Commits

Show all
Author Commit Message Labels Comments Date
Hakan Ardo
leftover debuging
Branches
jit-usable_retrace
Hakan Ardo
fixed merge error, only insert GUARD_NO_OVERFLOW when needed
Branches
jit-usable_retrace
Hakan Ardo
Test GUARD_OVERFLOW aswell. Currently it will not be moved out of the loop and end up in the short preamble
Branches
jit-usable_retrace
Hakan Ardo
Proper handling of resuming from a GUARD_NO_OVERFLOW inlined from the short preamble. This fix should probably go into trunk aswell.
Branches
jit-usable_retrace
Hakan Ardo
resolved merge
Branches
jit-usable_retrace
Hakan Ardo
hg merge default
Branches
jit-usable_retrace
Amaury Forgeot d'Arc
Add support for PyUnicode_FromUnicode(NULL, size), which allocates a (temporarily) mutable unicode string. Also implement PyUnicode_Resize. See comments in stringobject.py for a complete explanation
Amaury Forgeot d'Arc
(lambacck) Implement PyExceptionInstance_Class, simplified because 2.7 strongly discourages old-style exceptions.
Amaury Forgeot d'Arc
Appdirect tests really run too fast. Slow down a little
Amaury Forgeot d'Arc
MixedModule tests make no sense with appdirect.
Amaury Forgeot d'Arc
Don't even try to load a dynamic module when cpyext is not enabled. Also kill dead code.
Amaury Forgeot d'Arc
cpyext: Fix call to the tp_getattro type slot.
Amaury Forgeot d'Arc
Ensure that a PyTypeObject is "realized" only once, either directly with PyType_Ready or recursively when it is referenced in tp_base.
Armin Rigo
Rewrite in the same way int_floordiv.
Armin Rigo
Remove the scary sequence of operations for int_mod, and replace it with a call to a helper. The helper contains an 'if:else:' path, but it should generate far better code, both normally and when JITted.
Armin Rigo
By default, disable 'withsmalllong'. It seems to not give the expected speed-ups right now (will confirm it by looking at the performance in the following days).
Carl Friedrich Bolz
remove some oopspecs in rdict to make the JIT trace the hash functions in dicts. this makes it necessary to hide some interior field manipulation in a helper function.
Carl Friedrich Bolz
- setdefault only does one lookup now - update doesn't recompute hashes all the time
tav
Added myself to the authors list.
tav
Added cpyext/*/*.o to the .gitignore.
Armin Rigo
Fix for the failing test in test_typedef. The cause was ultimately that the global _subclass_cache was populated with various 'config's, and now there was too many of them for the test to pass.
Carl Friedrich Bolz
refactor rdict a bit: - make ll_dict_lookup return the index plus a flag bit set if the entry is unused. this removes a second access to the entry in the caller - make setdefault hash only once(!)
Amaury Forgeot d'Arc
cpyext: finally managed to support metatypes
Amaury Forgeot d'Arc
cpyext: Add support for user-allocated PyObjects, when the initialization only calls _Py_NewReference(obj).
Amaury Forgeot d'Arc
Implement imp.load_dynamic(), which loads a cpyext dll.
Amaury Forgeot d'Arc
Translation fix
Amaury Forgeot d'Arc
Implement PyImport_ReloadModule
Amaury Forgeot d'Arc
_PyLong_FromByteArray() first argument is an "unsigned char*"
Amaury Forgeot d'Arc
Implement thread.interrupt_main(), which is absent from CPython tests :-( Add PyErr_SetInterrupt() in cpyext.
tav
merge heads
  1. Prev
  2. Next