Juho Snellman [Thu, 9 Mar 2006 12:58:44 +0000 (12:58 +0000)]
0.9.10.25:
Port dynamic-extent vector support from x86 to x86-64.
Christophe Rhodes [Wed, 8 Mar 2006 18:49:54 +0000 (18:49 +0000)]
0.9.10.24:
Fix (I think) the %INSTANCE-REF build problems on PPC and
probably SPARC.
... the CTYPE-OF failure came about because the system didn't
know that (AND INSTANCE FUNCTION) was NIL. Make it
so...
... and then delete a stale KLUDGE workaround in primtype.lisp
... this then built as far as stream.lisp, where the problem
was that (AND INSTANCE STREAM) was being "simplified"
to just STREAM, and then primtype didn't know that
STREAMs are subtypes of INSTANCE...
... which in fact they're not; CLOS allows us to construct
funcallable streams. So...
... instead teach the system that (AND INSTANCE STREAM)
shouldn't be simplified...
... but (AND INSTANCE FD-STREAM) should.
... (also delete some crufty classoids: BASIC-STRUCTURE-CLASS
and FUNCALLABLE-STRUCTURE-CLASS were never used)
... tests for all the failing things in type.{before,after}-xc
Nathan Froyd [Tue, 7 Mar 2006 22:11:51 +0000 (22:11 +0000)]
0.9.10.23:
Unify the FOO and FOO-C VOPs in DEFINE-FULL-{REFFER,SETTER} on x86.
Saves ~100K in core file size.
Nathan Froyd [Tue, 7 Mar 2006 19:06:41 +0000 (19:06 +0000)]
0.9.10.22:
Remove spurious VOPs for translating %INSTANCE-{REF,SET}.
... These VOPs were based on SLOT-{REF,SET} and in all ports
there were also translations based on DEFINE-FULL-REFFER or
WORD-INDEX-REF. However, at the very least, the
%INSTANCE-SET VOP based on SLOT-SET was bogus, as it did not
return a value, thereby generating incorrect code.
Therefore, this VOP was either disabled (#+nil) or the
alternative via DEFINE-FULL-REFFER was carefully defined
with a lower :GENERATOR cost so that the VOP based on
SLOT-SET would never be called. This patch eliminates such
magic and eliminates the corresponding %INSTANCE-REF VOP as
well, since it does not appear to have any advantages over
the one created via DEFINE-FULL-REFFER.
... See also sbcl-devel, 2002-02-11, "Dubious VOPs". This patch
does not implement the DEFINE-VOP magic discussed in the
email.
Nathan Froyd [Tue, 7 Mar 2006 18:47:37 +0000 (18:47 +0000)]
0.9.10.21:
Add CMOV micro-optimization in GENERATE-CALL-SEQUENCE, too.
Christophe Rhodes [Tue, 7 Mar 2006 12:04:24 +0000 (12:04 +0000)]
0.9.10.20:
Move the stub definitions for contrib-type stuff into the
main runtime
... and don't build unix-dsos in contribs any more;
... should make binary distribution easier.
Christophe Rhodes [Tue, 7 Mar 2006 07:34:15 +0000 (07:34 +0000)]
0.9.10.19:
Add Alastair Bridgewater's chapter about calling conventions
Nathan Froyd [Mon, 6 Mar 2006 21:20:28 +0000 (21:20 +0000)]
0.9.10.18:
Micro-optimize DEFAULT-UNKNOWN-VALUES on x86{,-64} by using CMOV
... guard it with *BACKEND-SUBFEATURES* on the x86.
Nikodemus Siivola [Mon, 6 Mar 2006 17:48:10 +0000 (17:48 +0000)]
0.9.10.17: fix bug #400, aka more correct CONSTANTP
* To know that a function call is a constant we need to check that
it doesn't signal an error in addition to knowing that the function
is foldable.
* To know that a THE form is constant we need to check that the
value is of the right type and be prepared to deal with malformed
type specifiers.
* Tests, tests, tests.
...all in all, I'm starting to miss the evaluator branch.
Cyrus Harmon [Mon, 6 Mar 2006 09:59:33 +0000 (09:59 +0000)]
0.9.10.16:
x86/Darwin support
... Add Config support and ape the darwin/ppc space fixup stuff.
... Align stack to 16-bytes as per ABI in %alien-funcall,
c-call and show.
... Add darwin space params.
... Use an altstack for handling signals in hopes of giving signal
handlers properly aligned stacks.
... Get registers and flags out of the mcontext properly. Needed
to play some games to get sys/_types included in order to get
the full definition of the mcontext struct which Apple tries
to hide from us.
... Moved os_get_runtime_executable_path from ppc-darwin-os.c to
new file darwin-os.c.
... Fix debugging #ifdefs in runtime.h
... #define hacks to x86-assem.S to keep Apple's brain-dead
assembler happy. In particular, it doesn't like things like
$NIL, so I had to play some games with the precompiler to make
it happy, and it wants .globl instead of .global. Also fixed a
couple raw .aligns that should have used the macros
... Align stack in ppc-assem.S:call_into_c
... Added x86-darwin-os.h
... #- out debug.impure test that causes bad things to happen on
darwin/ppc as it seems to cause bad things on x86/darwin as
well.
... adjust fails for float tests
... LDSO stubs for x86/darwin
Alexey Dejneka [Mon, 6 Mar 2006 04:35:50 +0000 (04:35 +0000)]
Oops.
Alexey Dejneka [Mon, 6 Mar 2006 04:33:26 +0000 (04:33 +0000)]
0.9.10.15:
Fix typo in (defconstant m-v-prog1); partially fix the bug
in constant-folding reported by Paul F. Dietz in "New compiler
bugs", sbcl-devel 2006-03-05.
Juho Snellman [Sat, 4 Mar 2006 21:45:33 +0000 (21:45 +0000)]
0.9.10.14:
Kill some dead code.
Juho Snellman [Sat, 4 Mar 2006 21:12:06 +0000 (21:12 +0000)]
0.9.10.13:
Remove the x87 instruction definitions from the x86-64 backend for
a significant core file size reduction. Patch by Lutz Euler.
Juho Snellman [Sat, 4 Mar 2006 21:07:48 +0000 (21:07 +0000)]
0.9.10.12:
The error branch in the win32 DLOPEN-OR-LOSE was calling
SHARED-OBJECT-SAP instead of SHARED-OBJECT-HANDLE. (Patch from
by Yaroslav Kavenchuk, sbcl-devel "Re: Where define shared-object-sap"
on 2006-02-26).
Christophe Rhodes [Sat, 4 Mar 2006 19:58:26 +0000 (19:58 +0000)]
0.9.10.11:
Merge patch (Alastair Bridgewater 2006-02-06 / CSR 2006-03-03)
implementing a better calling convention on the x86 (and
probably a better one on the x86-64).
... increment fasl version, because mixing calling conventions
is a spectacular no-no.
... take this opportunity to put :gencgc on
*features-potentially-affecting...*, since mixing
gencgc and cheney allocation strategies on the ppc
is also a spectacular no-no.
Cyrus Harmon [Thu, 2 Mar 2006 05:21:16 +0000 (05:21 +0000)]
In aliencomp.c #+(and ppc darwin) should be #!+(and ppc darwin), which
allows ppc builds to be able to cross-compile to other
architectures. This bug was originially introduced by yours truly in
the process of bringing the VOPs in question and this call to the over
from CMUCL.
Christophe Rhodes [Wed, 1 Mar 2006 09:10:32 +0000 (09:10 +0000)]
0.9.10.9:
Another try at the COPY-MORE-ARGS microoptimization, which was
in fact reverted inadvertantly by the ppc/gencgc merge.
... this time, schedule the instructions by hand to try to
avoid a stall.
... (whitespace. *sigh*.)
Christophe Rhodes [Tue, 28 Feb 2006 14:22:54 +0000 (14:22 +0000)]
0.9.10.8:
Maybe fix some gc problems on ppc/gencgc?
... there's a workaround for non-descriptors in descriptor
registers which was previously !GENCGC but in fact should
have been !CONSERVATIVE or possibly !PARTITIONED_REGISTERS.
... whitespace in target-char.lisp
Nikodemus Siivola [Tue, 28 Feb 2006 05:01:20 +0000 (05:01 +0000)]
0.9.10.7: oops. re-version last
Nikodemus Siivola [Tue, 28 Feb 2006 04:58:55 +0000 (04:58 +0000)]
0.9.10.6: fix non-unicode build
...I was _sure_ I'd tested this...
Juho Snellman [Tue, 28 Feb 2006 04:47:50 +0000 (04:47 +0000)]
0.9.10.6:
Restore the old behaviour of never using mmap-tricks for memory
zeroing on Solaris/gencgc, since doing so was causing GC crashes
on Solaris/x86.
Also some whitespace canonicalization.
Nikodemus Siivola [Mon, 27 Feb 2006 15:03:08 +0000 (15:03 +0000)]
0.9.10.5: correct CONSTANTP
* fix thinko in interactions between PROGV & IF and add a test to catch
it.
Nikodemus Siivola [Mon, 27 Feb 2006 13:12:34 +0000 (13:12 +0000)]
0.9.10.4: better CONSTANTP
* Recognizes constant argument calls to foldable functions and also
deals with some simple special forms like.
* Replace a ton of EVAL calls with CONSTANT-FORM-VALUE.
Nikodemus Siivola [Mon, 27 Feb 2006 11:07:31 +0000 (11:07 +0000)]
0.9.10.3: Faster function calls via constant symbols
* Compile (FUNCALL 'FOO) as (FUNCALL (SB-C::GLOBAL-FUNCTION FOO))
which is like (FUNCALL (FUNCTION FOO)) except that inline
functions are not expanded and the lexical functions are ignored.
Christophe Rhodes [Mon, 27 Feb 2006 11:02:11 +0000 (11:02 +0000)]
0.9.10.2:
Fix remaining slot name publicness in standardized classes.
... be cowardly and just rename TYPE to %TYPE, but write a
comment about why this isn't really good enough.
... now we can test for our interpretation.
... document it, too
Nikodemus Siivola [Mon, 27 Feb 2006 05:13:29 +0000 (05:13 +0000)]
0.9.10.1: Unicode character names -- aka More Bloat
* Make CHAR-NAME and NAME-CHAR aware of the Unicode names when building
with :SB-UNICODE.
William Harold Newman [Mon, 27 Feb 2006 03:08:03 +0000 (03:08 +0000)]
0.9.10:
release, tagged as sbcl_0_9_10
Christophe Rhodes [Wed, 22 Feb 2006 12:11:26 +0000 (12:11 +0000)]
0.9.9.38:
gencgc/ppc fixups
... fix the allocator/gc on threaded builds. Whoops.
... STACK_GROWS_DOWNARD
... in pa_alloc/gencgc/!x86, actually do the stack manipulation
more plausibly. (Don't carefully write the value we
want to preserve past the end of the stack, for one)
... whitespace.
Juho Snellman [Tue, 21 Feb 2006 23:40:36 +0000 (23:40 +0000)]
0.9.9.37:
Implement os_get_runtime_exexutable_path() on Solaris
(patch by Daisuke Homma).
Move the os_get_runtime_exexutable_path() FreeBSD version
check from compile-time to runtime (patch by NIIMI Satoshi).
Christophe Rhodes [Tue, 21 Feb 2006 22:59:29 +0000 (22:59 +0000)]
0.9.9.36:
Merge Cyrus Harmon's 7th gencgc merge candidate
... with the addition of idempotent implementations of
arch_clear_pseudo_atomic_interrupted() for
sparc, mips, alpha and hppa. (the last three completely
untested).
... many, many changes, most of which are documented in
doc/internals-notes/GENCGC-PORTING-NOTES
(This commit may break horribly. Please read, please test)
Nathan Froyd [Fri, 17 Feb 2006 20:56:26 +0000 (20:56 +0000)]
0.9.9.35:
Eliminate calls to GENERIC-< in SUBTRACT-BIGNUM-BUFFERS{,-WITH-LEN}
by declaring a few types.
Christophe Rhodes [Fri, 17 Feb 2006 17:49:36 +0000 (17:49 +0000)]
0.9.9.34:
Merge fix from Peter Van Eynde (sbcl-devel 2006-02-11 "cosmetic
room bug") for a cosmetic room bug.
... also fix a distinctly non-cosmetic scrub-control-stack bug
resulting from the same issue. This scrubbing failure
appeared to cause heap corruption in powerpc/gencgc
test builds; I think I understand why.
The Cheney GC zeros the unused parts of the lisp control stack
after it has completed the garbage collection. This ensures
that, if the active stack had no garbage pointers at the start
of the collection, there is no region in the entire control
stack (used or unused) which contains a garbage pointer, since
every entry has either been scavenged or zeroed. But since by
assumption we start off with no garbage pointers, by
mathematical induction we never scavenge one, so everything is
safe.
GENCGC doesn't perform this zeroing. (Why?) However,
SCRUB-CONTROL-STACK does, before a GC. This is slightly more
dangerous, because we could in fact have incomplete stack frames
lying below the stack pointer with an entry from a previous
iteration of the heap, but I think it's OK by the same
reasoning as before. Failure to zero the stack, however, does
leave the possibility of bogus pointers open when stack frames
are extended but not every stack slot has yet been written to.
This wasn't so much of an issue when the stack is
scanned conservatively and ambiguous roots caused pinning, but
under a precise stack scanning regime disaster ensues.
Christophe Rhodes [Fri, 17 Feb 2006 17:32:37 +0000 (17:32 +0000)]
0.9.9.33:
Fix the gencgc verification code for raw-slot-at-end
structure layout.
Christophe Rhodes [Fri, 17 Feb 2006 17:15:34 +0000 (17:15 +0000)]
0.9.9.32:
Fix hideously embarrassing ppc assembly bug in reg_LRA
computation.
... no longer go wrong if bit 15 of lra is set.
(The symptoms from this have been reported many, many times:
segmentation faults in the first triggered GC. Kevin Rosenberg
reported it first from my trawl on sbcl-devel, but I think it's
been known for longer than that. Previously it had been
dismissed as gcc miscompilation problems, because the problem
disappeared when using a different version of gcc, for any
individual developer: in retrospect, the fact that it was our
bug after all is pretty obvious from the fact that we were never
able to characterize particular versions of gcc which were bad.)
Nathan Froyd [Tue, 14 Feb 2006 19:38:38 +0000 (19:38 +0000)]
0.9.9.31:
Add a declaration to GET-LATIN-BYTES to silence some efficiency
notes. Makes the code smaller, too.
Nathan Froyd [Tue, 14 Feb 2006 19:30:12 +0000 (19:30 +0000)]
0.9.9.30:
PPC micro-optimization of COPY-MORE-ARG VOP. Use LWZU/STWU to
avoid ADDI instructions.
Gabor Melis [Tue, 14 Feb 2006 17:14:59 +0000 (17:14 +0000)]
0.9.9.29
* fixed type= and csubtypep for arrays of unknown-type
Christophe Rhodes [Tue, 14 Feb 2006 09:27:20 +0000 (09:27 +0000)]
0.9.9.28:
Whoops. SB-INTROSPECT is (was) using one of the symbols I
renamed in 0.9.9.27...
Christophe Rhodes [Mon, 13 Feb 2006 15:59:16 +0000 (15:59 +0000)]
0.9.9.27:
Fix most use of slot-names colliding with external symbols /
symbols accessible from CL-USER
... prefix most such slots by %;
... rename METHOD-COMBINATION-TYPE to -TYPE-NAME (as in AMOP
FIND-METHOD-COMBINATION)
... only the TYPE slot in SPECIALIZER left to go, which is more
complicated because in fact it's not a TYPE at all; more
like a specifier (or maybe a typeoid)
Christophe Rhodes [Mon, 13 Feb 2006 10:56:29 +0000 (10:56 +0000)]
0.9.9.26:
Move the Linux/ppc heap around again, in an attempt to find a space
that everyone can use...
... lichtblau on #lisp reported that his ppc mapped stuff at
#x30000000
... also some whitespace unbreakification.
Christophe Rhodes [Fri, 10 Feb 2006 15:02:10 +0000 (15:02 +0000)]
0.9.9.25:
Fix bug in method-metacircle/discriminating function update.
... start defining SAFE-FOO variants of method- and
generic-function- accessors, concentrating the
horribleness. At the moment, we have separate SAFE-FOO
and EARLY-FOO logic; at some time in the future it might
be worth coalescing the two.
... test cases. Include both Jean and Pascal's variants of the
method code, and write similar generic-function code
(which, admittedly, seemed to pass anyway).
Christophe Rhodes [Thu, 9 Feb 2006 10:09:18 +0000 (10:09 +0000)]
0.9.9.24:
Fix build on ppc (sb- -> sb!)
Teemu Kalvas [Wed, 8 Feb 2006 19:41:10 +0000 (19:41 +0000)]
Fixed buffer overrun in resync restart for variable width external formats.
Nathan Froyd [Wed, 8 Feb 2006 19:19:28 +0000 (19:19 +0000)]
0.9.9.23:
More fixes courtesy of PFD's ansi-tests:
...%LOGBITP had its arglist reversed. Oops;
...be more proactive in declaring the types of constant arguments
in the %LOGBITP VOPs to avoid bogosity.
Christophe Rhodes [Tue, 7 Feb 2006 17:55:22 +0000 (17:55 +0000)]
0.9.9.22:
PPC64 build fixes
... uname returning ppc64 should give us an SBCL_ARCH of ppc;
... move the spaces around for MORE HEAP (512Mb vs 128Mb);
... turn on linkage tables on Linux/PPC.
Christophe Rhodes [Tue, 7 Feb 2006 17:44:43 +0000 (17:44 +0000)]
0.9.9.21:
Make disassembly on PowerPC (at least) less hideously ugly, by
removing weird WORD #x00000000 overflowingness
Nathan Froyd [Tue, 7 Feb 2006 16:23:41 +0000 (16:23 +0000)]
0.9.9.20:
Oops. Add functions for constant folding purposes and fix AVER
in the x86 backend. Add tests from (where else?) ansi-tests.
Juho Snellman [Tue, 7 Feb 2006 03:32:01 +0000 (03:32 +0000)]
0.9.9.19:
Fix build.
Nathan Froyd [Tue, 7 Feb 2006 02:35:25 +0000 (02:35 +0000)]
0.9.9.18:
Introduce new vm-support-routine COMBINATION-IMPLEMENTATION-STYLE
for letting the backend have a crack at implementing certain
functions directly (cf. OPTIMIZATIONS, #29);
...implement a few efficient cases for PPC and x86.
Nathan Froyd [Tue, 7 Feb 2006 02:21:43 +0000 (02:21 +0000)]
0.9.9.17:
Unbreak the test suite by exporting SBCL_HOME from run-tests.sh.
SB-POSIX is now found as one might expect.
Juho Snellman [Sun, 5 Feb 2006 23:29:15 +0000 (23:29 +0000)]
0.9.9.16:
Fix tests that modify constant data (patch by Lutz Euler).
Juho Snellman [Sun, 5 Feb 2006 23:14:08 +0000 (23:14 +0000)]
0.9.9.15:
Make some variables in the code code generated by
ALIEN-CALLBACK-LISP-WRAPPER-LAMBDA ignorable to get rid of
spurious warnings, as suggested by James Bielman on sbcl-devel
"Unused variable warnings in SB-ALIEN::ALIEN-LAMBDA".
Juho Snellman [Sun, 5 Feb 2006 22:26:11 +0000 (22:26 +0000)]
0.9.9.14:
Canonicalize some whitespace (the upstream version of
asdf.lisp committed last week).
Juho Snellman [Sun, 5 Feb 2006 22:22:01 +0000 (22:22 +0000)]
0.9.9.13:
Support for finding the executable path on FreeBSD. (thanks to
NIIMI Satoshi)
Juho Snellman [Sun, 5 Feb 2006 22:03:00 +0000 (22:03 +0000)]
0.9.9.12:
New SAVE-LISP-AND-DIE keyword argument :EXECUTABLE can be used
for bundling the runtime and the core file into one executable
binary. This feature is not currently supported on all SBCL
platforms. (thanks to James Bielman)
Christophe Rhodes [Wed, 1 Feb 2006 10:27:54 +0000 (10:27 +0000)]
0.9.9.11:
Change the way we find our asdf-based contribs, using a custom
*system-definition-search-function* rather than symlinks.
... $SBCL_HOME/systems directory can go away;
... various changes in build and install scripts.
Christophe Rhodes [Sun, 29 Jan 2006 22:15:26 +0000 (22:15 +0000)]
0.9.9.10:
Fix a bug in ctor optimization reported by kpreid on #lisp
... quote the ctor-initarg keys in the fallback (unoptimized)
constructor too.
Juho Snellman [Sun, 29 Jan 2006 22:04:14 +0000 (22:04 +0000)]
0.9.9.9:
Don't signal an error on (DOCUMENTATION 'FOO 'STRUCTURE) when
there's no structure named 'FOO (reported by Glenn Ehrlich).
Alexey Dejneka [Sat, 28 Jan 2006 08:52:35 +0000 (08:52 +0000)]
0.9.9.8:
* Merge a patch by Ivan Boldyrev fixing character code
conversions for OCTETS-TO-STRING and STRING-TO-OCTETS.
Christophe Rhodes [Fri, 27 Jan 2006 22:42:55 +0000 (22:42 +0000)]
0.9.9.9:
Fix bug #399 (gwking on #lisp / paste 16110; reduced case by
NJF)
... we need to be able to derive DATA-VECTOR-REF's return type
when we have a SIMPLE-STRING, even if the array's type
isn't represented directly as an ARRAY-TYPE
Nathan Froyd [Fri, 27 Jan 2006 20:42:19 +0000 (20:42 +0000)]
0.9.9.6:
Place yet another obstacle in the way of version 1.0. (bug #399)
Nathan Froyd [Fri, 27 Jan 2006 15:53:19 +0000 (15:53 +0000)]
0.9.9.5:
Reduce storage requirements for error trap information:
... shuffle around ANY-REG and DESCRIPTOR-REG storage classes
to make their SC-NUMBERs small, which makes SC-OFFSETs smaller,
which cuts down the size of SC-OFFSETs in error trap
information from three bytes to one byte;
... saves ~250KB in core files on PPC/OS X;
... only applies to RISCy platforms, as x86oid SC-OFFSETs already
fit into a single byte due to having a smaller number of
registers.
Bump the fasl file version, too.
Andreas Fuchs [Thu, 26 Jan 2006 23:29:07 +0000 (23:29 +0000)]
0.9.9.4:
Fix memory fault in sb-bsd-sockets:get-protocol-by-name.
Patch by Stelian Ionescu, sbcl-devel 2006-01-22.
* Also export unsupported-protocol condition
* Add 3 test cases.
Juho Snellman [Thu, 26 Jan 2006 23:16:13 +0000 (23:16 +0000)]
0.9.9.3:
Make the handling of errors outside WITH-TEST forms more
robust.
Juho Snellman [Thu, 26 Jan 2006 23:06:04 +0000 (23:06 +0000)]
0.9.9.2:
Test cleanups.
* Mark some tests as expected to fail on various platforms, based
on information from test reports on sbcl-devel
* Disable the failing external-format test completely, since it
was sometimes failing in unexpected ways due to WITH-TIMEOUT
races
* Try to avoid running TEST-BIG-BIT-VECTORS on platforms where
the big vector doesn't fit into the dynamic space
Juho Snellman [Thu, 26 Jan 2006 21:51:31 +0000 (21:51 +0000)]
0.9.9.1:
Use an opt-in strategy for SSE. By default use the base version
of fast_bzero unconditionally. The cpuid-based detection is only
enabled when we have reason to believe that the operating system
supports SSE. Patch by NIIMI Satoshi.
William Harold Newman [Thu, 26 Jan 2006 19:29:11 +0000 (19:29 +0000)]
0.9.9:
release, tagged as sbcl_0_9_9
Juho Snellman [Sun, 22 Jan 2006 18:55:05 +0000 (18:55 +0000)]
0.9.8.48:
Disable fast_bzero_sse for FreeBSD systems with kernels that
don't have SSE support. Patch by NIIMI Satoshi on sbcl-devel,
"Re: upcoming sbcl-0.9.9 release".
Juho Snellman [Fri, 20 Jan 2006 08:34:56 +0000 (08:34 +0000)]
0.9.8.47:
Clear the direction flag on Lisp -> C transitions, as
required by the x86-64 ABI. Fixes mysterious GC crashes on
SuSE. (reported by Andrej Grozin and Hendrik Maryns)
Rudi Schlatte [Thu, 19 Jan 2006 17:02:33 +0000 (17:02 +0000)]
0.9.8.46
Merge patch "Charsets: latin-N, N=2,...,8" from Ivan Boldyrev,
sbcl-devel 2006-01-17
Rudi Schlatte [Thu, 19 Jan 2006 15:15:07 +0000 (15:15 +0000)]
0.9.8.45
Briefly document bivalent streams.
Christophe Rhodes [Wed, 18 Jan 2006 12:57:46 +0000 (12:57 +0000)]
0.9.8.44:
Fix for (truename #p"/") (reported by tomppa on #lisp)
Christophe Rhodes [Wed, 18 Jan 2006 11:28:28 +0000 (11:28 +0000)]
0.9.8.43:
Commit patch from Luis Oliveira for alignment on #!+win32
Christophe Rhodes [Mon, 16 Jan 2006 15:39:57 +0000 (15:39 +0000)]
0.9.8.42:
Merge "first round of i/o fixes" (sbcl-devel 2006-01-13
from James Bielman)
... some extended horribleness, mostly isolated horribleness.
Christophe Rhodes [Mon, 16 Jan 2006 15:10:24 +0000 (15:10 +0000)]
0.9.8.41:
Fix the mop/sb-posix/interface.pure.lisp/PCL metacircularity
problem.
... treat GF-DFUN-STATE and (SETF GF-DFUN-STATE) specially.
... thanks to everyone who thought very hard about it!
... also make interface.pure.lisp actually pure.
Christophe Rhodes [Mon, 16 Jan 2006 14:45:46 +0000 (14:45 +0000)]
0.9.8.40:
Merge patch from Luis Oliveira "stdcall support for alien-funcall"
sbcl-devel 2006-01-12
... factor set-fpu-word-for-{c,lisp} out from number-stack-space
manipulating vops.
... magic to adjust for calling convention.
Andreas Fuchs [Sat, 14 Jan 2006 20:01:07 +0000 (20:01 +0000)]
0.9.8.39:
make sb-posix's asd file load sb-grovel with asdf
This change should make it possible to run tests even
if sb-grovel is not in $SBCL_HOME.
Juho Snellman [Fri, 13 Jan 2006 11:32:46 +0000 (11:32 +0000)]
0.9.8.38:
Log a BUG.
Andreas Fuchs [Thu, 12 Jan 2006 13:26:41 +0000 (13:26 +0000)]
0.9.7.37:
Make asdf-install use bivalent streams
* Fix warnings and style-warnings when compiling a-i/installer.lisp:
remove undefined variables / exported symbols *verify-gpg-signatures*
and *safe-url-prefixes*, add a few ignored declarations.
* Add :element-type :default to stream opening forms that need it
* Allow sb-executable's copy-stream to deal with bivalent streams:
Add an element-type &key argument that should be passed when copying
bivalent streams
Christophe Rhodes [Thu, 12 Jan 2006 09:31:21 +0000 (09:31 +0000)]
0.9.8.36:
->pure, not ->slots[15] in purify
... independent of whether 7 clos-hash slots get deleted, this
might prevent future maintainers making the same kind of
fool of themselves as I did with my "inconsequential
modifications" error...
Nathan Froyd [Wed, 11 Jan 2006 21:32:19 +0000 (21:32 +0000)]
0.9.8.35:
Add getpwnam and getpwuid to SB-POSIX...
* ...in a fit of frustration ("why does this work in sb-bsd-sockets
and the exact same thing not work here?"), remove the
SB-POSIX-INTERNAL package and throw everything into SB-POSIX.
No more lurking issues with packages.
Christophe Rhodes [Wed, 11 Jan 2006 14:28:35 +0000 (14:28 +0000)]
0.9.8.34:
Merge patch from nyef for load-shared-object on win32.
... there's a scary amount of duplication between
win32-foreign-load and foreign-load; hope some
gardening takes place at some point.
... attempt to make make-config.sh put the relevant target
features in place.
Christophe Rhodes [Wed, 11 Jan 2006 13:33:20 +0000 (13:33 +0000)]
0.9.8.33:
From James Bielman: dirname() for win32.
Christophe Rhodes [Wed, 11 Jan 2006 11:58:59 +0000 (11:58 +0000)]
0.9.8.32:
Sucks to be me.
... insert missing semicolon in win32-os.c. Why? Why why why?
Christophe Rhodes [Tue, 10 Jan 2006 17:39:28 +0000 (17:39 +0000)]
0.9.8.31:
Implement (following James Bielman) unix-access for windows.
... use access() not _access(); no, I have no idea why this
makes sense (in either direction).
Christophe Rhodes [Tue, 10 Jan 2006 16:17:04 +0000 (16:17 +0000)]
0.9.8.30:
Fix for --load on Win32 (patch from James Bielman)
Juho Snellman [Mon, 9 Jan 2006 23:13:19 +0000 (23:13 +0000)]
0.9.8.29:
Kill a couple of obsolete x86 instructions (illegal in 64-bit
mode). Patch from sbcl-devel "Tidying up
src/compiler/x86-64/insts.lisp" by Lutz Euler on 2006-01-08.
Juho Snellman [Mon, 9 Jan 2006 22:46:14 +0000 (22:46 +0000)]
0.9.8.28:
Oh, the embarrassment. x86-64 was using full calls to GENERIC-< and
GENERIC-> for (UN)SIGNED-BYTE-64-P and CHECK-(UN)SIGNED-BYTE-64.
Fix it.
* Conditionalize type predicate, type predicate wrapper and
typecheckfun creation on N-WORD-BITS. Add missing 64-bit
cases.
* Add missing SIGNED-BYTE-64 VOPs.
* Nuke the now-unused 32-bit VOPs.
Christophe Rhodes [Mon, 9 Jan 2006 13:00:17 +0000 (13:00 +0000)]
0.9.8.27:
Merge patch from James Bielman fixing self-build under Win32.
... a little bit of an accident ensued; I ran
canonicalize-whitespace on .sh files, which seemed to
work (except on wc.sh) until it scribbled all over
make.sh, confusing the shell interpreter utterly.
... the resulting files have built the system, nevertheless.
Christophe Rhodes [Mon, 9 Jan 2006 12:45:47 +0000 (12:45 +0000)]
0.9.8.26:
Merge patch from James Bielman fixing use of GNAME on
fast_bzero* assembly functions.
Christophe Rhodes [Sun, 8 Jan 2006 22:39:03 +0000 (22:39 +0000)]
0.9.8.25:
Merge patch from Luis Oliviera (sbcl-devel 2006-01-08) fixing
probe-file on Win32.
Juho Snellman [Sun, 8 Jan 2006 06:31:23 +0000 (06:31 +0000)]
0.9.8.24:
It's x86-64 sign-extension time, once again. The
SB!VM:SIGN-EXTEND VOP wasn't getting selected, causing a full
call for every use. Luckily it's only used only for alien
calls whose return value is declared as a signed byte of 32 or
fewer bits.
Juho Snellman [Sun, 8 Jan 2006 03:06:11 +0000 (03:06 +0000)]
0.9.8.23:
Add x86-64 support for passing alien callback parameters on the stack.
Juho Snellman [Sun, 8 Jan 2006 02:01:42 +0000 (02:01 +0000)]
0.9.8.22:
Some x86-64 SAP-REF setter VOPs for floats were trying to use
the MOVE macro on float registers, causing compile errors in
rare circumstances. Extend x86-64 MOVE to also handle
SINGLE-REG and DOUBLE-REG SCs.
Juho Snellman [Sun, 8 Jan 2006 00:59:34 +0000 (00:59 +0000)]
0.9.8.21:
Oops.
Juho Snellman [Sun, 8 Jan 2006 00:33:13 +0000 (00:33 +0000)]
0.9.8.20:
Final batch from sbcl-devel "Changes to GENCGC memory zeroing"
in 2005-12). Use hand-coded assembly for zeroing memory in
GENCGC instead of the platform memset/bzero.
* Use MOVNTDQ on x86-64
* Use MOVNTDQ on x86 that supports SSE2 (basically Pentium 4 and
newer)
* Difference to the version posted on sbcl-devel: Do *not* use
the MMX MOVNTQ for x86 that supports MMX but not
SSE2. MOVNTQ apparently had very bad performance on K7
Athlons/Durons.
* Use REP STOSL on remaining x86.
Juho Snellman [Sat, 7 Jan 2006 18:53:39 +0000 (18:53 +0000)]
0.9.8.19:
Changes to GENCGC memory zeroing behaviour that give a big
performance boost, especially on modern processors.
* Instead of zeroing memory by remapping memory with
munmap/mmap at GC time, pages are just marked as needing
zeroing and zeroed with memset when they're added to a new
allocation region. This reduces GC latency both for the
common and worst cases.
* To keep the memory footprint down, clear the pages by
remapping after major GCs (arbitrarily defined as a
collection of generation 2 or older). The memory freed from
a minor GC is just going to get used again immediately , so
releasing them back to the OS would make little sense.
* Add a GENCGC mode (#define READ_PROTECT_FREE_PAGES) for
catching attempts to read unallocated pages
* See sbcl-devel "Changes to GENCGC memory zeroing" in 2005-12
for more details and performance measurements. (Note
that many parts of this patch have already been committed
piecemeal over the last month, this is just the most
significant chunk).
* Performance effect on BSDs (which used a different zeroing
strategy than Linux before this) is unknown.
Juho Snellman [Sat, 7 Jan 2006 18:22:58 +0000 (18:22 +0000)]
0.9.8.18:
Rearrange the GENCGC "struct page" a bit to for a more compact
memory representation. Saves memory (about 15MB on x86-64
where the page table is large, a couple of MB on x86). Also
a minor performance improvement thanks to cache issues.
TODO: The size could be still improved by another 15MB on x86-64
by defining the ill-named first_object_offset as an int
instead of long (4 bytes less data and 4 bytes less of padding).
The naive implementation would then limit the maximum region size
to 4GB. Since some low bits in the field are guaranteed to be
zero, a smart implementation could do some shifts and store even
more data. It remains to be seen whether this would be worthwhile.
Christophe Rhodes [Fri, 6 Jan 2006 16:44:59 +0000 (16:44 +0000)]
0.9.8.17:
Fix a bug in ENSURE-DIRECTORIES-EXIST: merge in
*DEFAULT-PATHNAME-DEFAULTS*.
... this fix may also include a mostly-working set of pathname
functions for Win32. Or it may not. You have been
warned.