sbcl Log


Commit Date  
[a45df4] (flaky5_branch) by William Harold Newman William Harold Newman

0.pre7.37.flaky5.29:
incremented fasl file version number
I noticed (while making an abortive effort to simplify SBCL's
own code enough that the new simple-minded inlining of
DEFUN in non-null lexical environments could handle
it) that BUTLAST, NBUTLAST, PAIRLIS, REVAPPEND, LDIFF,
NRECONC, COPY-LIST, COPY-ALIST, COPY-TREE, and
LIST-LENGTH aren't likely to benefit much from
inlining anyway (since they involve loops and consing,
and since their argument lists are trivial), so I
removed their inline declarations.

2001-10-03 13:27:05 Tree
[6b031d] by William Harold Newman William Harold Newman

0.pre7.37.flaky5.28:
(This version builds, passes regression tests, and serves as
its own cross-compilation host. Maybe the long flaky5
misadventure is finally drawing to a close.)
moved "miscellaneous primitive stubs" out of byte-interp.lisp
into stubs.lisp
INLINE-SYNTACTIC-CLOSURE-LAMBDA was broken by the preceding
changes. Do I want to maintain something that hairy?
No, so I..
..rewrote DEFUN to avoid INLINE-SYNTACTIC-CLOSURE-LAMBDA
..deleted now-unused INLINE-SYNTACTIC-CLOSURE-LAMBDA
..defined LAMBDA-INDEPENDENT-OF-LEXENV-P to support this

2001-10-02 23:51:01 Tree
[134e08] by William Harold Newman William Harold Newman

0.pre7.37.flaky5.27:
(builds, passes regression tests, and serves as its own
cross-compilation host up to backend.lisp, where
it dies with 47 errors)
split LOOP-until-LOCALL-done logic out of %COMPILE and
COMPILE-TOP-LEVEL (for general tidiness and also to
make it easier to trace what's going on in the bug
that I'm *still* looking for..)
replaced PRE-ENVIRONMENT-ANALYZE-TOP-LEVEL with
PREALLOCATE-ENVIRONMENTS-FOR-TOPLEVELISH-LAMBDAS in
%COMPILE, to stop the cascade of events which causes
failure in (LOAD "src/cold/shebang.lisp")
defined LAMBDA-TOP-LEVELISH-P to support this
deleted my recent and apparently bogus experimental assertion
(MEMBER (LAMBDA-KIND LETLAMBDA) '(:LET :MV-LET))
from GET-LAMBDA-ENVIRONMENT, since the changes above
cause it to fail and since I can't really justify it
removed other dependent-on-LAMBDA-KIND=:LET assertions there
too, since now I see I can't safely assume :LETness
similarly, deleted (AVER (NULL (LAMBDA-RETURN LETLAMBDA)))

2001-10-01 02:16:50 Tree
[84a94a] by William Harold Newman William Harold Newman

0.pre7.37.flaky5.26:
(still hunting assertion failure in FIND-IN-ENVIRONMENT when
compiling shebang.lisp, still no luck)
one new assertion
mostly just some new comments and tweaking/reformatting some
old comments

2001-09-29 23:25:32 Tree
[d426c5] by William Harold Newman William Harold Newman

0.pre7.37.flaky5.25:
factored out COMPONENT-TOP-LEVELISH-P to support this

2001-09-28 15:00:09 Tree
[019b26] by William Harold Newman William Harold Newman

0.pre7.37.flaky5.24:
updated REPLACE-TOP-LEVEL-XEPS to reflect the new world
of HAS-EXTERNAL-REFERENCES-P

2001-09-27 17:21:46 Tree
[77bfae] by William Harold Newman William Harold Newman

0.pre7.37.flaky5.23:
(I've been trying to debug the assertion failure in
FIND-IN-ENVIRONMENT when compiling compiling
shebang.lisp. No luck so far..)
IR2-ENVIRONMENT-ENVIRONMENT is explicitly initialized and never
modified subsequently -- would that the same could be
said of more slots! -- so I declared it that way.
and similarly for IR2-ENVIRONMENT-RETURN-PC-PASS too
more-explicit variable names in ASSIGN-IR2-ENVIRONMENT; removed
archaic COLLECT too
added invariant assertions in GET-LAMBDA-ENVIRONMENT

2001-09-27 16:24:51 Tree
[121e9e] by William Harold Newman William Harold Newman

0.pre7.37.flaky5.22:
added #!-GENGC on main DEFUN IR2-CONVERT-CLOSURE to correspond
to the #!+GENGC variant
made FIND-REFERENCE-FUNCTIONS grok HAS-EXTERNAL-REFERENCES-P
(I was hoping the HAS-EXTERNAL-REFERENCES-P would make the
cross-compiler work on shebang.lisp, but it doesn't.
Oh well.)

2001-09-26 22:30:15 Tree
[d35695] by William Harold Newman William Harold Newman

0.pre7.37.flaky5.21:
(This version can be built under sbcl-0.6.13 and passes
regression tests, but can't bootstrap itself.)
reenabled the structure slot accessor type declarations in
defstruct.lisp
added (DECLARE (OPTIMIZE (SAFETY 0))) for !LATE-TYPE-COLD-INIT
to suppress inline structure slot accessor type checks
there
added some regression tests for bug #126.
noticed that bugs IR1-1, IR1-2, IR1-3, and IR1-3a are gone

2001-09-26 16:45:23 Tree
[ec480d] by William Harold Newman William Harold Newman

0.pre7.37.flaky5.20:
The disassembler could be really handy for debugging the
structure accessor type check problems, so..
..uncommented GEN-PRINTER-DEF-FORMS-DEF-FORM call
..hacked on GEN-PRINTER-DEF-FORMS-DEF-FORM stuff until it
used unique function names for all the
automatically-generated functions (so that new
static linking code can cope with it)

2001-09-25 22:21:35 Tree
[65ca48] by William Harold Newman William Harold Newman

0.pre7.37.flaky5.19:
added a regression test to catch current structure accessor
carelessness about input types

2001-09-25 02:25:50 Tree
[27ff18] by William Harold Newman William Harold Newman

0.pre7.37.flaky5.18:
(Oops! The last couple of versions didn't work as well as
I thought: I was just seduced by slam.sh. It looks
as though flaky5.14 was the last one which would
build. This version seems to've fixed the problem,
so that make.sh actually works again, and even
passes regression tests again, but its structure
accessor type checking is still slightly messed
up in a way that the current regression tests
don't catch.)
commented out the structure accessor DECLAREs introduced
in flaky5.15

2001-09-25 01:56:37 Tree
[e92abc] by William Harold Newman William Harold Newman

0.pre7.37.flaky5.17:
(passes regression tests, but can't build itself)
tweaked foreign.lisp so that it doesn't try to compile
(WITH-ALIEN ((.. :EXTERN "summish")) ..) when
LOAD-FOREIGN isn't defined. (It'd be better to make
compilation work even when "summish" isn't loaded, but
I don't see an easy way to do that, and static-linking
pain isn't too unreasonable in the FFI anyway, so I'm
not going to work too hard on the problem.)

2001-09-24 19:18:37 Tree
[4eea8c] by William Harold Newman William Harold Newman

0.pre7.37.flaky5.16:
(passes type.impure.lisp, fails foreign.test.sh)
merged MNA's 2001-09-05 reworked defclass patch, since he
pointed on sbcl-devel today out that it's related to
the failure in type.impure.lisp
tweaked DECLAIMs in %EVAL so that SAFETY isn't trashed in
favor of COMPILATION-SPEED

2001-09-24 18:20:59 Tree
[520571] by William Harold Newman William Harold Newman

0.pre7.37.flaky5.15:
(This version still fails in type.impure.lisp.)
added more TYPE declarations to structure slot accessor
definitions to keep them from indiscriminately
working on any old structure type without signalling
errors

2001-09-24 16:59:27 Tree
[6b6b77] by William Harold Newman William Harold Newman

0.pre7.37.flaky5.14:
(This version builds and even runs many of its regression
tests, but fails in type.impure.lisp.)
made INLINE-SYNTACTIC-CLOSURE-LAMBDA &OPTIONAL ENV argument
into a required LEXENV-valued argument, since the
&OPTIONAL defaulting was just confusing when NIL
was passed
made COERCE-TO-LEXENV to encapsulate this idiom
suppressed more byte-compiler-related stuff

2001-09-24 14:47:44 Tree
[e7ea55] by William Harold Newman William Harold Newman

0.pre7.37.flaky5.13:
made FIND-ERROR-CONTEXT test BOUNDP *CURRENT-PATH* (instead of
assuming it'll be NIL outside the compiler)
got frustrated with stupid FUNCTIONAL-KIND dependencies in
particular and unmaintainability in general in the
byte compiler, and suppressed it..
..removed *byte* files from stems-and-flags.lisp-expr
..set *BYTE-COMPILE-DEFAULT*, and other
byte-compilation-related stuff in compiler/main.lisp,
to NIL
..removed "SB!BYTECODE" package
..removed BYTE-CLOSURE and BYTE-FUNCTION types e.g. in
EXTRACT-FUNCTION-TYPE

2001-09-24 00:27:09 Tree
[34e2a9] by William Harold Newman William Harold Newman

0.pre7.37.flaky5.12:
(This version gets all the way through building and loading
PCL, then dies with a compiler error in describe.lisp.)
FDEFINE-CAREFULLY shouldn't need to do %%DEFUN.
removed lotso REMOVEMEs in constraint.lisp

2001-09-22 17:43:25 Tree
[af110f] by William Harold Newman William Harold Newman

0.pre7.37.flaky5.11:
(This version makes it all the way into PCL warm init, where
it dies because %%DEFUN no longer exists.)
made cold FOP-FSET refuse to statically link duplicate
definitions (like the VECTOR-PUSH-EXTEND and COPY-SSET
problems a few versions ago)
deleted duplicate DEFUNs: OPEN-STREAM-P, DUMP-COMPLEX
made FIND-PRINTER-FUN use unique !WITH-CACHED-FUNCTION :STEM
to avoid duplicate DEFUNs
similarly for FIND-PREFILTER-FUN
ended up suppressing instruction printers completely for
now, since it's too hard to make the printer
function names unique for things like
PUSH-BYTE-IMM-WORD
The #!+SB-XC in DUMP-COMPLEX didn't do what it was intended
to do (should've been #+SB-XC) and didn't seem to be
strictly necessary anyway, so I deleted it completely.
removed some REMOVEMEs

2001-09-22 15:38:43 Tree
[e08cab] by William Harold Newman William Harold Newman

0.pre7.37.flaky5.10:
(This version gets through COMPILE-FILE of all
src/assembly/target/*.lisp files in warm.lisp.)
fixed errors in writing debug info in %COMPILE called from
%EVAL by requiring/providing non-NIL values of PATH
(e.g. the magical idiom '(CURRENT-SOURCE-START 0 0))
for %COMPILE and friends
SUB-SUB-COMPILE-FILE doesn't need to clear *SOURCE-PATHS*
any more, since we only use it once.
*CURRENT-PATH* should only be bound inside the compiler.
made SSET do :COPIER NIL so that default shallow copy doesn't
compete with explicitly-defined deep copy elsewhere

2001-09-21 16:58:13 Tree
[32b5b4] by William Harold Newman William Harold Newman

0.pre7.37.flaky5.9:
(This version does usable error reporting in warm init.)
added (ALIEN ..) declaration in CONTEXT-REGISTER
--
Remove REMOVEMEs.

2001-09-21 01:46:34 Tree
[6f3ad1] by William Harold Newman William Harold Newman

0.pre7.37.flaky5.8:
(The last version got through assem-rtns.lisp before dying,
but that was without SB-SHOW. I'd like to use
SB-SHOW to debug things, and that's causing
new problems. This version has fixes for
some of those problems, but still goes astray,
calling %COMPILE at runtime from somewhere
inside FIND-INTERRUPTED-NAME.)
made make-target-2.sh use SETQ to initialize *PRINT-FOO*,
so that /SHOWs for debugging the compiler at init
don't cause infinite amounts of output
made DEF-ALIEN-ROUTINE DECLAIM the return type of its DEFUN,
since if the return type isn't known, then code
downstream which tries to do should-be-simple alien
operations, like DEREF, on it will be implemented as
horrendous interpreter calls even when we try to
compile it
added (THE (ALIEN (* UNSIGNED-INT)) ..) in CONTEXT-PC too

2001-09-20 22:40:09 Tree
[ebc814] by William Harold Newman William Harold Newman

0.pre7.37.flaky5.7:
(This version gets hung up somehow in warm init in
src/assembly/target/array.lisp, after successfully
compiling and loading assem-rtns.lisp.)
still aiming for COMPILE with no toplevel stub..
..redid COMPILE in terms of %COMPILE
tweaked ASSIGN-MINIMAL-VAR-NAMES so that it can find SB!DEBUG
as well as SB-DEBUG, so it can work early in init
before packages receive their warm names

2001-09-19 23:26:44 Tree
[2390f0] by William Harold Newman William Harold Newman

0.pre7.37.flaky5.6:
still aiming for COMPILE with no toplevel stub..
..redid most of PROCESS-COLD-TOP-LEVEL-FSET as %COMPILE (to be
shared with CL:COMPILE later, I hope)
..redid FASL-DUMP-COLD-FSET to accept a dump handle instead
of an IR1 LEAF
..exported FASL-OUTPUT-ENTRY-TABLE so %COMPILE can see it

2001-09-17 18:34:54 Tree
[e09e9a] by William Harold Newman William Harold Newman

0.pre7.37.flaky5.4:
(This version gets all the way through cold top level forms
and into TOPLEVEL-INIT before somehow getting confused
and going into what seems to be an infinite loop in GC.)
restructured definitions of needed-at-cold-init stuff
so that their DEFUNs are at top level so they
will be statically linked by GENESIS..
..PACKAGE-INTERNAL-SYMBOL-COUNT
..INTERSECTION-SIMPLE-SUBTYPEP-TYPE-METHOD
also got rid of unused second return value from
PACKAGE-FOO-SYMBOL-COUNT
tweaked *COMPILE-PRINT* conditionalization in %COMPILER-DEFUN,
since (1) it's wrong in principle to use it outside
the compiler, and (2) in practice it was causing
problems in cold init before *COMPILE-PRINT* is
initialized
rewrote %COMPILER-DEFUN so it doesn't try to use compiler
variables when it's not in the compiler

2001-09-16 19:33:04 Tree
Older >