series-old Log


Commit Date  
[4bc70b] (SERIES_2_0) by matomira matomira

Series 2.0 - Change details in RELEASE-NOTES.

2000-03-03 19:17:15 Tree
[47c979] (SERIES_1_54) by toy toy

Even more changes from Fernando.

2000-03-02 17:49:29 Tree
[9729d7] by toy toy

Some more changes from Fernando (actually two sets of changes):

- A little bit more bottom-up def motion (notably defS and
COLLECT because of a warning if gatherers are compiled
first)
- Added UNINITIALIZED and DEFAULTED typedefs (unused)
- Added aux and output variable initialization (fragL aux
form now accepts also (var decl init)
- Removed any fragL prolog code initializing vars to
constants and moved to aux decl form.
- Fixed bug in COLLECT. CL: qualifier missing before a
MULTIPLE-VALUE-BIND

- REMOVED LONGSTANDING defS cleanup BUG (inner function in
flet defined before top-level macro def)
- Additional defS and defS-1 simplification (and slet*)
- renamed LETIZE as DESTARRIFY

2000-03-01 14:46:22 Tree
[e11f9c] by toy toy

Forgot to export COLLECT-PRODUCT (caught by Fernando).

2000-02-28 17:40:15 Tree
[33dee5] by toy toy

o Make when-bind always available
o Add deftype generator for Allegro. I'm not sure this is
correct, but it let's ACL compile series, albeit with
warnings about generator type being defined twice. (Need
to find a better solution.)

2000-02-28 17:37:31 Tree
[2e915f] by toy toy

o Fernando added an indefinite-extent declaration and uses
it in the one place where it's needed.
o Fernando renamed split-assignment to detangle2 and
corrected some bugs in my version.

2000-02-23 15:27:02 Tree
[1b2b42] by toy toy

Remove the cmu version from scan-range. It was generating bad
initialization code for things like (scan-range :length 10 :type
'single-float).

2000-02-22 23:37:22 Tree
[ff1404] by toy toy

o Let's try to set the correct package for CLISP depending if we're in
ANSI mode or not. (Not sure this really works.)

o In test 280, we probably really want to use the COMMON-LISP-USER
package instead of just USER. Mostly for CLISP where the USER
package is not a nickname for COMMON-LISP-USER.

2000-02-22 23:25:39 Tree
[215283] by toy toy

o One of Fernando's uses of dynamic-extent was wrong, as Fernando
points out.

o CLISP apparently has a bug in loop such that split-assignment is
broken. Replace that with an equivalent do loop.

2000-02-22 22:25:51 Tree
[445af8] by toy toy

Fernando added dynamic-extent declarations wherever needed.

2000-02-22 15:21:38 Tree
[c72e20] by toy toy

Let's not use fix-types for CMU in optimize-producing. This means the
compiler can't optimize things as well as it could, and I (RLT) want
to see these warnings.

2000-02-11 14:45:42 Tree
[8a3964] by toy toy

Fix a typo that got in the last few patches: LET should really be
CL:LET. (From Fernando.)

2000-02-10 17:15:11 Tree
[ed8322] by toy toy

Fernando made these changes: Replace deftest with defok,
defcmumismatch, and defcmukernel if some older versions of CMUCL
produce bad results on these tests. Newer versions of CMUCL pass
these without problems.

2000-02-09 22:52:00 Tree
[9c3eba] by toy toy

Changed all occurrences of defunique to be just defun and added a
comment on where the function is called.

2000-02-09 22:46:00 Tree
[55ccdb] by toy toy

o As discussed with Fernando, the "optional" type is renamed to
null-or.
o Cleaned up and indented some of the comments.

2000-02-08 17:08:36 Tree
[403574] by toy toy

A few more changes from Fernando:

o All functions called from a single site are defined with DEFUNIQUE
for documentation purposes (and eventual inlining via DEFEMBEDDED).
o Fixed the long standing bug that install uninterned everything that
it didn't like. Shadowing import is used now.

There were some other changes that I (RLT) don't understand.

2000-02-04 23:05:57 Tree
[eaaf6b] by toy toy

In collect-nth, counter is really a fixnum.

2000-02-04 22:52:45 Tree
[1c3985] by toy toy

o Some more changes from Fernando. This fixes some bugs that show up
in the test suite. The test suite now passes on CMUCL.

o Fixed up some declarations that used the optional type when, in
fact, it didn't. (Hope I got these all right.)

2000-02-04 16:34:30 Tree
[d3f59c] by toy toy

Two major fixes:

o Bug in collect (missing set of parens)
o Change some defconstants back to defvar. (Tickles CMUCL inf loop).

2000-02-03 17:30:08 Tree
[54eaa6] by toy toy

Here are the changes that Fernando made. I (RLT) don't claim to
understand everything that was changed or why.

1. Removed 1 redundant eval-when
2. Sorted functions so that it can be loaded w/o `undefined function'
warnings.
3. Added inline declarations for all functions called from a single
site.
4. Sorted functions so that inlining will work even if a compiler does
not inline forward references to functions defined in the same
file.
5. Setup eval-when's so that it can be compiled w/o having to load the
source first.
6. Simplified redundant (OR NULL T) to T
7. Simplified redundant #'(lambda (x) (foo x)) to #'foo where foo is a
function.
8. Fixed so it won't complain when LispWorks adds
CLOS::VARIABLE-REBINDING declarations after CLOS macro
transformations (general support provided via the constant
allowed-generic-opts).
9. Added support for MACROLET on LispWorks (necessary because of CLOS
macro transformations).
10. Only declare variables as (OR foo NULL) on implementations that
won't allow to store NIL otherwise (currently, only CMUCL).
11. Added specialization to series declarations (eg: (SERIES FIXNUM)).
12. Do more precise type propagation in PRODUCING forms.
13. Allow `SETF like SETQ' in PRODUCING forms.
14. Added COLLECT-PRODUCT.
15. Extended SETQ-P to take into account multiassignments (not used
yet). This should still be trivially generalized to support PSETQ
and SETF, BTW.
16. Added DEFTYPE for GENERATOR so that LispWorks and CMUCL won't
complain "because it's not a list" (IT IS!!)
17. Replaced PROCLAIMS with DECLAIMS.
18. Replaced DEFVARs with DEFCONSTANTs where appropriate.
19. Removed function namespace pollution by defS-generated code.

2000-02-02 21:31:44 Tree
[04f681] by toy toy

Merged 1.32.2.2 (Fernando's changes) with 1.37.
I hope I got this right.

2000-01-20 18:19:21 Tree
[e8f1dc] by toy toy

Need to import compiler-let from the extensions package in CMUCL.

1999-12-01 16:09:05 Tree
[6ab183] by toy toy

o The backquoting of make-sequence added in 1.35 is wrong, I think.
To make it worse, I lost my test cases that caused me to do this.
Stupid!

o Updated a few doc strings. Should really add some more so that
series is a little bit more self-documenting.

1999-10-04 17:46:55 Tree
[87b016] by toy toy

The change in 1.17 for scan-range was backwards: The CMUCL version
should get the coercion stuff and the non-CMUCL version shouldn't.
(Hmm, should I remove this dependency on platform? Probably.)

1999-09-15 14:30:55 Tree
[269718] by toy toy

Backed out the changes in INIT-ELEM: they were mostly correct I
think. Just needed to stick a backquote before the make-sequence
stuff so we create the sequences at run-time, not compile-time.

1999-09-15 14:11:29 Tree
Older >