (RTTI)
  from std_util.m. Any users impacted by this change should look for required
  functionality in the construct, deconstruct and type_desc modules of the
  standard library, in forms that have been mostly unchanged since the
  0.11 release. In most cases, the differences are quite minor, but provide
  more expressive power.
* We have moved the all-solutions predicates from std_util.m into a new
  library module, solutions.m.  These predicates are still available in
  std_util.m but these versions are now deprecated.
* We have moved the univ type, and related predicates, from std_util.m
  into a new library module, univ.m.
* We have moved the maybe type, and related predicates, from std_util.m
  into a new library module, maybe.m.
* We have moved the pair type, and related predicates, from std_util.m
  into a new library module, pair.m.
* We have moved the unit type from std_util.m into a new library module,
  unit.m.
* We have made the predicates semidet_succeed/0, semidet_fail/0 and
  cc_multi_equal/2 into builtins.  Formerly these were exported by std_util.m.
* We have added an `injection' module, for reversible maps that are injective.

Changes to the Mercury compiler:
* The compiler now generates error messages for mismatches between format
  strings and lists of values to be printed in calls to string.format and
  io.format.
* The compiler now generates better error messages for determinism errors
  involving single-solution contexts.
* We have significantly improved the compiler's performance on predicates
  with many clauses.
* We have deleted the old --split-c-files option, as it conflicted with the
  implementation of module initialisation and finalisation.

Portability Improvements:
* We've ported Mercury to the x86_64 (AMD64 / Intel EMT64) architecture.
* We've made the implementation compatible with gcc 4.1.

Changes to the Mercury debugger:
* Users can now see a listing of the source code lines referred to by the
  current environment (see the documentation for the `list' command in
  the Mercury Users' Guide).
* Users can now keep hold of a term, referring to it even when execution has
  left the goal at which the term was available as the value of a program
  variable.
* Users can now see the set of places where two terms differ from each other.
* The `set' command has been replaced by several other commands: the `format',
  `format_param', `list_context_lines', `list_path', `xml_browser_cmd',
  `xml_tmp_filename', `fail_trace_counts', `pass_trace_counts' and
  `max_io_actions' commands.
* The `save_to_file' command has been renamed the `dump' command.
* The `save' command now saves the entire persistent state of the debugger
  (with one small exception that cannot be reestablished by an mdb command from
  an arbitrary point of execution).
* The declarative debugger now supports an `undo' command, and allows users to
  select the search algorithm.
* The declarative debugger can now exploit information from the "code
  footprints" of passed and failed test cases to find bugs with fewer
  questions. We have also added two tools, mslice and mdice, to manipulate
  files containing such footprints.  See the "Trace counts" section of the
  Mercury User's Guide for details.
* Subterm dependency tracking in the declarative debugger is now significantly
  faster.

Changes to the compiler backends:
* We have implemented an optimization, --optimize-constructor-last-call,
  that can turn recursive calls that are followed only by unifications that
  construct output arguments into tail calls. This can reduce the stack space
  requirements of the predicates to which it is applicable from linear
  in the size of the input data to constant.
* We have implemented an optimization, --tuple, that can replace several
  arguments that are usually passed to predicates together with a single
  tuple. This can reduce parameter passing overheads.
* The compiler can now optimize away the trail manipulation code from parts
  of the program that cannot affect the trail.
* The compiler now optimizes away any instructions referring to values of dummy
  types. A type is a dummy type if it has one function symbol of arity zero.
* Higher order calls are now cheaper on the low level C backend.

Changes to the extras distribution:
* We've added a library of data structures designed to work with solver types. 
* We've added a library to generate Windows installer packages.
* We've added a program to generate optimisation flags for the compiler.


DETAILED LISTING
================

Changes to the Mercury language:

* We have added support for functional dependencies to the typeclass system.
  See the "Type classes" section of the Mercury Language Reference Manual for
  details.

* A new language construct allows programmers to promise that any given
  goal is pure or semipure. Given Goal, a goal that uses impure and/or
  semipure code, the goal

	promise_pure ( Goal )

  promises that Goal presents a pure interface. Given Goal, a goal that
  uses impure code, the goal

	promise_semipure ( Goal )

  promises that Goal presents a semipure interface.

* A new language construct allows programmers to promise that all solutions
  of a given goal are equivalent with respect to the relevant equality
  theories. Given Goal, a goal that computes values for two variables,
  X and Y, the goal

  	promise_equivalent_solutions [X, Y] ( Goal )

  promises that all solutions of Goal are equivalent with respect to the
  equality theories of the types of X and Y. This means that the
  promise_equivalent_solutions goal will be det if Goal is cc_multi,
  and that the promise_equivalent_solutions goal will be semidet if Goal
  is cc_nondet.

  A related language construct allows programmers to promise that although
  the solutions of a given goal are not necessarily equivalent with respect
  to the relevant equality theories, it is nevertheless immaterial which one
  is chosen in a particular context. The language construct is the `arbitrary'
  goal, and the context is established by a `promise_equivalent_solution_sets'
  goal. Consider a type representing maps from keys to values which is
  implemented using 2-3 trees. In such a type, the precise shape of the tree
  doesn't matter; two trees should be considered equal if they contain the same
  set of keys and map them to the same values:

  :- type tree23(K, V)
  	--->	two(tree23(K, V), K, V, tree23(K, V)
  	;	three(tree23(K, K, V, tree23(K, V), K, V, tree23(K, V))
	where equality is tree23_equal
	and comparison is tree23_compare.

  Two values of e.g. type tree23(int, string) may differ in their top level
  function symbol even through they denote the same map. Deconstructing a
  value of such a type may therefore theoretically yield either "two" or
  "three" as the top level function symbol, although in practice which one
  you get is determined by the concrete structure of the term. Unifications
  of such values with specific function symbols are therefore permitted only
  in committed choice contexts. Unfortunately, one cannot simply put the
  deconstruction into the scope of a promise_equivalent_solutions goal,
  since the solutions are not equivalent in all contexts. However, the
  solutions will be equivalent in *some* contexts. Consider this function
  to count the number of key-value pairs in the map:

  count(Tree) = Count :-
  	promise_equivalent_solution_sets [Count] (
		(
			arbitrary [Tree1, Tree2] (
				Tree = two(Tree1, _Key, _Value, Tree2)
			),
			Count = 1 + count(Tree1) + count(Tree2)
		;
			arbitrary [Tree1, Tree2, Tree3] (
				Tree = three(Tree1, _Key1, _Value1, Tree2,
					_Key2, _Value2, Tree3)
			),
			Count = 2 + count(Tree1) + count(Tree2) + count(Tree3)
		)
	).

  The construct `arbitrary [Tree1, Tree2] Goal', where Goal computes Tree1
  and Tree2, tells the compiler that it is OK to commit to the first solution
  of Goal, because regardless of whether the goal succeeds and if so with
  which values of Tree1 and Tree2, the set of solutions of the surrounding
  `promise_equivalent_solution_sets [Count] Goal' will not be affected.
  Regardless of whether Tree is bound to "two" or "three", the body of count
  will compute the right value for Count.

  A goal of the form `arbitrary [Vars] Goal' will be det if Goal is cc_multi,
  and it will be semidet if Goal is cc_nondet. Goals of that form may occur
  only inside `promise_equivalent_solution_sets' goals. There is no restriction
  on the determinism of `promise_equivalent_solution_sets' goals.

* We have added support for optional module initialisation.  See the 
  "Module initialisation" section of the Mercury Language Reference
  Manual for details.

* We have added support for optional module finalisation.  See the
  "Module finalisation" section of the Mercury Language Reference
  Manual for details.

* We have added support for module-local mutable variables.
  See the "Module-local mutable variables" section of the Mercury Language
  Reference Manual for details.

* We now have support for recognizing switches in which multiple switch arms
  have shared code. Where previously programmers had to write code like this

  (
  	X = a,
	... code for a ...
  ;
  	X = b(...),
	... code for b ...
  ;
  	X = c,
	... code for c ...
	... shared code ...
  ;
  	X = d(...),
	... code for d ...
	... shared code ...
  )

  to have the disjunction recognized as a switch on X, they can now write
  code like this:

  (
  	X = a,
	... code for a ...
  ;
  	X = b(...),
	... code for b ...
  ;
  	(
		X = c,
		... code for c ...
	;
		X = d(...),
		... code for d ...
	),
	... shared code ...
  )

* If a predicate or function is defined by mode-specific clauses, like this:

	reversible_sort(Raw::in, Sorted::out) :-
		list.sort(Raw, Sorted).
	reversible_sort(Raw::out, Sorted::in) :-
		is_sorted(Sorted),
		list.perm(Sorted, Raw).

  the compiler by default assumes that the definitions of the different modes
  have different semantics. Programmers can tell the compiler that the
  mode-specific definitions, though syntactically distinct, are semantically
  equivalent by including a pragma:

  :- pragma promise_equivalent_clauses(reverse_sort/2).

* To ensure soundness, goals in negated contexts using non-local variables
  with dynamic modes (inst "any") must now be marked as impure.

  If a goal uses a variable with a dynamic mode (inst "any"),
  and that goal occurs inside a negated context (such as the
  condition of an if-then-else, or a lambda expression),
  and the variable also occurs outside of that negated context,
  then the compiler will infer that goal to be impure,
  and so such goals must normally be marked as "impure".

  This change was required because Mercury implements negation using
  the standard negation-as-failure approach, which is not sound if the
  negated goal binds any non-local variables.

  As usual, the programmer can use "promise_pure" if they are
  sure that the goal is in fact pure, e.g. because they know that
  the goal inside the negation will not instantiate the variable.

Changes to the Mercury standard library:

* We have added the function `divide_equivalence_classes' to the `eqvclass'
  module.

* We have added an `injection' module, for reversible maps that are injective.

* We have added list.foldl_corresponding/5, list.foldl2_corresponding/7,
  list.map2_foldl2/8 and list.det_split_list/4.

* We have added string.word_wrap/2.

* We have added set.fold4/10.

* We have added semidet_true/0 and semidet_false/0 as synonyms for
  semidet_succeed/0 and semidet_fail/0.

* We have added impure_true/0 and semipure_true/0.

Changes to the Mercury compiler:

* The compiler now generates error messages for known mismatches between format
  strings and lists of values to be printed in calls to string.format and
  io.format, unless the user specifies the --no-warn-known-bad-format-call
  option.

  If the user specifies the --warn-unknown-format-call option, the compiler
  will also generate error messages for calls to string.format and io.format
  in which the format string or the structure of the list of values to be
  printed are not statically available.

Changes to the extras distribution:

* We've added a library of data structures designed to work for solver types. 

  The module extras/solver_types contains versions of the standard
  library's array, assoc_list, list and map modules that are designed to
  work with terms that have inst `any'.

* We've added a library to generate Windows installer packages.

  The directory extras/windows_installer_generator contains a library to
  generate Wix source files.  WiX is an XML language that is used to generate
  Microsoft Windows Installer (.msi) packages.

* We've added a program to generate optimisation flags for the compiler.

  The directory extras/gator contains a program to search for the
  optimal set of compiler flags for a given program.  The search
  algorithm used is a genetic algorithm, which can run in parallel over
  multiple hosts (by default, 1).


Mercury 0.13.1, 1 December 2006
-------------------------------

This release is primarily a bug-fix release.
The problems fixed include:

* polymorphic insts and mode-specific clauses did not work together.
* polymorphic insts and export pragmas did not work together.
* the compiler was not correctly enforcing the restriction that type
  variables in instance declarations should be distinct.
* the compiler sometimes performed superclass reduction incorrectly,
  causing it to reject valid typeclass constraints.
* installation of static archives on Mac OS X using mmc --make now
  works correctly.  Previously, static archives caused linking problems
  because the table of contents was not being updated after installation.
* non-exported typeclasses sometimes resulted in incomplete interface
  files being generated.

In addition to the above bug-fixes we have the following addition to the 
standard library.

Changes to the Mercury standard library:
* We have added a new module `rtree', that provides region trees.  
  These are a standard data structure for querying spatial information.


NEWS for Mercury 10.04.2, 5 October 2010
-----------------------------------------

This release is a bug-fix release.  A number of problems that caused
the compiler to abort have been fixed, some broken RTTI operations
in the java grade have been fixed, and we have added a workaround
for a problem with GCC version 4.4.


NEWS for Mercury 10.04.1, 30 August 2010
----------------------------------------

This release is primarily a bug-fix release.  It fixes some problems in
the code generator and the source-to-source debugger.  In addition,
there are some changes to the standard library.

Changes to the Mercury standard library:
* We have added cc_multi modes for map.foldl2/6 and tree234.foldl2/6.
* We have improved the performance of cords, hash tables and
version hash tables.


NEWS for Mercury 10.04, 19 July 2010
------------------------------------

Changes to the release numbering:

Stable releases are now numbered according to the year and month the
release was made, eg 10.04 is the release made in April 2010.
Previously the release naming scheme was 0.x where x was the x'th
major release of the system.

Changes to the Mercury language:

* We have removed support for automatic initialisation of solver variables.

* A new pragma, foreign_enum, allows the constructors of Mercury 
  enumeration types to be assigned values from foreign language code.

* A new pragma, foreign_export_enum, allows the constructors of Mercury
  enumeration types to be referred to in foreign language code.

* Some of the restrictions on typeclass instances have been relaxed, allowing
  support for polymorphic instances with functional dependencies.

* We now support trace goals, which can be used to print progress messages or
  log messages in the middle of arbitrary computations.

* Mutables can now be marked as constant, which is useful when working with
  constant data structures that cannot be conveniently represented as constant
  terms.

* Mutables can now be marked as thread-local, which can take on different
  values for each thread.

* promise_equivalent_solutions scopes (et al.) must now also list variables
  with inst any that may be constrained by the goal.

* We now support !X ^ field_list := Term as a synonym for
  !:X = !.X ^ field_list := Term.

* We now support higher-order `any' insts.

* We now support "implementation-defined literals", such as `$file', `$line',
  `$pred', which are useful for producing better run-time error messages.

* We now support currying of multi-moded predicates or functions when the
  mode to curry can be determined from the insts of the higher-order
  arguments.

* We now support `try' goals for catching exceptions.

Changes to the Mercury standard library:

* A new module, parsing_utils, has been added to provide support for
  implementing recursive descent parsers.

* The string.to_int family of predicates now fails (or throws an exception
  for the det_ versions) on base 10 numbers that do not fit in the range
  [int.min_int+1, int.max_int].  Numbers outside this range lead to overflow.
  Numbers not in base 10 are assumed to denote bit patterns and are not
  checked for overflow.

* A module for handling directed graphs, digraph.m, has been added.  This
  supersedes relation.m and svrelation.m in that it has a more consistent
  interface (which supports state variable notation), provides more type
  safety by using phantom types, and has a number of efficiency improvements.
  Further use of relation.m and svrelation.m is deprecated.

* An improved pretty printer module, pretty_printer.m, has been added.  This
  supersedes pprint.m in that it is more economical, produces better
  quality output (line overruns are completely avoided wherever possible),
  has better control over the amount of output produced, and supports
  user-specifiable formatting for arbitrary types.  Further use of pprint is
  deprecated.

* We have added extra modes to many of the fold style predicates in the
  library in order to better support (mostly-)unique accumulators.

* The foldr family of functions and predicates has been added to the map
  and tree234 modules. We have also extended the arities for map_foldl
  to map_foldl3 in both modules, and added versions of both map_foldl*
  and just plain foldl* in which the higher order argument does not take
  the key as an argument.

* We have added the following predicate to the int module:
	int.nondet_int_in_range/2.

* We have added the following functions to the integer module:
	integer.from_base_string/2
	integer.det_from_base_string/2

* We have added a new builtin predicate, unsafe_cast_any_to_ground/1, that
  can be useful when manipulating polymorphic values that have inst any.

* Predicates and functions which create strings from lists of characters
  now fail, throw an exception or return an error value if they find
  a null character. Unexpected null characters in strings are a potential
  source of security vulnerabilities.

  We have added the predicates string.semidet_from_char_list/2 and
  string.semidet_from_rev_char_list/2. These fail rather than throwing
  an exception if they find a null character.

* We have added string.remove_suffix_det, a version of string.remove_suffix
  that throws an exception if the suffix is not there.

* string.float_to_string now trims redundant trailing zeroes (although
  at least one fractional digit is always present).  This change affects the
  output from the debugger and io.print etc.

* The globals field in the I/O state is no longer unique.  The modes of
  the access predicates, io.set_globals/3 and io.get_globals/3 have been
  changed accordingly.

* We have added io.update_globals/3 which allows for atomic updates to
  the globals field in the I/O state in the presence of multiple threads.

* We have moved some of the concurrency primitives out of the extras
  distribution and into a new standard library module called `thread',
  and its submodules `thread.channel', `thread.mvar', and `thread.semaphore'.
  The predicates `thread.can_spawn', `thread.channel.try_take'
  and `thread.mvar.try_take' have also been added.

* Processes no longer terminate until all threads have finished.  Previously
  a process would terminate as soon as the original thread finished.

* The following predicate has been added to the set module:
	set.filter_map/3

* The following predicates have been added to the array modules:
	array.fold/4
	array.foldl2/6
	version_array.foldl/4

* The following predicates have been added to the list module:
	list.filter_map_foldl/5
	list.map_corresponding/4
	list.map2_foldl3/10
	list.map_corresponding_foldl/6
	list.map_corresponding_foldl2/8
	list.map_corresponding_foldl3/10
	list.map_corresponding3_foldl/7
	list.negated_filter/3
	list.foldl3_corresponding/9
	list.foldl_corresponding3/6
	list.foldl2_corresponding3/8
	list.foldl3_corresponding3/10
	list.foldl4_corresponding3/12
	list.split_upto/4
	list.contains/2
	list.find_index_of_match/4
	list.find_first_match/3

   We have also added versions of list.foldl/4 and list.foldr/4 that have
   determinism multi.

* The following functions have been added to the string module:
	string.split_at_separator/2
	string.split_at_char/2
	string.split_at_string/2
	string.remove_suffix_if_present/2
	string.remove_prefix_if_present/2
	string.is_all_digits/1
	string.all_match/2
	string.remove_prefix/3

* The following functions and predicates have been added to the bag module:
	bag.count/1
	bag.count_unique/1
	bag.member/2
	bag.member/3

* A unique mode has been added to cord.foldl_pred/4

* The following function has been added to the pqueue module
	pqueue.length/1

* The following predicate has been added to the maybe module
	maybe_is_yes/2

* We have changed the interface of the ops module to make lookups of operators
  more efficient.

* We have added string.c_pointer_to_string/{1,2} and string.from_c_pointer/1
  to convert c_pointers to a human readable form.

* The bitmap module has been modified and extended to make it more suitable
  for use as a general container for binary data.  See runtime/mercury_types.h
  for the new definition of the bitmap type for interoperability with C code.

  Bitmaps now have fields `bit', `bits' and `byte' for getting and
  setting a single bit, a group of bits (up to machine word size),
  and an aligned eight bit byte respectively.

  bitmap.get/2 has been deprecated; use bitmap.bit/2 instead.

  There is a new type bitmap.slice/0 to represent segments of bitmaps.

  There are new functions to move data around in bulk:
	copy_bits/5
	copy_bits_in_bitmap/4
	copy_bytes/5
	copy_bytes_in_bitmap/4

  Other added functions include:
  	shrink_without_copying/2
	append_list/1
	to_byte_string/1

* The operations in bitmap.m and version_bitmap.m which treat bitmaps
  as sets have been modified to throw an exception when the input
  bitmaps are not the same size.  Before this change bitmap.intersect/2
  computed the wrong answer when the input bitmaps were of different sizes.

* bitmap.difference/2 and version_bitmap.difference/2 now compute difference,
  not xor.  bitmap.xor/2 and version_bitmap.xor/2 have been added.

* bitmap.to_string(BM) now returns "<0:>" for an empty bitmap BM.
  Previously it returned "<0:00>".

* Version bitmaps now have a field `bit' for getting and setting a single bit.

  version_bitmap.get/2 has been deprecated; use version_bitmap.bit/2 instead.

* The io module now contains predicates io.read_bitmap/{4,5,6,7},
  io.write_bitmap{3,4,5,6} and io.read_file_as_bitmap/{3,4}.
  io.write_bytes/{3,4} are now marked as obsolete.  Note that the
  interface of io.read_bitmap/* has changed since the first release
  of the day implementation.

* There are new modules bit_buffer, bit_buffer.write and bit_buffer.read
  which give a bit-oriented interface to byte-oriented streams.

* There is a new typeclass stream.bulk_reader/5.  Instances of
  stream.bulk_reader/5 support reading of multiple items at once
  into a container such as an array or a bitmap.

* Comparison of version_arrays is now the same as for arrays.

* We have added predicates char.is_hex_digit/2 and char.int_to_hex_char/2.

* We have changed term.variable so that it records the context where
  the variable was used.  This required the backward mode of
  term.var_list_to_term_list to be removed.  The backwards mode is
  now accessed via term.term_list_to_var_list.

  There is a new function, get_term_context, to return the context of any term.

* We have renamed some library predicates whose names were ambiguous.

* The type software_error/0 has been moved from the require module into
  the exception module.

* construct.num_functors/1 now fails rather than returning -1 for types
  with no functors.  There is a new function construct.det_num_functors/1
  which aborts for types which do not have functors.

* We have added predicates deconstruct.functor_number/3 and
  deconstruct.deconstruct_du/4 which return functor numbers suitable
  for use by construct.construct, rather than functor strings.

* We have added a function construct.get_functor_lex/2 which converts
  an ordinal functor number into a lexicographic functor number.

* A new module string.builder has been added to the standard library.
  The new module provides instances of the stream typeclasses that can
  be used to build up a string using char and string writers.

* We have added the types `string.line' and `string.text_file' and made
  input streams instances of stream.reader/4 with those unit types.
  This means stream.get/4 can be used to efficiently read lines
  and files as string.

* We have added a predicate io.remove_file_recursively/4
  which can remove non-empty directories.

* We have added the predicates `dir.current_directory',
  `dir.relative_path_name_from_components'.

* We have added the predicates rev_list, split_last, get_first, get_last,
  filter, foldl_pred, foldr_pred, map_foldl, map_foldl[23], cord_list_to_cord
  and cord_list_to_list to the cord module.

* We have added two predicates that are useful for custom term construction:
	construct.find_functor/5
	type_desc.same_type/2 

* We have added new predicates to the tree234 and map modules for constructing
  2-3-4 trees and maps directly (without the creation of intermediate trees)
  from sorted lists:
	tree234.from_sorted_assoc_list/2
	tree234.from_rev_sorted_assoc_list/2
	map.from_rev_sorted_assoc_list/2
  map.from_sorted_assoc_list now also constructs the tree directly, so now
  it requires its input list to be duplicate-free.

* We have added tree234.map_values_only and map.map_values_only, which are
  versions of tree234.map_values and map.map_values which do not give the
  higher order argument the key associated with the value being transformed.

* We have replaced the hash_table and version_hash_table implementations
  with separate chaining schemes.  Consequently delete works, and double
  hashing predicates are not required.

* We have added optional synchronisation to the version_array and
  version_hash_table implementations.  They are now thread safe, but slower,
  by default.

* We have added a calendar module to the standard library. This module
  contains utilities for working with the Gregorian calendar.

Changes to the Mercury compiler:

* The Java backend has been substantially improved and now supports
  all the core features of the language and most of the standard library.
  The Java backend is roughly three times slower (after allowing for JIT)
  than the C backends.

  The Java backend improvements were contributed by Mission Critical IT
  <http://www.missioncriticalit.com/>.

* Interfacing with Mercury code from Java is also improved:
    * Polymorphic Mercury types are now translated to Java classes
      with generics, allowing for greater type safety in the Java code.
    * Exported Mercury procedures retain their argument order in the
      corresponding Java versions (output arguments are handled with a new
      Ref Java type).

* We have added support for trail segments, which allow programs to grow
  the trail on demand.

* Shared libraries are now used by default on Linux/x86 systems.

* Support for the reserve tag grades has been removed.

* We have added an Erlang back-end.

  The Erlang back-end was contributed by Mission Critical IT
  <http://www.missioncriticalit.com/>.

* In parallel grades we now support thread-local trailing.

* The compiler now issues a warning when an inst declaration is not
  consistent with any of the types in scope.

* We have added support for simultaneous execution of jobs with `mmc --make'.

* We have added support for `mmc --make' to recompile modules if options have
  changed.

* We have added an option for `mmc --make' to compile more recently modified
  source files first.

* We have added support for stack segments, which allows programs to grow
  stacks on demand.

* We have made it easier to use single-precision floats, which do not need
  to be boxed on 32-bit machines.

* A new option, `--generate-standalone-interface', simplifies the task
  of calling Mercury procedures from programs written in other languages.

* We have added a new option, `--inform-ite-instead-of-switch'. If this is
  enabled, the compiler will generate informational messages about
  if-then-elses that it thinks should be converted to switches for the
  sake of program reliability.

* We have removed support for Managed C++ as a foreign language for the IL
  backend.

* The width of error message lines can be adjusted with a new option,
  `--max-error-line-width'.

* Generation of 64-bit code on Mac OS X is now supported,
  i.e. the x86_64*apple*darwin* architecture is now supported.

* We have added another debugger, called the source-to-source debugger (ssdb).
  It is more limited than mdb, but it does work with backends other than the
  low-level C backend, e.g. the Java backend.

Changes to the Mercury deep profiler:

* The deep profiler now supports measuring a proxy for time: a counter that
  is incremented at each call. Since calls happen a lot more frequently than
  clock interrupts, this can yield useful profiles for shorter-running
  programs.

Changes to the samples directory:

* The samples directory now includes an example of how to implement a
  solver type.

Changes to the extras distribution:

* The extras distribution now includes a binding to the Allegro and
  AllegroGL game programming libraries.

* `mtogl', the Mercury binding to the Tk widget `togl' has been removed
  from the distribution.

Changes to the Mercury debugger:

* A `track' mdb command has been added.

* The `dd' command now accepts a `--reset-knowledge-base' option.

* You can now put breakpoints on individual events inside procedures.

* mdb now ignores lines beginning with a `#' character
  in sourced files.  This is useful for commenting mdb scripts.

DETAILED LISTING
================

Changes to the Mercury language:

* Support for the automatic initialisation of solver variables has been
  removed because it was source of confusing errors and was not used in
  practice anyway.

  A consequence of this is that solver types are now properly polymorphic,
  i.e. types like `foo(bar)' where:

  	:- solver type foo(T).
	:- solver type bar.
  
  are now supported.

* The new pragma, foreign_enum, can be used to establish a 
  mapping between values in a foreign language and the constructors
  of a Mercury enumeration type.  This can be useful when writing
  Mercury bindings to foreign code libraries.

  For example, 

  	:- type matrix_mode
		--->	texture
		;	modelview
		;	projection.

	:- pragma foreign_enum("C", matrix_mode/0, [
		texture    - "GL_TEXTURE" 
		modelview  - "GL_MODELVIEW"
		projection - "GL_PROJECTION"
	]).
  
  When passed to C foreign clauses, values of type matrix_mode/0 will have
  the corresponding C value specified by the foreign_enum pragma.

* The new pragma, foreign_export_enum, can be used to establish a
  mapping between the constructors of a Mercury enumeration type and
  a symbolic name for values of that type in the foreign language.

  For example, given the type

  	:- type status ---> ok ; error.

  the declaration

  	:- pragma foreign_export_enum("C", status/0, [prefix("STATUS_")]).
  
  allows code in C foreign_proc and foreign_code pragma bodies to refer
  to the value `ok' via the name `STATUS_ok' and to the value `error'
  via the name `STATUS_error'.

* The restriction on typeclass instances that all type variables appearing in
  the range of a functional dependency must be monomorphic has been relaxed.
  We now support cases where the type variables in the range are determined
  by the type variables in the domain, using the functional dependency
  information from any instance constraints.

  For example, given the typeclass

	:- typeclass foo(A, B) <= (A -> B).

  the following instance is now valid:

	:- instance foo(list(S), list(T)) <= foo(S, T).

  since the variable T, in the range, is determined from the variable S by the
  functional dependencies on the foo(S, T) constraint.

* A new language construct allows programmers to include debugging and/or
  logging code in the middle of arbitrary computations. Trace goals
  may have both compile time and run time conditions placed on their execution.
  However, if they are enabled, then they can perform I/O (even if the
  surrounding code can't); they can also access the values of mutables.

  Their capabilities, syntax and intended use are shown by the following
  example.

	:- mutable(logging_level, int, 0, ground, []).

	:- pred time_consuming_task(data::in, result::out) is det.

	time_consuming_task(In, Out) :-
		trace [
			compile_time(flag("do_logging") or grade(debug)),
			run_time(env("VERBOSE")),
			io(!IO),
			state(logging_level, !LoggingLevel)
		] (
			io.write_string("Time_consuming_task start\n", !IO),
			( !.LoggingLevel > 1 ->
				io.write_string("Input is ", !IO),
				io.write(In, !IO),
				io.nl(!IO)
			;
				true
			)
		),
		...
		% perform the actual task

* Higher-order terms can now have an `any' inst, which means that they can
  legally use non-local solver variables even though these variables may
  become further bound when in it is called.  Higher-order terms with an
  `any' inst cannot be called or applied in a negated context (a negation,
  the condition of an if-then-else, or the body of a higher-order expression
  that does not itself have an `any' inst).

  Such terms can be expressed by using `any_pred' and `any_func' in place of
  `pred' and `func', as in the following examples:

	AnyPred = (any_pred(X::ia, C::in) is semidet :-
			geqc(X, A, C)		% X >= A + C
		),
	...
	call(AnyPred, Z, 5)			% Z >= A + 5

	AnyFunc = (any_func(X::ia) = (Y::ia) is semidet :-
			X = Y + A
		),
	...
	Z = apply(AnyFunc, W)			% W = Z + A

* Try goals provide syntactic sugar for catching exceptions. An example is:

	:- pred p_carefully(io::di, io::uo) is det.

	p_carefully(!IO) :-
		(try [io(!IO)] (
			io.write_string("Calling p\n", !IO),
			p(Output, !IO)
		)
		then
			io.write_string("p returned: ", !IO),
			io.write(Output, !IO),
			io.nl(!IO)
		catch S ->
			io.write_string("p threw a string: ", !IO),
			io.write_string(S, !IO),
			io.nl(!IO)
		catch 42 ->
			io.write_string("p threw 42\n", !IO)
		catch_any Other ->
			io.write_string("p threw something: ", !IO),
			io.write(Other, !IO),
			% Rethrow the object.
			throw(X)
		).

Changes to the Mercury compiler:

* The option `--trail-segments', or grade component `.trseg', causes
  programs to execute using trail segments, where segments can be allocated
  at runtime, instead of using a fixed size trail.  This can prevent trail
  exhaustion, but execution time will be increased.

* There's a new back-end that targets Erlang.

  Compiler support for this new back-end is mostly complete,
  but large parts of the standard library are still not yet
  implemented for this new port.

  For more details, see the README.Erlang.

* The compiler now issues a warning when an inst declaration isn't
  consistent with any of the types in scope.

  This makes it easier to diagnose mode errors caused by insts that are not
  consistent with the type they are intended to be consistent with.

* Simultaneous execution of jobs with `mmc --make' can be enabled with
  the `--jobs <n>' option.

* `mmc --make' can record what compiler options were used for each module
  by enabling the `--track-flags' option.  Then `mmc --make' can know to
  recompile modules whose options have changed, even if the files haven't
  been touched.

* `mmc --make' can compile more recently modified files first, if the option
  `--order-make-by-timestamp' is enabled.

* The option `--stack-segments', or grade component `.stseg', causes
  programs to execute using stack segments, where segments can be allocated
  at runtime, instead of using fixed sized stacks.  The program won't run out
  of stack space and stack sizes can be much smaller, but execution time will
  be increased.

* Single-precision floats can now be selected for the C backends by using the
  `.spf' grade component, or passing the `--single-prec-float' option to the
  compiler.

* The option `--generate-standalone-interface', causes the compiler to
  create a "stand-alone" interface to the Mercury runtime and a set of
  Mercury libraries.  This interface allows programs written in languages
  such as C or C++ to initialise the Mercury runtime and libraries prior to
  calling any foreign exported procedures defined in those libraries.

* We have removed support for Managed C++ as a foreign language for the IL
  backend.  This was desirable because it wasn't used, because it has been
  deprecated by Microsoft, and because it complicated the dependencies for the
  IL backend.

Changes to the Mercury standard library:

* The predicates io.seek_binary/5 and io.binary_stream_offset/4 have been
  deprecated.  They have been replaced by the predicates:
  io.seek_binary_input/5, io.seek_binary_output/5,
  io.binary_input_stream_offset/4 and io.binary_output_stream_offset/4.

Changes to the Mercury debugger:

* A `track' mdb command has been added.  This command invokes the declarative
  debugger and executes it's `track' command, before returning to the mdb
  prompt.

* The `dd' command now accepts a `--reset-knowledge-base' option. 
  This option resets the declarative debugger's knowledge base of previous
  question answers.

* You can now put breakpoints on individual events inside procedures.
  Commands of the form "break <procedure-specification> <portname>"
  will cause execution to stop only at the specified port in the specified
  procedure. If there is more than event of the given port type in the
  procedure, mdb will prompt the user to select one.

* mdb now ignores lines beginning with a `#' character
  in sourced files.  This is useful for commenting mdb scripts.


NEWS for Mercury 11.01, 27 April 2011
-------------------------------------

Upcoming changes to the Mercury standard library:

* In the next release after 11.01 we intend to change the argument order of
  many of the predicates in the standard library so as to make them more
  conducive to the use of state variable notation.

  For example, map.insert/4 will be changed from
      
     map.insert(map(K, V)::in, K::in, V::in, map(K, V)::out) is semidet.

  to

     map.insert(K::in, V::in, map(K, V)::in, map(K, V)::out) is semidet.

  As part of this change, the sv* modules will be deprecated; they will
  be removed from the library in a later release.

  Note that none of these upcoming changes affect the 11.01 release.


HIGHLIGHTS
==========

Changes to the Mercury language:

* We have added two new kinds of scopes to the language.

  A scope introduced by one of the new keywords require_det, require_semidet,
  require_multi, require_nondet, require_cc_multi, require_cc_nondet,
  require_erroneous and require_failure, as the name implies, requires
  the goal inside the scope to have the given determinism.
  
  A scope introduced by the keyword require_complete_switch requires the
  goal inside the scope, if it is a switch on the variable named by the scope,
  to be a complete switch, i.e. to have arms for all the function symbols
  in the type of the switched-on variable.

Changes to the Mercury standard library:

* We have added semidet modes for hash_table.fold/4 and
  version_hash_table.fold/4.

* We have added new predicates and functions added to the assoc_list module.
  The predicates map_keys_only/3 map_values_only/3 and map_values/3 complement
  the functions of the same name. The predicates filter/3, negated_filter/3,
  filter/4, merge/3, which also have function versions, do jobs very similar
  to the predicates of the same name in the list module, but do the relevant
  operations on keys instead of entire list elements.

+ We have moved the lazy evaluation module out of the extras distribution and
  into a new standard library module called `lazy'.  It has also been made
  backend-agnostic.

* We have made changes to the list module of the standard library:

  + We added a new predicate list.member_index0/3.  It is like list.member/2
    except that it also takes a parameter representing the zero-based index of
    the element within the list.

  + We added a new predicate list.map3_foldl/7 which maps over a list producing
    three lists and one folded value.
  
  + We added semidet modes with unique accumulators for list.foldl3/8,
    list.foldl4/10, list.foldl5/12, and list.foldl6/14.

* We have added the predicates divide/4 and divide_by_set/4 to the tree_bitset
  module of the standard library.

* We have added the predicates set_ctree234.member/2 and
  set_ctree234.non_empty/1.  We have add the function
  set_ctree234.from_list/1.

* We have added the predicate set_bbbtree.count/2 and the
  function set_bbbtree.count/1.  These replace the predicate
  set_bbbtree.size/2 which is now deprecated.  

* We have added the predicate set_ordlist.non_empty/1.

* We have added the predicate set_tree234.non_empty/1 and the
  function set_tree234.from_list/1.

* We have added the predicates set_unordlist.non_empty/1 and
  set_unordlist.count/2, and the function set_unordlist.count/1. 

* All of the modules in the standard library that implement the set ADT,
  (set, set_ordlist, set_unordlist, set_bbbtree, set_tree234,
  and set_ctree234), now support folding over sets with up to six
  accumulators.  Modes that provide unique and mostly-unique accumulators
  for set fold have also been added.

* We have made the following changes to the array module of the standard
  library:

  + We have added the functions unsafe_elem/2 and append/2.

  + We have added the predicates svset/4, unsafe_svset/4, foldl2/4, foldl2/6,
    foldr/4, foldr2/6, map_foldl/5, map_corresponding_foldl/6, and member/2.

* We have added the predicates version_array.foldl2/6, version_array.foldr/4,
  and version_array.foldr2/6 to the standard library.

* We have added semidet modes with unique and mostly-unique accumulators for
  the following predicates:
	bimap.foldl2/6
	bimap.foldl3/8
	cord.foldl_pred/4
  	cord.map_foldl/5
	list.map_corresponding_foldl2/8
	list.map_corresponding_foldl3/10
	list.map_corresponding3_foldl/7
	map.foldl3/8

* We have added the predicate unsorted_aggregate2/6 to the solutions module
  of the standard library.

* We have added several predicates and functions to the require module
  of the standard library. The predicates sorry/2 and sorry/3 report
  the absence of a feature, while the predicates unexpected/2 and unexpected/3
  report an internal error in the program; all have function versions too.
  The predicate expect/3 calls unexpected if a condition isn't satisfied.
  We now have expect/4 as well as expect/3. For expect/4 as well as expect/3,
  the first and last arguments are the expected condition and the error message
  respectively, but with expect/4, there are two arguments in the middle
  to specify the location of the error (normally the name of the module
  and of the predicate respectively). We also added expect_not/3 and
  expect_not/4, which are like expect/3 and expect/4 respectively,
  except they expect the condition to be *false*, not true.

Changes to the Mercury compiler:

* We have added a new backend that generates C#.

* Support for building and linking against frameworks on Mac OS X has
  been improved.

Changes to the extras distribution:

* We have added a binding to the Cairo 2D graphics library.


DETAILED LISTING
================

Changes to the Mercury compiler:

* We have added a new backend that generates C#.
  For more details, see the README.CSharp.

  The new backend was contributed by Mission Critical IT
  <http://www.missioncriticalit.com/>.

* We have added two new options, --framework and --framework-directory
  in order to simplify building and linking against frameworks on Mac OS X.
  (-F is supported as a synonym for --framework-directory.)

* Switches on strings in which all output arguments in all switch arms are
  bound to constants are now implemented using lookup tables on the LLDS
  back end. This should make the generated code more compact as well as faster.


NEWS for Mercury 11.07.2, 3 August 2012
---------------------------------------

This is a bug-fix release: it fixes the following:

* The name of the C# compiler in newer versions of Mono is now recognised.

* The configure script now distinguishes between the MS C# compiler and the
  Chicken Scheme compiler.

* Thread local storage now works on Mac OS X when using clang as the C
  compiler.

* Various build issues on Solaris 10 have been resolved.

* The --erlang option now sets all the relevant grade components correctly.

* The binary input and output streams are now set to binary translation
  mode when using the MSVC CRT.

* The standard library's lexer module now correctly tokenizes binary, octal
  and hexadecimal integer literals.

* Some bugs in the UTF-8 routines used by the C grades have been fixed.
  Also, the performance of these routines has been improved.

* The exit status is now correctly set when --warn-non-tail-recursion
  is being used.

Changes to the Mercury standard library:

* We have added additional modes to map.foldr/4, map.foldr/6 and the
  list.foldl<N>_corresponding3 predicates.

* The predicates parsing_utils.float_literal/3 and
  parsing_utils.float_literal_as_string/4 now accept an optional plus sign in
  the exponent.

* The obsolete predicates string.remove_suffix_det/2 and string.index_det/2
  have been removed.

* The predicate time.times/4 is now implemented for the csharp grade.

* The version_hash_table module is now supported in the csharp and java
  grades.

* The unsafe versions of the predicates in the array2d module are now more
  efficient.

Changes to the Mercury compiler:

* The compiler now outputs #line directives in C# code it generates.


NEWS for Mercury 11.07.1, 2 March 2012
--------------------------------------

This is a bug-fix release.  It fixes a problem that was preventing the standard
library from building in the csharp grade on Cygwin and also adds support for
using GCC in C99 (or GNU99) mode with Mercury.  In addition, there are a number
of minor changes to the standard library.

Changes to the Mercury standard library:

* We have added the predicate map.keys_and_values/3.

* We have added the predicates set.is_singleton/2, set_bbbtree.is_singleton/2,
  set_ctree234.is_singleton/2 and set_unordlist.is_singleton/2.

* We have added the function list.foldl_corresponding/4.


NEWS for Mercury 11.07, 22 December 2011
----------------------------------------

HIGHLIGHTS
==========

Changes to the Mercury language:

* The `char' type now represents a Unicode code point.

* Unicode characters can now be encoded in string literals using an
  escape sequence.  The escape sequence \uXXXX (or \UXXXXXXXX), where XXXX
  (or XXXXXXXX) is a Unicode character code in hexadecimal, is replaced with
  the corresponding Unicode character.

* Expressions used to initialise mutables may now contain impure
  or semipure function calls.

Changes to the Mercury standard library:

* We have improved Unicode support in the standard library.

* We have deprecated substring procedures which take start and count
  arguments in favour of procedures which take start and end arguments.

* We have changed the argument order of many of the predicates in the array,
  bag, bimap, eqvclass, map, multi_map, queue and set modules in order
  to make them more conducive to the use of state variable notation.

Changes to the Mercury compiler:

* Support for using clang (<http://clang.llvm.org/>) as a C compiler with
  Mercury has been added.  See README.clang for further details

* We have significantly improved support for using Microsoft Visual C as
  a C compiler with Mercury.  See README.MS-VisualC for further details.

* The Java code generated by the Java backend is now compatible with Java 1.7.

Changes to the Mercury profiler:

* We have added a new form of profiling, memory retention profiling, to mprof
  that allows it to generate reports describing the origin of objects that
  are live on the heap at selected points in a program.
  See the Mercury User's Guide for details.
  

DETAILED LISTING
================

Changes to the Mercury standard library:

* We have improved Unicode support in the standard library.

    + Procedures in the string module now understand either UTF-8 or
      UTF-16 encodings, depending upon the backend.

    + Text I/O routines now read and write files in UTF-8 encoding.

* We have deprecated substring procedures which take start and count
  arguments in favour of procedures which take start and end arguments.
  The new procedures are more convenient to call, make more sense with
  variable-width character encodings, and pave the way for potentially
  replacing string offsets with an abstract type.

* We have added additional modes for set.map/3 and set.map_fold/5.

* The argument order of the following predicates has been changed so as to
  make them more conducive to the use of state variable notation:
  array.set/4, array.semidet/4, array.slow_set/4, array.semidet_slow_set/4,
  array.resize/4, array.shrink/3, bag.insert/3, bag.insert_list/3,
  bag.insert_set/3, bag.remove/3, bag.det_remove/3, bag.remove_list/3,
  bag.remove_set/3, bag.delete/3, bag.remove_all/3, bag.delete_all/3,
  bag.remove_smallest/3, bimap.insert/4, bimap.det_insert/4, bimap.set/4,
  eqvclass.ensure_element/3, eqvclass.new_element/3,
  eqvclass.ensure_equivalence/4, eqvclass.new_equivalence/4,
  eqvclass.remove_equivalent_elements/3, map.insert/4, map.det_insert/4,
  map.det_insert_from_corresponding_lists/4,
  map.det_insert_from_assoc_list/3, map.set_from_corresponding_lists/4,
  map.set_from_assoc_list/3, map.update/4, map.det_update/4, map.delete/3,
  map.delete_list/3, map.remove/4, map.det_remove/4, map.remove_smallest/4,
  multi_map.insert/4, multi_map.det_insert/4, multi_map.update/4,
  multi_map.det_update/4, multi_map.det_replace/4, multi_map.set/4,
  multi_map.add/4, multi_map.delete/4, multi_map.remove/4, and
  multi_map.remove_smallest/4, queue.put/3, queue.put_list/3,
  queue.get/3, queue.delete_all/3, queue.put_on_front/3,
  queue.get_from_back/3, queue.put_list_on_front/3,
  queue.get_from_back/3, rbtree.insert/4, rbtree.update/4, rbtree.set/4,
  rbtree.delete/3, rbtree.remove_smallest/4, rbtree.remove_largest/4,
  set.insert/3, set.insert_list/3, set.delete/3, set.delete_list/3,
  set.remove/3, set.remove_list/3, set.remove_least/3, tree234.insert/4,
  set_bbbtree.insert/3, set_bbbtree.insert_list/3, set_bbbtree.delete/3,
  set_bbbtree.delete_list/3, set_bbbtree.remove/3, set_bbbtree.remove_list/3,
  set_bbbtree.remove_least/3, set_bbbtree.remove_largest/3,
  set_ordlist.insert/3, set_ordlist.insert_list/3, set_ordlist.delete/3,
  set_ordlist.delete_list/3, set_ordlist.remove/3, set_ordlist.remove_list/3,
  set_ordlist.remove_least/3, set_unordlist.insert/3,
  set_unordlist.insert_list/3, set_unordlist.delete/3,
  set_unordlist.delete_list/3, set_unordlist.remove/3,
  set.unordlist.remove_list/3, set_unordlist.remove_least/3,
  sparse_bitset.insert/3, sparse_bitset.insert_list/3, sparse_bitset.delete/3,
  sparse_bitset.delete_list/3, sparse_bitset.remove/3,
  sparse_bitset.remove_list/3, sparse_bitset.remove_leq/3,
  sparse_bitset.remove_gt/3, sparse_bitset.remove_least/3,
  term.create_var/3, tree234.set/4, tree234.remove/4, tree234.remove_smallest/4,
  tree234.update/4, tree_bitset.insert/3, tree_bitset.insert_list/3,
  tree_bitset.delete/3, tree_bitset.delete_list/3, tree_bitset.remove/3,
  tree_bitset.remove_list/3 and tree_bitset.remove_least/3

* We have added the following new functions for creating singleton
  maps: bimap.singleton/2, injection.singleton/2, map.singleton/2,
  rbtree.singleton/2 and tree234.singleton/2 .

* The following procedures have been added to the standard library:

	array2d.init/3
	bitmap.init/2
        cord.init/0
	hash_table.init/3
	hash_table.init_default/1
	mvar.init/0
	mvar.read/4
	mvar.try_put/5
	store.init/1
	semaphore.init/1
	semaphore.init/3
	version_array.init/2
	version_array2d.init/3
	version_bitmap.init/2
	version_hash_table.init/3
	version_hash_table.unsafe_init/3
	version_hash_table.init_default/1
	version_hash_table.unsafe_init_default/1
	version_store.init/0

   They replace the following procedures, which are now obsolete and will be
   removed in a later release:

	array2d.new/3
	bitmap.new/2
	hash_table.new/3
	hash_table.new_default/1
	semaphore.new/1
	semaphore.new/3
	store.new/1
	string.foldl2_substring/8
	string.foldl_substring/5
	string.foldl_substring/6
	string.foldr_substring/5
	string.foldr_substring/6
	string.substring/3
	string.substring/4
	string.unsafe_substring/3
	string.unsafe_substring/4
	version_array.new/2
	version_array2d.new/3
	version_bitmap.new/2
	version_hash_table.new/3
	version_hash_table.new_default/1
	version_hash_table.unsafe_new/3
	version_hash_table.unsafe_new_default/1
	version_store.new/0

* The following procedures are have been deprecated and will be removed in
  a future release: dir.basename_det/1, list.replace_nth_det/3,
  list.replace_nth_det/4, list.index0_det/2, list.index1_det/2,
  list.index0_det/2, list.index1_det/2, list.last_det/1, stack.top_det/1,
  stack.pop_det/3, string.remove_suffix_det/2, string.index_det/2,
  string.index_det/3, string.set_char_det/3, string.set_char_det/4 and
  type_desc.ground_pseudo_type_desc_type_to_type_desc_det/1.

  The versions of these procedures that use "det_" as a prefix should be
  used instead.  (We have added these where they did not already exist.) 

* The deprecated modules graph, group and relation are no longer included
  in the library.

* The following predicates have been added to the modules that provide sets
  in the library: set.is_empty/1, set_bbbtree.is_empty/1,
  set_ctree234.is_empty/1, set_ordlist.is_empty/1, set_tree234.is_empty/1,
  set_unordlist.is_empty/1, sparse_bitset.is_empty/1, tree_bitset.is_empty/1,
  set.filter/4, set_bbbtree.filter/3, set_bbbtree.filter/4, set_ctree.filter/3,
  set_ctree.filter/4, set_ordlist.filter/3, set_ordlist.filter/4,
  set_tree.filter/3, set_tree.filter/4, set_unordlist.filter/3,
  set_unordlist.filter/4, sparse_bitset.filter/3, sparse_bitset.filter/4,
  tree_bitset.filter/3, tree_bitset.filter/4.

  All the is_empty predicates are synonyms for the existing empty/1 predicates
  in those modules.

* We have added the predicate pqueue.det_remove/4.  It is like pqueue.remove/4
  except that it throws an exception instead of failing if the priority queue
  is empty.

* The new modules svlist, svstack and svpqueue provide state variable friendly
  versions of predicates in the list, stack and pqueue modules.
  (As with the other sv* modules these modules are intended to pave the
  way for an eventual change of the predicate argument ordering in the
  list, stack and pqueue modules.)

* We have added additional modes with unique and mostly-unique accumulators
  to rbtree.foldl/4, rbtree.foldl2/6 and tree_bitset.foldl/4.

* A new function, array.generate/2, and new predicate, array.generate_foldl/5,
  can be used to create and fill new arrays.

* We have added the new predicate assoc_list.foldl_keys/4 and
  assoc_list.foldl_values/4 for folding over just keys or values
  an association list.

* We have added the new function array.from_reverse_list/1, which creates
  a new array from a list with the elements of the array occurring in
  the reverse order to that of the list.

* We have added a version of the field update function for arrays that
  omits the bounds check, 'unsafe_elem :='/3.

* We have added the predicates array.foldl3/8, array.foldl4/10, array.foldl5/12,
  array.foldr3/8, array.foldr4/10 and array.foldr5/12.

* We have added the predicate maybe.map_fold3_maybe/9 and also added
  additional modes for maybe.fold_maybe/4, maybe.map_fold_maybe/5,
  and maybe.map_fold2_maybe/7.

* The implementation of multi_map.det_update/4 has been fixed so that it
  conforms to the documented behaviour.  The new predicate multi_map.replace/4
  has been added.

* We have added a version of math.sqrt/1, math.unchecked_sqrt/1, that omits
  the domain check.


.NET CLR back-end history
-------------------------
As mentioned above, we started working on the MLDS back-end in July 1999.
The MLDS back-end was also used as the basis of the .NET CLR and Java
back-ends.  An initial version of the compiler, which was capable of
compiling programs such as "hello world" and "eliza", was committed to
the public Mercury CVS repository on 14th Oct, 2000.  The compiler first
completed a successful bootstrap using the .NET CLR back-end grade on
February 21st, 2003.