1. Jasper St. Pierre
  2. pypy-avm

Commits

Show all
Author Commit Message Date Builds
Armin Rigo
The basic threshold is now larger by 2% than it used to be.
Branches
counter-decay
Armin Rigo
Fix the test.
Branches
counter-decay
Armin Rigo
hg merge default
Branches
counter-decay
Armin Rigo
- fix the tests to expect the new format of descrs. - fix match() to always raise InvalidMatch if the match fails because I found that at least one test was not doing "assert match(..)" but just "match(..)" which always passes... - test_generator still fails, but that should be fixed on the new version of the 'counter-decay' branch.
Armin Rigo
Change again the repr of CallDescrs, this time to standardize the format with the other Descrs and to include more information.
Armin Rigo
Fix the repr of CallDescrs to not include a "," because that confuses the jitlogparser to no end...
Armin Rigo
A minimal version of counter decaying.
Branches
counter-decay
Armin Rigo
Python 2.5 compat
Armin Rigo
Simplify and "stand-alone-ize" the clean-up of old jitcells from the jitcell dict.
Branches
counter-decay
Armin Rigo
Merge default again, but keeping the pieces that I want to keep. The idea is to keep some mecanism to clean-up jitcell dicts, and to re-add some minimal form of decaying.
Branches
counter-decay
Armin Rigo
Un-merge counter-decay, which was definitely not really good. More work should be going on in the branch. This cancels 5309a1389556, e790db7af776 and 15811e23d71a.
Armin Rigo
hg merge default Reactivate this branch, and cancel its merge on trunk, which was definitely not really good.
Branches
counter-decay
Alex Gaynor
Make the Python StringBuilder raise exceptions in the same place as the translated one, and get str.replace in Python raising the correct exception for overlfows on 32-bits again. Remove a duplicate test.
Armin Rigo
This was meant to say (2, 6) and not (2.6), I am pretty sure. Fixed, but then the test fails on top of Python 2.6...
Armin Rigo
Merge op_malloc_gc: move out of the jit backend the NEWxxx operations. Now they are replaced during a rewrite phase with explicit CALL_MALLOC_GC's, which are almost equivalent to plain CALLs, as well as CALL_MALLOC_NURSERY, which are a simplified version of "bump the nursery pointer". The point, besides simplifying the backends, is that the rewriting is done a bit more cleverly now: it will merge several consecutive NEWxxx into a single CALL_MALLOC_NURSERY, which also avoids some write barriers. CALL_MALLOC_GC is only used for mallocs that are either too large or of a non-constant size.
Armin Rigo
Close branch before merge
Branches
op_malloc_gc
Armin Rigo
fix
Branches
op_malloc_gc
Armin Rigo
- choose an arbitrary upper limit, and beyond it, generate a regular malloc_array even if the length is constant - fix boehm translation
Branches
op_malloc_gc
Armin Rigo
Fix on 64-bit
Branches
op_malloc_gc
Armin Rigo
fix
Branches
op_malloc_gc
Armin Rigo
fix
Branches
op_malloc_gc
Armin Rigo
hg merge default
Branches
op_malloc_gc
Armin Rigo
fix on 64-bit, found by test_zll_stress again
Armin Rigo
don't overwrite a previous exception with MemoryError, if any
Armin Rigo
Don't raise MemoryError explicitly; it is not needed, and confuses test_runner:test_memoryerror.
Branches
op_malloc_gc
Armin Rigo
hg merge default
Branches
op_malloc_gc
Armin Rigo
Fix if the malloc() functions provided by gc.py don't raise MemoryError themselves, as e.g. the Boehm ones.
Armin Rigo
skip this test, does not contain anything really interesting any more
Branches
op_malloc_gc
Armin Rigo
fix
Branches
op_malloc_gc
Armin Rigo
fix
Branches
op_malloc_gc
  1. Prev
  2. Next