Switch branches/tags
Commits on Oct 24, 2003

    telent committed Oct 24, 2003
    	Removed debugging message about unexpectedly enabled signals.
    	These days we're older and wiser and expect that stuff to
    	contrib/experimental-thread.patch, as the name suggests, is
    	a patch that may help thread stability problems, but hasn't
    	been sufficiently well tested to get merged into SBCL proper
    	this late int he release cycle
Commits on Oct 22, 2003

    csrhodes committed Oct 22, 2003
    	CLISP build "fixes"
    	... don't emit structure-sharing code from macros
    		(fixes "SB-IMPL::|,| does not name a function" clisp
    	... rewrite bad LOOP as per TFBurdick sbcl-devel 2003-10
    	I say `"fixes"' because new sbcl code has exposed some
    	clisp bugs, so the resulting codebase doesn't build from
    	clisp-2.31; however, I have patched my clisp and it works to the
    	extent that the resultant sbcl binary only fails one test.

    csrhodes committed Oct 22, 2003
    	Solaris fixes (from Nikodemus Siivola sbcl-devel 2003-10-22)
    	... sh(1) apparently doesn't like $(foo).  Make it `foo`.
    	... allow building from cmucl/sparc, by suppressing underflow
    	... /dev/random doesn't exist.  /dev/zero better had, though
Commits on Oct 20, 2003

    csrhodes committed Oct 20, 2003
    	... add a test
    	... some other BUGS filtrage (BIT-VECTOR is required to be
    		disjoint from SIMPLE-VECTOR; other fixed bugs)

    csrhodes committed Oct 20, 2003
    	Fix bug 213a
    	... CONS-TYPE-LENGTH-INFO to walk CONS-TYPE lists
    	... delete the neat but ultimately flawed (CONS NIL T) test
    		and use a proper test instead
    	... test suite additions.
    	Add idea from Michael Hudson (sbcl-devel 2003-08-26) to exit
    		early from Darwin compilations when the stack size
    		limit is too small.
Commits on Oct 19, 2003

    csrhodes committed Oct 19, 2003
    	One last little PCL frob
    	... when declaring defgenerics, don't foolishly assume that all
    		extended function names are of the form (SETF ..)
    	... now internal PCL generic functions can have lambda lists
    	... and that's all of the ones at startup, so let's write an
    		internalsish test to ensure that it stays that way

    csrhodes committed Oct 19, 2003
    	Make MODULE-PROVIDE-CONTRIB obey the protocol
    	... hacky solution as per CSR sbcl-devel 2003-10-18

    antifuchs committed Oct 19, 2003
    	Add a check for working CONSTANTLY in OpenMCL (some CVS versions
    	throw an error)
Commits on Oct 18, 2003

    telent committed Oct 18, 2003
    	More for SB-INTROSPECT, shamelessly inspired by Helmut
    	Eller's SLIME code for CMUCL.
    	objects for FDEFN references and return functions.  Users of
    	the former interface are probably advised to cache the
    	answers, or to accept that it's presently rather slow.

    csrhodes committed Oct 18, 2003
    	Minor PCL fixette
    	... when defining early generic functions in the bootstrap
    		process, clobber the lambda list if the gf has already
    		been created (e.g. NO-APPLICABLE-METHOD) by an early

    csrhodes committed Oct 18, 2003
    	Be more careful over automatically generated generic function
    	lambda lists
    	... when generating PCL-internal GFs, pass :LAMBDA-LIST to
    	... when generating accessor GFs, pass :LAMBDA-LIST if the
    		function is not already created (where you want to
    		preserve the user's lambda list instead)
    	... tests for required behaviour
    	Adjust INVALID-FASL patch slightly
    	... comment in package-data-list.lisp-expr
    	... remove unneccessary sb!ext:: prefixes
Commits on Oct 17, 2003

    antifuchs committed Oct 17, 2003
    	LOOP fixups - whee, I love digging around in code from 1986
    	* make SB-LOOP::LOOP-SEQUENCER no longer choke on NIL
    	  as a name for for-as-arithmetic counters
    	* also make it throw a PROGRAM-ERROR when it encounters
    	  a list as a counter variable.

    antifuchs committed Oct 17, 2003
    	ASDF users, rejoice!
    	* Add a condition SB-EXT:INVALID-FASL (subtype ERROR)
    	* Make SB-FASL::CHECK-FASL-HEADER raise conditions that are subtype
    	  SB-EXT:INVALID-FASL on the appropriate errors (or the other way
    	  around, I forget)
    	* Make SB-FASL::INTERNAL-LOAD also raise a condition if the FASL
    	  header is broken.

    csrhodes committed Oct 17, 2003
    	OK, here's the deal.
    	... *MODULE-PROVIDER-FUNCTIONS* no longer call PROVIDE.  It's
    		now the job of user code to do that (as it probably
    		should always have been.
    	... OK, so as a result, our one-file contribs now need to
    		PROVIDE themselves.  Make it so.
    	... fix asdf-install/asdf autorequiration; an installed sbcl
    		can now require asdf-install (as well as all the other
    		asdfized contribs) as the first command in a clean core
    	... update contrib/{README,STANDARDS} to match what we currently
Commits on Oct 16, 2003

    telent committed Oct 16, 2003
    	Innie?  Outie?
    	Take the first debug-fun from the fun-map vector in debug-info
    	instead of looking for one with a matching name, which may not
    	exist.  We only want it for the top-level form number anyway
    	compiler/debug-dump.lisp; write the tlf number at all debug
    	quality levels.  0 would take up just as much space as n does

    telent committed Oct 16, 2003
    	Update text files in contrib/ directory

    csrhodes committed Oct 16, 2003
    	Since it's rapidly becoming an FAQ
    	... additional magic in to make synthetic fasls
    		allowing REQUIRE on an asdf module to work without
    		a loaded ASDF.
    	... does not work for the ASDF-INSTALL contrib, because REQUIRE
    		goes for the file named "asdf-install" rather than
    		"asdf-install.fasl" in preference.  Probably want to fix
    		REQUIRE for that, but that's pending the other REQUIRE
Commits on Oct 15, 2003

    csrhodes committed Oct 15, 2003
    	Fix for bug 191c (and some of PFD's tests)
    	... do proper keyword argument checking in the effective method
    	... not the cleanest fix in the world (note especially
    		the use of PROGN as an optimization inhibitor)
    	... I'm not telling you how long it took me to find the
    		NCONC -> APPEND bug in fngen.lisp
Commits on Oct 14, 2003

    Alexey Dejneka committed Oct 14, 2003
            * Fix problem reported by salex on #lisp: SLOT-VALUE was not
              known to return exactly one value.
Commits on Oct 13, 2003

    csrhodes committed Oct 13, 2003
    	A couple of filesystem-related fixes from Milan Zamazal
    	... :IF-EXISTS OPEN behaviour corrected
    	... don't error if a file is deleted from under us in DIRECTORY
Commits on Oct 11, 2003

    telent committed Oct 11, 2003
    	Is that lint?
    	DEFINITION-SOURCE now has both FORM-PATH (a la CMUCL source
    	path, renamed because "source-path" is just too similar to
    	"source-pathname") and CHARACTER-OFFSET accessors.
    	DEFINITION-SOURCE now works to some extent on struct accessors
    	and predicates.  (It gets the pathname right, but I can't find
    	anywhere to get a within-file offset)
    	Commentary and stuff.

    telent committed Oct 11, 2003
    	Is that lint?
    	DEFINITION-SOURCE now has both FORM-PATH (a la CMUCL source
    	path, renamed because "source-path" is just too similar to
    	"source-pathname") and CHARACTER-OFFSET accessors.
    	DEFINITION-SOURCE now works to some extent on struct accessors
    	and predicates.  (It gets the pathname right, but I can't find
    	anywhere to get a within-file offset)
    	Commentary and stuff.
Commits on Oct 10, 2003

    William Harold Newman committed Oct 10, 2003
    	Clean more cleanly:
    	...Delete more doc/html/ stuff.
    	...also contrib/sb-bsd-sockets/

    csrhodes committed Oct 10, 2003
    	Fix DOCUMENTATION bug reported by dan_b on #lisp
    	... set documentation slot of braid classes to NIL

    Alexey Dejneka committed Oct 10, 2003
            * Remove CLOSURE-FUN-HEADER-WIDETAG; preserve its number for
              some time in order to keep FASL compatibility.

    telent committed Oct 10, 2003
    	More navel-gazing ...
    	FIND-DEFINITION-SOURCE takes a thing and returns the source
    	pathname and location at which it was defined.  Currently
    	works for only a small number of things (simple functions
    	and methods of gfs) and returns the location as the number
    	of source forms from the start of the file.  Suspect that
    	a character offset would be a better bet.
    	Poor pretence for a test case added as well.  So far no
    	framework to automate same.
Commits on Oct 9, 2003

    Alexey Dejneka committed Oct 9, 2003
            * Change modularization of ASH
              ... add per-function cutter;
              ... s/ash-left-constant-modxx/ash-left-modxx/;
              ... put DEFKNOWN and modular function optimizer for
                  ASH-LEFT-MODxx to src/compiler/generic/sm-tran.lisp;
              ... compile src/compiler/generic/vm-tran.lisp before
                  src/compiler/target/arith.lisp (in fact, immediately
                  after src/compiler/srctran.lisp);
            * strength reducer for * wraps LOGAND around the whole form.

    csrhodes committed Oct 9, 2003
    	Fix suboptimality reported by piso on #lisp
    	... catch bogus types like (fixnum 10) before they get too far
    		in the type system
    	... doesn't work for logically-built-in-but-made-by-PCL types
    	Comment out newly-broken assertion that constant multiplies
    		don't produce compiler-notes
    	... if that's all that broke, count your lucky stars :-)

    telent committed Oct 9, 2003
    	 Added sb-introspect contrib to SBCL.  Mostly for
    	 experimentation at this stage: do not use this stuff in
    	 client code unless you are willing to revisit it later, and
    	 please do say if it does things differently than you wanted.

    csrhodes committed Oct 9, 2003
            I'm not proud of this.
            HEALTH WARNING: this may not work.  It does for me, on Linux/PPC.
    		If your sourceforge-fu is strong, please try it.
            HEALTH WARNING: this is ugly as sin.  Unexported symbols, special
                    assumptions, KLUDGEs thrown in with gay abandon.
            In partial mitigation, it does fix a bug :-)
            Fix for lying-to-the-compiler bug in
            ... turn TRULY-THEs into suitable LOGANDs
                    (inefficient in compile-time space; we only need one
                    LOGAND wrapping the resulting form)
            ... likewise in x86 OPTIMIZE-MULTIPLY
                    (even less efficient: constant mask is first :-)
            but that would be slow at runtime if we just left it there, so
            ... add - as a modular function (that was easy)
            ... add preliminary support for ASH as a modular function
                    (for constant right shifts):
            ... delete ASH-RIGHT-[UN]SIGNED from the sparc backend
                    (will be restored eventually, fear not, probably more
            ... hack in special knowledge about ASH into CUT-TO-WIDTH
            ... ensure that all backends have a suitable VOP for translation
                    of new ASH function
            ... (alpha version is 64bit, oh yes)
            ... don't forget out-of-line version (for xc also!)
                    (aside: might we not need out-of-line versions of
                    other modular functions in the xc?)

    Alexey Dejneka committed Oct 9, 2003
            * Fix bug found by WHN and Paul Dietz: do not replace optional
              dispatch with an entry point in a block to be deleted.
Commits on Oct 7, 2003

    telent committed Oct 7, 2003
    	"brown paper bag" bug fix: initialise the thread->state slot
    	when it's created, to fix the most obvious threading problem
    	Search order for sbcl.core and sbclrc simplified.
    	Manual page edits: updates for contrib/, threading, platforms,
    	new sbcl.core search order, etc
    	Make "debugger invoked" message print thread id
    	asdf-install contrib: new UNINSTALL function deletes a
    	package's files.
    	New installation variable $(BUILD_ROOT) for use by package
    	installers that want to install into a fake root directory

    csrhodes committed Oct 7, 2003
    	Fix bug in CONS types/SUBTYPEP
    	... actually a bug in the CONS :SIMPLE-INTERSECTION2 type
    		method, which wasn't doing enough work when one of the
    		types produced a useful intersection and the other
Commits on Oct 6, 2003

    telent committed Oct 6, 2003
    	More thread fixes: now passes the cl-ppcre thread test
    	... all_threads_lock needed to be declared volatile
    	... new 'state' field in struct thread, to avoid race where
    	the head of the all_threads list is reaped during stop_for_gc
    	Removed "waitpid : child %d %x exited \n" message (rather,
    	conditionalised it on show_thread_exit, which is default 0
    	Remove unexplained (setsigmask 0) call in the repl, replaced
    	with a call to warn_when_signals_masked so we can find out
    	what it was for anyway.
    	Default repl calls (get-foreground) - this is a unithread nop