* Export SB-KERNEL::READER-EOF-EROR (bug reported by Nikodemus Siivola).
Comment out the meat in ppc/linux os_restore_fp_control, because although that superficially worsens our floating point handling on that platform, ...RUN-PROGRAM no longer segfaults with such enthusiasm; ...our floating point handling was quite poor in any case, as evidenced by (/ 1.0 0.0).
* Update BUGS.
Alpha fixes before 0.8.2 ... correct type declaration for FOREIGN-SYMBOL-ADDRESS-AS-INTEGER (add N-MACHINE-WORD-BITS to all backends for this purpose) ... correct argument order in the depths of DATA-VECTOR-SET-C/ (DEF-SMALL-DATA-VECTOR-FROBS), so Alpha now passes bit array tests run-tests.sh now prints the filename of the failing test just before stopping, which should save on the effort required to misinterpret the backtrace delete final full stop from " cd tests && sh ./run-tests.sh." in make.sh so I can triple-click the line to do exactly that commit patch for BASE-CHAR-COMPARE/C that Christophe sent me (alpha/char.lisp). I hope this is what he wanted ...
* Fix bug 262: LOAD-DEFMETHOD-INTERNAL checks whether GF has any methods before calling FIND-METHOD; ... late condition slot installers do not overwrite GF lambda list.
Frobs to previous patches ... Fix off-by-one error in column number reporting absent any #\Newlines in READER-ERROR ... change the test for BUFFER-P in fd-stream (and add entry in BUGS)
Significantly rearrange locking code for thread waitqueues, to eradicate races shown up in McCLIM applications. It's better than it was, and I can't trigger the lockup any longer: I don't claim it's _correct_, just that I can no longer see anything wrong with it Tidy up some constants in sb-bsd-sockets Amend spec in SB-POSIX to deal better with the differences between a CL namestring and a string-representing-a-filename- as-known-to-the-OS
A +FASL-FILE-VERSION+ a month reduces bug hunth.
Merge patch from David Lichteblau (sbcl-devel 2003-06-01) for (signed-byte 8) streams. ... no real insight, so merge it and hope for the best
Merge patch from Nikodemus Siivola for nicer error reporting on READER-ERRORs ... frob it a bit to reduce the number of erring possibilities
Merge patch from Nikodemus Siivola for FILE-POSITION on STRING-INPUT-STREAMs ... :START and :END are valid file position designators, so accept them.
Get SB-MOP:CLASS-PROTOTYPE right in more cases ... new :PROTOTYPE-FORM property in SB-KERNEL:*BUILT-IN-CLASSES* list; ... use it to generate the prototype in SB-PCL::*BUILT-IN-CLASSES*; bonus: since this means we have an instance of almost every built-in-class by SAVE-LISP-AND-DIE time, we can detect longstanding bugs in PURIFY ... make complex bit-vectors and simple-array-nils purifyable.
Update to upstream asdf ... now with better restarts and less nonsensical warning texts
Remove time calls in make.sh :-(
* Fix optimizer of BIT-NOT; * remove explicit type check in PEEK-CHAR.
WARNING: here be low-quality code ... implement more disassembly notes for PPC. ... full of horrible literal constants. Needs refactoring.
Make it slightly easier on the type inferencer: don't even threaten to return NIL from a string array reference. ... use ETYPECASE instead of TYPECASE.
* Cleanup of MACROLET processing; ... fix bug 264: interpreted version of SYMBOL-MACROLET did not check for a bound SPECIAL declaration.
* Remove several DECLAIMs for internal functions; * DEFSTRUCT: PROCLAIM return type of structure predicate.
Remove leakage from host to target of ARRAY-FOO-LIMIT via the INDEX type ... don't UNCROSS everything in DEF!TYPE, just the name. ... test for one of the symptoms (thanks to Patrik Nordebo for providing a manageable test case)
Merge vector_nil_string_branch ... many other incremental fixes, including * decrease of number of places array properties need to be specified; * rework of build order so that unknown types are never specialized; * primitive types need to know the specifier, not the ctype, so make it so; * fixes to the kernel classoid hierarchy, so more likely to be consistent internally. The good news is that, should it prove necessary, reverting this patch so that (vector nil) isn't a string is probably not very much work; all that needs to be changed are the kernel classoid supertypes and the STRING and SIMPLE-STRING definitions (and unparses). On the other hand, I'd be interested in trying to fix any performance problem "the right way" before reverting this behaviour.
* Fixed bug 263: coerce logical block affixes to SIMPLE-STRING; * COMPLEX-= type method: don't reparse totally unknown type specifier; * !DEF-BOOLEAN-ATTRIBUTES: create an attributes to list decoder; * FUN-INFO: print attributes; * optimizer for ARRAY-HEADER-P: reimplement as DERIVE-TYPE.
Since I previously made a note that the applicability of the REPLACE transform was critical for the performance of string-streams ... move types around sufficiently that the REPLACE transform becomes applicable again in string-stream machinery. (not that I observe the dramatic speedup I was hoping for)
Minor frobs, mostly to put behaviour back to as I found it in unimportant bits. I think there's still a small performance problem, which I'm investigating.
* Condition slot accessor installer: call ENSURE-GENERIC-FUNCTION; * fixed type method (VALUES :SIMPLE-=); * SB-C::DOMAIN-SUBTYPEP: merged patch by DTC 1999/01/23.
Delete unused %TEST-LOWTAG-AND-HEADERS type-vop function Minor textual edit to the text at the end of make.sh
fiddled with return code of condition.impure.lisp to keep run-tests.sh script happy trivial text/whitespace tweaking
* DELETE-BLOCK: if the last continuation has a destination, delete it. (bug reported by ohler on #lisp 2003-07-10) * make condition slot accessors be methods.
* s/*THE-PCL-PACKAGE*/*PCL-PACKAGE*/ (reported by Istvan Marko).
* Derive types of variables, bound with MV-BIND, even if we don't know the number of values, returned by the argument.
* tests of sb-bsd-sockets: SOCKET-ERROR is not exported.
* SB-BSD-SOCKETS: test of local sockets tries both stream and datagram kinds; * COMPLEX-= type method: if the second type is unknown, try to reparse it.