Commits

Show all
Author Commit Message Labels Comments Date
Armin Rigo
Comments.
Branches
minimark-no-mass-free
Armin Rigo
First step, to see that it works: change minimark.py to adhere to a new interface, documented in minimarkpage.py but not implemented yet.
Branches
minimark-no-mass-free
Armin Rigo
Make a branch in which to refactor ArenaCollection.mass_free(), moving minimarkpage.py even further away from its original general alloc-and-free interface and into the land of only-usable-from-a-GC.
Branches
minimark-no-mass-free
Carl Friedrich Bolz
one of the more obscure commits: change the thresholds slightly to be prime numbers. the reason for this is that if there are nested loops that have combined iteration times of exactly the threshold you get bad traces. probably not important in practice, but seemed fun to do.
Branches
default
Carl Friedrich Bolz
merge improve-heap-caching-tracing: - a much better heap cache during tracing, this makes the traces that are produced before optimization much shorter, thus improving warmup time, reducing GC pressure, etc. - change the maximum trace length to half its original value. Now that the traces are much shorted due to the better optimization, this seems to give much better results for translation.
Branches
default
Carl Friedrich Bolz
close to-be-merged branch
Branches
improve-heap-caching-tracing
Armin Rigo
merge heads
Branches
default
Armin Rigo
Translation fix
Branches
default
Armin Rigo
More informative message
Branches
default
Antonio Cuni
fix for ootype, llstrings are not supported as input arguments
Branches
default
Carl Friedrich Bolz
change the trace length limit to half its previous value. this seems to be better now with the better heap caching, will do a benchmark run to confirm.
Branches
improve-heap-caching-tracing
Armin Rigo
permute() should also ignore non-initialized continulets, just like switch().
Branches
default
Armin Rigo
merge heads
Branches
default
Armin Rigo
Rename start_state -> global_state
Branches
default
Carl Friedrich Bolz
remove outdated comment
Branches
default
Armin Rigo
Oups! The call to the C functions stacklet_new() and stacklet_switch() did not include save-and-restore-the-roots code. Argh. Fix it by saying, reasonably, that any external function with 'random_effects_on_gcobjs' needs to have this code.
Branches
default
Armin Rigo
Debugging code. Needs to be enabled explicitly for now.
Branches
default
Armin Rigo
Fix.
Branches
default
Armin Rigo
merge heads
Branches
default
Armin Rigo
Fold the 4 cases into 1 for greenlets and coroutines calling continulet.switch(), by changing the rule: passing a non-initialized continulet as 'self' or 'to' does not raise, but instead is ignored.
Branches
default
Justin Peel
make all files use space.isinstance_w instead of space.is_true(space.isinstance(...)) in objspace/std/
Branches
default
Justin Peel
move the new isinstance_w up to objspace/std/objspace.py so that it actually is used.
Branches
default
Justin Peel
make isinstance_w not wrap and unwrap a bool. (flow space is happy too).
Branches
default
Amaury Forgeot d'Arc
"Merge" Issue #12012 from CPython 2.7: ssl.PROTOCOL_SSLv2 becomes optional
Branches
default
Justin Peel
Backout 12b651d04110 for now. Fun with the spaces.
Branches
default
Armin Rigo
merge heads
Branches
default
Armin Rigo
Bah. test_stackless.py fails completely nowadays. I'm not sure I understand why e.g. it doesn't show up on buildbot, but the previous version was too naive in a few places. Fix fix fix...
Branches
default
Justin Peel
make isinstance_w not wrap a bool and unwrap it again
Branches
default
Armin Rigo
Fix the test by a hack. Also as discussed on IRC the test is really about frozen objects; so far it does not work as is on non-frozen objects.
Branches
default
Maciej Fijalkowski
a failing test
Branches
default
  1. Prev
  2. Next