sbcl-cvs-import Log


Commit Date  
[89987c] by Christophe Rhodes Christophe Rhodes

0.9.8.4:
Beginnings of a Win32 merge.
... rearrange the build scripts to use input from files rather
than <<HERE documents.
... (no other changes; just working to get the meaty changes
isolated from the fluff)

2005-12-29 16:08:31 Tree
[3ceaa0] by Juho Snellman Juho Snellman

0.9.8.3:
Make EQUAL faster (about 50% improvement for short lists on x86-64).
As amazing as it might seem, there are actually real-world
applications where significant time is spent in EQUAL.

* Inline EQL in EQUAL
* Rearrange things a bit to enable the inlining
* Rewrite EQUAL to use a local helper function

2005-12-28 22:37:14 Tree
[b34a35] by Gabor Melis Gabor Melis

0.9.8.2: constraint propagation
* add (EQL LAMBDA-VAR LAMBDA-VAR) constraints on BIND and CSET
* add cast and test constraints to all eql lambda vars
* calculate the intersection of two lambda-var types when
encountering (IF (EQL X Y) ...) and neither X or Y is a subtype of
the other
* there is now only one invocation of FIND-AND-PROPAGATE-CONSTRAINTS
* since GEN depends on IN there is no COMPUTE-BLOCK-OUT shortcut
anymore, a full FIND-CONSTRAINTS-IN-BLOCK is done
* for blocks whose constraints won't be recalculated after the
prepass (see LEADING-COMPONENT-BLOCKS) do what
USE-RESULT-CONSTRAINTS would do is done on the prepass to save
time
* support test constraint propagation for blocks with with multiple
predecessors

2005-12-28 13:56:23 Tree
[cbfca6] by Juho Snellman Juho Snellman

0.9.8.1:
Make LDB quit on a ^D from a TTY instead of just ignoring the EOF.

* More consistent with the normal REPL
* The old behaviour would occasionally cause an infinite loop
of printing the LDB prompt, trying to read input, finding that
the tty was closed, printing the prompt again, etc.

2005-12-28 12:39:12 Tree
[242bb9] (sbcl_0_9_8sbcl.0.9.8) by William Harold Newman William Harold Newman

0.9.8:
release, tagged as sbcl_0_9_8

2005-12-27 17:18:54 Tree
[58ef9d] by Christophe Rhodes Christophe Rhodes

0.9.7.35:
Document the "SYS" logical host, as required by ANSI.

2005-12-22 10:10:47 Tree
[0aa292] by Alexey Dejneka Alexey Dejneka

0.9.7.34:
* New bug.

2005-12-21 13:00:49 Tree
[a01045] by Christophe Rhodes Christophe Rhodes

0.9.7.33:
Merge patch (Daisuke Homma sbcl-devel 2005-12-19) for stack
exhaustion on x86/SunOS.

2005-12-19 10:49:33 Tree
[2fb5b1] by Gabor Melis Gabor Melis

0.9.6.32:
* added support for (EQL LAMBDA-VAR LVAR) constraints
* fixed bug #233.b by paying attention to said constraints

2005-12-17 22:38:17 Tree
[fec361] by Christophe Rhodes Christophe Rhodes

0.9.7.31:
Merge essentially as "Pathname goodness" from CSR sbcl-devel
2005-12-15.
... define pathname host-specific PARSE-NATIVE and
UNPARSE-NATIVE methods.
... define NATIVE-PATHNAME, NATIVE-NAMESTRING and
PARSE-NATIVE-NAMESTRING in a direct analogy with
PATHNAME, NAMESTRING and PARSE-NAMESTRING.
... use NATIVE-PATHNAME both on what POSIX-GETCWD/ returns and
on files the user has asked us to load at the command
line. (Fixes bug #296 and *DEFAULT-PATHNAME-DEFAULTS*
being wrong when a component of the current
directory contains a pathname metacharacter in
"[*?\\")
... don't create a string from --load (and --disable-debugger)
that just gets read again; instead allow
process-eval-options to deal with non-strings too.
... tease *physical-host* (the default physical host on the
platform) and *unix-host* apart ever so slightly, with
obvious knock-on benefits for ports to non-Unixoid
platforms.
... sb-posix no longer needs its own implementation of
NATIVE-FILENAME.
... delete unused UNIX-MAYBE-PREPEND-DIRECTORY.
... some tests and some documentation.

2005-12-16 15:06:09 Tree
[a01412] by Nikodemus Siivola Nikodemus Siivola

0.9.7.30: robustify sb-introspect when used with extended function names
...so that (find-definition-sources-by-name '(setf foo) :class) will
return NIL instead of signaling an error.
...this makes it simpler to support M-. for SETF-names in Slime.

2005-12-15 22:08:38 Tree
[559d0d] by Juho Snellman Juho Snellman

0.9.7.29:
Oops, fix cheneygc build (thanks to Jon Allen Boone).

2005-12-15 20:18:09 Tree
[707695] by Juho Snellman Juho Snellman

0.9.7.28:
Make MAP-ALLOCATED-OBJECTS use gencgc pagetable information to
skip over free dynamic space regions.

* More accurate ROOM results (previously all empty space up to
the fake allocation pointer would get counted as conses)
* Usually faster
* Needed for some GC work that might get committed in the future
* Genesify the GENCGC page size, while since it's now needed
in Lisp

2005-12-14 03:39:23 Tree
[b392b3] by Juho Snellman Juho Snellman

0.9.7.27:
Oops. Off-by-one in 0.9.5.68 meant that the shorter MOV instruction
wasn't getting used quite as often as possible. Use a check for
(signed-byte 32) instead of (signed-byte 31) to pick between the
C7 and B8 opcodes. Shrinks the x86-64 core by 70k.

2005-12-14 02:02:11 Tree
[2e4a90] by Christophe Rhodes Christophe Rhodes

0.9.7.26:
Log a bug in fill-pointer streams, from tichy on sbcl-help.

2005-12-13 13:56:00 Tree
[f9b113] by Juho Snellman Juho Snellman

0.9.7.25:
Miscellanea.

* Don't return structures from FIND-DEFINITION-SOURCE-BY-NAME when
asking for classes
* Fix typo in error message for --eval '1 2' (reported by ignas
on #lisp)
* Add a couple of comments
* Add myself to the initials glossary

2005-12-11 04:23:04 Tree
[7201fe] by Thiemo Seufer Thiemo Seufer

0.9.7.24:
Don't flush icaches for each moved code vector when using
cheneygc, it is more efficient to do that once for the whole
dynamic space.

2005-12-10 22:27:59 Tree
[4e0e25] by Christophe Rhodes Christophe Rhodes

0.9.7.22:
Oops II. Make (SETF CLASS-NAME) and (SETF
GENERIC-FUNCTION-NAME) return the right values.

2005-12-09 15:57:43 Tree
[0f807a] by Christophe Rhodes Christophe Rhodes

0.9.7.22:
Whoops. Make (setf class-name) and (setf generic-function-name)
generic functions again; AMOP may not specify it, but ANSI
does...
... also log some bugs

2005-12-09 11:02:28 Tree
[b49f85] by Christophe Rhodes Christophe Rhodes

0.9.7.21:
Make SB-PCL::MAP-ALL-CLASSES hit each class Once And Only Once.

2005-12-08 18:12:59 Tree
[4ace72] by Christophe Rhodes Christophe Rhodes

0.9.7.20:
Add documentation of the various SLOT-VALUEish optimizations
performed.
... and some FIXMEs where either (a) I'm not sure what's going on
or (b) something fishy is happening.

2005-12-08 17:43:45 Tree
[830db4] by Christophe Rhodes Christophe Rhodes

0.9.7.19:
Merge patch (DL "external-formats problem..." 2005-12-03
sbcl-devel)
... and (fixed) test case 2005-12-07

2005-12-07 15:57:21 Tree
[7a7a52] by Christophe Rhodes Christophe Rhodes

0.9.7.18:
Merge patch (David Lichteblau sbcl-devel
"define-optimization-quality compute-debug-fun" 2005-12-03)

2005-12-07 15:24:41 Tree
[b544f7] by Christophe Rhodes Christophe Rhodes

0.9.7.17:
Document what we've learnt about discriminating functions
in the internals manual
... now we depend on dot (graphviz) to draw pretty state
transition graphs.
... lots o' text. I hope it's clear.

(it would be good to upload this to the web somewhere, in
order to be the top hit for e.g. "PCL CONSTANT-VALUE" on search
engines, so that the next person to hit the swamp has a chance
to get out before it's too late. Hack the Makefile enough so
that we can build an html version of an internals manual.)

2005-12-07 14:20:04 Tree
[fbde18] by Christophe Rhodes Christophe Rhodes

0.9.7.16:
More PCL smallification.
... the ACCESSOR-SLOT-VALUE optimization creates a generic
function (using load-time-value) when it sees
a form like (slot-value x 'constant). That's fine,
but...
... these generic functions are also created at class
initialization time, three per slot. This hurts
now that we're creating classes eagerly (so that the
mop functionality works) as even condition classes
and structure classes cause these gfs to come into
being.
... so, rearrange things so that only those generic functions
which are needed are created. Never create one with
a slot-missing method, as the automatically-generated
method will fall through to the full call and get there
eventually, anyway.
... this causes slot-missing from slot-value outside of methods
to be slower if (and only if) no slot of that name
exists in the image anywhere. We could potentially
improve fall-through-to-slot-missing performance.

(this shaves off 2.5Mb from sbcl.core on my x86)

2005-12-05 18:01:27 Tree
Older >