Commits

Show all
Author Commit Message Labels Comments Date
Daniel Roberts
Added support code for array broadcasting, along with a small (unfinished) test.
Branches
micronumpy-resync
Daniel Roberts
Problem with .svn directory should be fixed now.
Branches
micronumpy-resync
Daniel Roberts
Taking the easy way out and just replacing this.
Branches
micronumpy-resync
Daniel Roberts
Going to rebase micronumpy changes into this then replace the original micronumpy branch.
Branches
micronumpy-resync
Alex Gaynor
Don't use relative imports.
Armin Rigo
Fix the test.
Armin Rigo
Improve the error message.
Armin Rigo
Merge branch/jit-generator. It re-adds some supports for generators, and it also avoids the issue of call_assemblers whose target code is not yet compiled, by providing in this case a dummy piece of machine code that can later be patched.
Maciej Fijalkowski
Merge rsocket-improvements branch. This branch intends to provide support for AF_PACKET type addresses (wherever supported)
Hakan Ardo
cpython compability for array subclasses overriding __new__ but not __init__
Hakan Ardo
Merged jit-bounds, which splits jit/metainterp/optimizeopt into separate optimizations steps and adds one additional step keeping bounds on integers and removes guards when they are not needed.
Amaury Forgeot d'Arc
Use opaque pointers to SSL structures (instead of rffi.VOIDP which becomes char*) This remove almost all compilation warnings in interp_ssl.
Amaury Forgeot d'Arc
Turn rwin32.HANDLE into an Opaque type that emits "HANDLE" in C. this removes a lot of compilation warnings on Windows.
Amaury Forgeot d'Arc
Fix translation on Windows
Armin Rigo
The portal_calldescr should not be attached to the CPU, but to the JitDriverSD. (almost not used any more but still)
Armin Rigo
In rffi_platform, return an integer of some RPython type that fits it: either int, r_uint, r_longlong or r_ulonglong.
Amaury Forgeot d'Arc
Fix test_rffi on Windows, which aborts() in os.write when an invalid fd is used. Use a non-writable file instead.
Armin Rigo
These constants also need a r_uint() now.
Armin Rigo
Temporary fix. It seems that on 32-bit linux, compiling with -fPIC gives assembler that asmgcc is not happy about.
Armin Rigo
Allow constant-folding to occur early for r_uint(long-value). It must be done before annotation, because now annotation explodes when seeing the long value.
Armin Rigo
Revert r76928. It breaks tons of tests and there is no reason for this checkin alone.
Armin Rigo
Move the constant 0xffffffff out of the RPython functions.
Armin Rigo
Merge branch/asmgcc-64. I *think* that after a couple of minor changes it does not break asmgcc on 32 bits, and it seems to work on 64 bits too.
Armin Rigo
More fixes for r76920.
Armin Rigo
Fix for r76920. Oups, breaks tests a bit all over the place.
Maciej Fijalkowski
on posix, define sockaddr_ll
Armin Rigo
Fix for the test.
Maciej Fijalkowski
A test for untested piece of code (initializing raw addresses). Also kill a sanity check that was in wrong place at the very least. Reintroduce if needed, but with tests please
Armin Rigo
Fix.
Armin Rigo
Change annotation to say that int(r_uint(x)) is invalid: you have to call intmask(). Fix one place that use it in rlib. Also fix an issue where unexpected constants of type "long", but actually that fit an "int", show up during translation (e.g. constants from .pyc files generated under 32-bit and loaded under 64-bit).
  1. Prev
  2. Next