This fixes the ANSI non-compliance passed down from early
CMUCL (1991-10-31). The bug was that FDEFINITION did not return the
same value as SYMBOL-FUNCTION for symbols (lp#799533). The fix is an
incompatible change but only affects encapsulated (e.g. TRACEd and
SP-PROFILE:PROFILEd) non-generic functions.
However, encapsulations belong to the definition (the name) and not to
the function associated with the name, so (SETF (FDEFINITION NAME) FN)
strips encapsulations from FN and keeps the encapsulations of NAME.
- SB-EXT:UNENCAPSULATED-FUNCTION was added.
- Encapsulated functions are printed as
#<FUNCTION FOO ENCAPSULATED>
or, if they have no proper name, as
#<FUNCTION FOO ENCAPSULATED {B1782332}>
This is the same as for unencapsulated functions except for the
additional " ENCAPSULATED".
- Unused slots were removed from SB-PROFILE:PROFILE-INFO.
- Since generic function identity is stable in face of encapsulation,
CLOS is largely unaffected by this change.
Any closure can be given a name, via a global hashtable.
But the special case of storing the name in the closure is used if either
it had a spare slot, or the namer does not require preserving identity.
The limitation still exists that the debugger does not see closure
names in a backtrace unless the creator of the closure contained
(DECLARE (OPTIMIZE SB-C::STORE-CLOSURE-DEBUG-POINTER))
Otherwise we lose the backpointer from the simple-fun to the
closure that invoked it. I think that's a separate concern.
Also:
- rewrite %FUN-NAME and (SETF %FUN-NAME) to be less ugly and
dispatch on type more efficiently.
- use SB-MOP:GENERIC-FUNCTION-NAME instead of SLOT-VALUE.
- explain a fixme in gc-common
* Add various subclasses of STYLE-WARNING in src/code/condition.lisp
for most of the cases signaled around the system. (A few style
warnings remain SIMPLE-STYLE-WARNINGS for now.)
* Make STYLE-WARN be like WARN (take a condition class name and
arguments).
* Change over many callers of STYLE-WARN to supply names of new
STYLE-WARNING classes and initargs for those conditions.
* Set the default value of SB-EXT:*MUFFLED-WARNINGS* to a type that
includes uninteresting redefinitions of functions and methods.
FINALLY!
Fixed bugs related to DOUBLE-FLOAT-EPSILON on x86. Die, bug
#45, die a horrible death.
... make lisp code run with the fpu set to 53-bit mantissa;
... add code in number stack allocation to set the fpu to
64-bit precision when calling out to C, conditional on
new optimization quality SB-C::FLOAT-ACCURACY
(unexported, undocumented, etc)
... use FLOAT-ACCURACY around syscalls, since they don't
involve the FPU.
Also add code to allow fpu precision control in the (nominally
private) sb-int:set-floating-point-modes.
(this checkin fixes not only bugs #45 and #118, but also six of
PFD's EPSILONS tests and 29 ieeefp-tests related to +, -, *, /
and sqrt)
More refactoring around COMPILER-WARN
... new src/code/cross-condition file to house cross-compiler
definitions of conditions
... new condition types for format warnings
... (note that this separation between xc and target definitions
allows us to make warnings of different severity
during cross-compilation easily)
CLISP build megapatch
... mostly putting #-SB-XC in front of :COMPILE-TOPLEVEL,
because clisp gives a full warning for function and
macro redefinition;
... workaround clisp's buggy pretty printer by not exercising it
as much: use (INHIBIT-WARNINGS 3);
... explicit :INITIAL-ELEMENT 0 when we're using 0 to mean
"uninitialized" in MAKE-ARRAY;
... SPECIAL-OPERATOR-P isn't a good test on the host for what
can become a target macro;
... slightly more portable floating point logic:
Explicitly set *READ-DEFAULT-FLOAT-FORMAT* so that we
don't create host LONG-FLOATs by accident;
LOAD-TIME-VALUE magic for negative floating point zeros;
Minor associated text file frobbage
... braindump some unrelated TODO items
Obligatory runtime code improvement
... fix one warning in gencgc.h
=== Threads merge, 12.5 metres ===
Added ppc vop for CURRENT-THREAD-OFFSET-SAP : note that other
ports (alpha, sparc, etc) wil need this too
Fix the PPC symbol hashing problem with symbol names < 4
characters long (thanks CSR)
Replace CONTROL_STACK_FOO with thread-> or SymbolValue as
appropriate in various non-x86 places (cheneygc, ldb backtrace
etc)
get_spinlock defn moved into $arch-arch.c, as unlikely to vary
much between one OS and another on a given arch. Other arches
need to add this too, but for non-threaded ports the dummy
version in ppc-arch.c will do fine.
Stub arch_os_get_current_thread, arch_os_thread_{init,cleanup}
added for Linux/PPC: needs adding to other ports
Add missing UNIX-SETSID definition, used in MAKE-LISTENER-THREAD
Fix nasty bug introduced in 0.pre8.1 or so, wherein #S no longer
worked. (mea maxima culpa)
... export STRUCTURE-CLASSOID
... write a test for it, so it doesn't happen again
Now we can build from ourselves again, probably :-/
merged Robert E. Brown shush-the-compiler patch (sbcl-devel
2002-12-13)
minor changes...
...removed DECLAIM FTYPE for
SLOT-ACCESSOR-INLINE-EXPANSION-DESIGNATORS on the
theory that it's too fragile (since (1) S-A-I-E-D does
currently return functions, but could validly return
nonfunctions in some later implementation, and (2)
SBCL's declarations-are-assertions still doesn't work
right for DECLAIM FTYPE)
...sometimes used THE instead of DECLARE
(didn't do yet, but still intend to: add some documentation
related to drichards' #lisp question about :NOT-HOST)
Adjust test file running environment so that *.pure.lisp tests
are run with assertoid loaded.
... give assertoid its own package
... use it in run-tests.sh for *.pure.lisp
... also use it where applicable in *.impure.lisp
Fix one "outstanding embarrassment" (as per CSR sbcl-devel
2002-12-10): no more building of compiler files in
warm load.
... play symlink games instead, to build the troublesome
assembly files twice under different names (will need
adjustments to work on non-Unix filesystems (e.g.
NTFS) but no more than the rest of the build).
Working INLINE inside MACROLET for user code.
(more or less as per CSR sbcl-devel 2002-12-03)
... change tricky cross-compiled inline functions to source
transforms (including adding some functions to the
function database)
... use now-working FUNCTION-LAMBDA-EXPRESSION to get the
expander function for local macros
... build a complex LAMBDA-WITH-LEXENV for inlining user code
(like CMUCL's INLINE-SYNTACTIC-CLOSURE-LAMBDA does)
Some miscellaneous frobs
... actually signal a style warning for array code (though it
might actually be in currently-dead-but-should-be-alive
code)
... actually test the return value in the second half of
filesys.test.sh
... minor text adjustments (no more calling this period "early
0.7.x")
embarrassment reduction: transformed ugly duplicate
DEF-BOOLEAN-ATTRIBUTE into ugly OAOO (and also renamed
it to !DEF-BOOLEAN-ATTRIBUTE, since not aspiring to
make it exist on the target seemed to make it easier
to think about)
Also, as long as I'm renaming it anyway, it has no special
connection to the compiler and is used in SB!ASSEM too,
I moved !DEF-BOOLEAN-ATTRIBUTE from SB!C to SB!INT.
new reports and restart names trying to clarify what those
redefine-DEFSTRUCT CERROR restarts mean (cribbing
somewhat from CMU CL CVS)
While I'm at it, fix the behavior of the ABORT restart in
%REDEFINE-DEFSTRUCT, so that it leaves the system
unaffected as one might reasonably expect. (Before,
ABORTing then retrying an incompatible definition a
second time would allow the incompatible definition to
take place the second time with no error.)
As long as I'm proof-reading restart reports, use similar
FORMAT ~< tricks in some other restarts too.
Delete some stale symbols from, and do one or two other fixups
to, package-data-list.lisp-expr
... e.g. long-float symbols conditional on #!+LONG-FLOAT
... also s/#+x86/#!+x86/
Add Eric Marsden's script to contrib/
READ-VAR-INTEGER and READ-PACKED-BIT-VECTOR are external to
SB!C, so don't need double colons in SB!C: prefixes.
made (FORMAT "foo" "bar") fail earlier, at
FILL-POINTER-OUTPUT-STREAM ctor time, instead of
when the FILL-POINTER-OUTPUT-STREAM is first used for
output
got rid of *TOPLEVEL-LAMBDA-MAX* and
*PENDING-TOPLEVEL-LAMBDAS* (and FORCE-P arg to
SUB-COMPILE-TOPLEVEL-LAMBDAS and
COMPILE-TOPLEVEL-LAMBDAS, and TOPLEVEL-CLOSURE
in COMPILE-TOPLEVEL, and various now-redundant
FORCE-P-only calls to COMPILE-TOPLEVEL-LAMBDAS)
Alpha build fix
... define the relevant types earlier in the build
... s/INTEGER-WITH-A-BITE-OUT/UNSIGNED-BYTE-WITH-A-BITE-OUT/
Array performance enhancement
... remove the (SAFETY 3) declaration from HAIRY-DATA-VECTOR-{REF,SET}
... write tests for AREF beyond array bounds
Buglet fix in pack.lisp
... put FILL arguments the right way round
minor OAOO FIXME for GENESIS...
..made the FOO-ENTRY-TYPE-CODE parameters propagate
automatically into sbcl.h instead of maintaining
separate copies of their definititions in core.h (and
renamed them, s/entry-type-code/core-entry-type-code/,
to help make them more painfully specific now that
their scope is wider)
...similarly propagated FOO-SPACE-ID automatically into
sbcl.h, and s/foo-space-id/foo-core-space-id/
The missing piece in the OpenMCL build...
... move some clauses around in cross-compilation float logic.
NB: I am _not_ sure that this is correct in any sense
other than the empirical "it works". The IEEE-representation
logic for the cross-compiler needs review by someone
who knows what an IEEE float looks like. For now, though...
... now SBCL builds under OpenMCL!
Fix bug 177 (bogus type warnings from CALL-NEXT-METHOD) with
a hack, wrapping the offending code in an opaque
identity function.
Add comments in INVOKE-EFFECTIVE-METHOD-FUNCTION
explaining CSR's insight that the ,@(WHEN ...)
clauses are optimizing slot access (with slots
encoded as FIXNUMs). Also remove DECLARE
NOTINLINE GET-SLOTS-OR-NIL now that I understand
what the code is doing.
catching stack overflow, part III...
...redid *STACK-EXHAUSTION* value as SAP, because although
FIXNUM is cute and should be implementable efficiently,
it seems to be awfully annoying to work with
...made %DETECT-STACK-EXHAUSTION actually check for the
problem
...set up machinery to try to handle the problem reasonably
gracefully
...added basic regression test
moved EXPORT of '*SHEBANG-BACKEND-SUBFEATURES* alongside
EXPORT of '*SHEBANG-FEATURES* so that chill.lisp
works again
merged CSR SPARC port patch (sbcl-devel 2002-02-12, plus
lotso new files through FTP)
tweaking patch...
...s/ARCH_HAS_FOO/ARCH_HAS_FOO_IN_SIGCONTEXT/
...updated sbcl.1
...added 'typedef os_context_register' for OpenBSD
...added ARCH_HAS_STACK_POINTER for x86 (and added
x86-arch.h to hold it)
...renamed fpregister and fp.register to float.register (to
avoid the ambiguity with "frame pointer" that I
experienced when first trying to figure this out,
since even though for a given architecture it's pretty
unambiguous, in architecture-neutral code it's not)
...added x86-bsd-os.h
these versions just piled up on my computer and then I checked 'em
into CVS all at once.)
0.7.1.5:
made TRANSFORM-CALL provide more informative DEBUG-NAMEs
factored out COMBINATION-FUN-SOURCE-NAME and used it to
support this
0.7.1.6:
tweaked comments
(hunted fruitlessly for bug 147 fix)
0.7.1.7:
(hunted fruitlessly for bug 148 fix)
rewrote MAYBE-EXPAND to try to increase clarity
0.7.1.8:
factored out FUNCTIONAL-SOMEWHAT-LETLIKE-P and
FUNCTIONAL-LETLIKE-P
fixed part of the misbehavior in the bug 148 test case (but
not bug 148 itself, alas) by removing the assumption
that non-null FUNCTIONAL-KIND implies
FUNCTIONAL-SOMEWHAT-LETLIKE-P
0.7.1.9:
still trying to fix bug 148...
...stopped MAYBE-REANALYZE-FUN from trying to reanalyze :DELETED
functionals
s/maybe-reanalyze-fun/maybe-reanalyze-functional/
s/reanalyze-funs/reanalyze-functionals/
s/new-funs/new-functionals/
0.7.1.10:
still trying to fix bug 148...
...IR2-CONVERT-CLOSURE shouldn't be called on :DELETED functionals!
...Given that the :DELETED functional is making it all the way
to the IR2-CONVERT-CLOSURE stage, maybe the failure
in MAYBE-REANALYZE-FUNCTIONAL that I made go away in
0.7.1.9 was a good thing. Reinstate it, though more
clearly (as "shouldn't be reanalyzing :DELETED functional"
rather than a type error when trying to find the COMPONENT
of a LAMBDA) than before.
...stopped IR2-CONVERT-CLOSURE from trying to intensively
check CLAMBDA-to-COMPONENT relationship invariants
for :DELETED CLAMBDAs
made INVALID-FREE-FUN-P return true for :DELETED FUNCTIONALs
just on general principles
0.7.1.11:
s/local-call-lossage/locall-already-let-converted/
various puttering and tidying trying to understand bug 148
specifically and code deletion generally
0.7.1.12:
Having walked through the bug 148 problem more carefully, I can
see that before KIDIFY1 is deleted, it's first LET
converted. Ergo, a :DELETED value is consistent with
LET conversion after all, so...
...relaxed the change in IR1-CONVERT-LOCAL-COMBINATION
made in 0.7.1.8, so that now :DELETED is assumed
to be due to LET conversion after all
0.7.1.13:
made :ENCAPSULATE T the default for TRACE, since the
breakpoint-based version still doesn't work reliably
and since the ANSI description of TRACE is partial to
tracing named things anyway
APD bug 111 patch sbcl-devel 2001-12-30
APD PCL INHIBIT-WARNINGS patch ("Re: [sbcl-devel] sbcl style"
sbcl-devel 2002-01-23)
nibbling away at bug 137: making functions defined by
DEFMETHOD have debug names not e.g.
"#'(LAMBDA (SB-PCL::.PV-CELL. SB-PCL::.NEXT-METHOD-CALL. COMMON-LISP-USER::X) (DECLARE #) ...)"
but instead (:METHOD FOO (INTEGER))...
...added NAME-METHOD-LAMBDA and BODY-METHOD-NAME, and used 'em
...tweaked %METHOD-NAME declared values to look more like
modern CLOS syntax
...made NAMED-LAMBDA treat not-legal-as-source-name names as
debug names, so it barfeth not when fed method names
...tweaked BACKTRACE printing so that it won't truncate the
shiny new method names into e.g. (:METHOD FOO #)
deleted unused WALK-NAMED-LAMBDA
(I hope this looks a lot like 0.7.0. I'll putter on and around
it for a while, xc'ing under CMU CL, building some
applications, etc. Then if there isn't too much
surprise, maybe I can finally ship it!)
late-breaking incompatible cleanup: The default USE list
for DEFPACKAGE and MAKE-PACKAGE is
implementation-defined. Let's just make it NIL, the
way God intended, instead of trying to guess. (Anyone
who actually wants something random can just
:USE (PACKAGE-USE-LIST :CL-USER) anyway.:-) So...
...Implementation-defined things are supposed to be
documented. Put this in the doc string of MAKE-PACKAGE.
...Putting it in NEWS would be good too.
more s/argument/arg/...
...s/entry-argument/entry-arg/
...s/argument-list/arg-list/
...s/compile-argument/compile-arg/
...s/associate-argument/associate-arg/
...s/compute-argument/compute-arg/
...s/reference-argument/reference-arg/
...*not* s/one-argument/one-arg/, since CL defines
:IDENTITY-WITH-ONE-ARGUMENT as a kw arg for
DEFINE-METHOD-COMBINATION
...s/from-argument/from-arg/
abbreviating ARGUMENT as ARG...
...s/error-argument/error-arg/
...s/argument-count/arg-count/
...s/bogus-argument/bogus-arg/
...s/key-argument/key-arg/
...s/constant-argument/constant-arg/
also changed ARGUMENT to ARGS in the context of ARG-COUNT-ERROR
NJF patch I from sbcl-devel 2002-01-14: "Keywords are
self-evaluating, right? They don't need to be quoted?
Then I wonder why the writers of PCL felt it necessary
to do so. This patch fixes that."
NJF patch II from sbcl-devel 2002-01-14: "This patch does
away with the strange business of the PUSHP argument to
DEFINE-FOP (and DEFINE-COLD-FOP) being
(MEMBER '(T NIL :NONE)) and implements the suggestion
at the top of src/code/fop.lisp for fixing it."
chased down ramifications of s/offs-hook-function/offs-hook-fun/,
as per Alexei Dejneka broken disassembler bug report
sbcl-devel 2001-01-14
the poor neglected user manual...
...added some information about compiler-only-ness in the
introduction
...removed some "I'm sorry about this bug, give me time" stuff,
since time has passed, and the bugs are gone:-)
belatedly incremented fasl file version
that it's log message got lost. So this checkin has two log messages,
.131 for the changes in the previous checkin and .132 for the changes
in this checkin.)
0.pre7.131:
s/variable/var/ in...
...fun-variable
...symbol-variable
...find-variable
...variables*\*
...make-variable
...iteration-variable
...declare-variable
...when-it-variable
...named-variable
...1-variable
...compute-variable
...minimal-variable
...free-variable
...let-variable
...lexenv-variable
...(not set-variable because there's already a VAR slot for the
SET structure type, gah, so try s/set-variable/setq-var/)
...convert-variable
s/lexenv-functions/lexenv-funs/
Rename NAMED-VAR function to LOOP-NAMED-VAR, for consistency
with other LOOP operators and with the LOOP comments
which refer to LOOP-NAMED-VAR.
0.pre7.132:
learning that I don't even get people's *names* right
in CREDITS:-(
addressed APD bug report 2001-01-13...
...made a BUGS entry for pretty-printing nested backquotes
...scattered references to compiler-only implementation
through doc/sbcl.1
other corrections and tweaks in doc/sbcl.1
MNA "patch for bug 105 and some other cleanups" (sbcl-devel
2002-01-13)...
...fixed bug 105: made (DESCRIBE 'STREAM-READ-CHAR) work right
even when there're no methods on the GF yet
...tweaked format of DESCRIBE of methods, simplifying
GENERIC-FUNCTION-PRETTY-ARGLIST stuff
...tidied up tests/ stuff, deleting temporary files after use
...deleted BUGS entries (105 entry, and stale 99 entry too)
tweaked MNA patch...
...explicitly saying "It has no methods." for bare GF
(There were >300 matches to
"egrep -sn '^\(def[^(;&]*function' ..." before.)
s/to-function/to-fun/
s/hook-function/hook-fun/
s/describe-function/describe-fun/
s/bogo-function/bogo-fun/
s/fop-fun/fop-fun/
s/not-function/not-fun/
s/named-function/named-fun/
s/nil-function/nil-fun/
s/\<t-function/t-fun/
s/simple-function/simple-fun/
s/function-call/fun-call/
s/move-function/move-fun/
s/traced-function/traced-fun/
s/function-or-lose/fun-or-lose/
s/disassemble-function/disassemble-fun/
s/get-function/get-fun/
s/output-function/output-fun/
s/1[-a-z0-9]*function/1$1fun/
s/note-function/note-fun/
s/check-function/check-fun/
s/function-cache/fun-cache/
s/disassem-function/disassem-fun/
s/function-cost/fun-guessed-cost/
s/function-value/fun-value/
not just s/function/fun/, but also clearing up mess of (1)
LOSSAGE/SLIME (esp. SLIME), what do they mean? and
(2) "ERROR" and "WARNING" used in names even though
the condition system isn't involved...
...s/error-function/lossage-fun/
...s/warning-function/unwinnage-fun/
...s/slime/unwinnage/
other confusion/inconstency...
...s/compiler-style-warning/compiler-style-warn/ when used
as verb (as opposed to used as condition name)
...and similarly for s/compiler-warning/compiler-warn/
s/test-function/test-fun/
s/\*test-fun\*/*ctype-test-fun*/
(There are now some 230 hits to the egrep at the start, mostly in
src/pcl/*, where I'm trying to tread lightly to avoid
gratuitously breaking quasistandard MOPish things.)
hunting bug 138...
...Poking around in the bug 138 test case, I found that the
failure is occurring for a LAMBDA which represents
the inline expansion of LAYOUT-INHERITS. It seems
correct for :INLINE LAMBDAs not to be in
COMPONENT-LAMBDAS, so it looks as though the bug is
in the AVER, not the code it's trying to protect,
so I added a new :INLINE case to the check logic.
Alas, that only converted bug 138a into bug 138b.
...added some assertions and comments hunting for 138b
redid DUMP-CODE-OBJECT loop for clarity as LOOP not DO
function name mess, continued...
...used NAMED-LAMBDA in DEFMACRO-MUNDANELY DEFUN
...tweaked non-toplevel handling of COLD-FSET so that it
notifies the compiler about the source name (using
NAMED-LAMBDA to do so)
deleted dead symbols from package-data-list.lisp-expr as per
APD sbcl-devel 2001-12-30
deleted bug 128 as per NJF sbcl-devel 2001-12-30
merged DB patch sbcl-devel 2001-12-30 for (mostly)
conditional-on-SB-LDB gross errors dating back to
all the global substitutions I was doing a while
ago (and added :SB-LDB to my default
customize-target-features.lisp, in hopes that maybe
I can catch more such problems myself)
