Commits

Show all
Author Commit Message Labels Comments Date
Simon Cross
(fijal, hodgestar) Add test_pypy_c for generators.
Branches
inline-simple-generators
Simon Cross
(antocuni, hodgestar) Rename unroll argument in test to make test more understandable.
Branches
inline-simple-generators
Maciej Fijalkowski
(fijal, hodgestar) unroll one iteration of generators
Branches
inline-simple-generators
Simon Cross
(fijal, hodgestar) Unroll only one iteration of the loop.
Branches
inline-simple-generators
Maciej Fijalkowski
(fijal, hodgestar) make the test pass
Branches
inline-simple-generators
Simon Cross
(fijal, hodgestar) Start adding should_unroll_one_iteration.
Branches
inline-simple-generators
Armin Rigo
Issue #781: improve the error message.
Armin Rigo
Fix issue 793 (thanks mistuhiko). Generally allow visit_sequence() to be called on None.
Armin Rigo
Test and fix for issue #804 (thanks albert). Regenerated ast.py.
Armin Rigo
Change the interface of deldictvalue() to take an unwrapped attribute name, like getdictvalue() and setdictvalue(). Just for unification purposes.
Armin Rigo
Test and fix. Thanks Trundle for the test. The fix is done in error.py instead, by calling exception_is_valid_class_w() after calling exception_getclass(), as was already done in one place.
Armin Rigo
Fix for f523aecab18f, and rewrite to move the loop in a subfunction that is not called at all if step==1.
Armin Rigo
Tests prompted by issue #801 (thanks justinpeel).
Armin Rigo
Fix for issue #806.
Armin Rigo
Remove a needless try:finally: that goes in the way of printing RPython tracebacks.
Armin Rigo
Fix for the previous bug: declare that the list should be really 'old_objects_pointing_to_young', i.e. only contain old objects. This is a bit costly to ensure in the write barrier, so between collections, we let young arrays be added to it "by mistake" and then clean it up at the start of the next minor collection.
Armin Rigo
There is another case that is similarly buggy (it traces too much)...
Armin Rigo
Moving code around. The goal is to reduce the cost of space.isinstance(), probably by a large factor: according to valgrind, for every second we spend running W_TypeObject.issubtype(), we spend almost 4 extra seconds in the rest of space.isinstance(), because of the levels and levels of indirections.
Armin Rigo
merge heads
Armin Rigo
Support writing inline code in the write barrier for large arrays, for the simple case where calling it would just set a flag just before the object.
Armin Rigo
Add some encodings for the 8-bit instruction OR.
Armin Rigo
Add the encoding for the instruction BTS.
Armin Rigo
merge heads
Armin Rigo
Oooops. Found and fixed a subtle bug: young arrays that use card marking but die young would still be scanned for objects. That's wrooong as it means that a lot more objects stay alive.
Armin Rigo
Backout 4543677d758a and abc993e8d795, removing PYPY_GC_LOSTCARD again. Will try another approach: inlining the write barrier logic for GCFLAG_HAS_CARDS in the jit backend code.
Alex Gaynor
merged upstream
Alex Gaynor
Change a bunch of sys.platform == 'linux2' to be sys.platform.startswith('linux') to be ready for linux3.
Armin Rigo
Tentatively let StackOverflow and MemoryError exit the JIT tracing and be propagated outside.
Armin Rigo
Tentative: set acceptable_as_base_class to False by default on TypeDefs that don't define a custom __new__. CPython tests appear to pass.
Maciej Fijalkowski
fix and don"t compare pointers by is
  1. Prev
  2. Next