1. Olemis Lang
  2. pypy

Commits

Show all
Author Commit Message Date Builds
Jean-Paul Calderone
implement os.openpty
Benjamin Peterson
simplify by using space.finditem
Benjamin Peterson
reload loaded modules from sys.modules during import like cpython does
Armin Rigo
Fix test_boehm: we cannot ignore keepalives in case of Boehm, because we might be unlucky and at the C level the variable gets overridden with another variable. No clue why test_boehm started to fail now, though.
Armin Rigo
A missing keepalive_until_here() to keep the list alive. Unfortunately the test keeps failing on wyvern, even with this change.
Armin Rigo
Print errors to stdout instead of sending them to SystemExit to avoid a broken order of output (they are printed to stderr in that case).
Armin Rigo
Merge branch/morearraycopy: use rgc.ll_arraycopy more or less systematically.
Armin Rigo
More precise comment.
Armin Rigo
Test and fix for the case of an object defining __getslice__ but not __len__, and slicing with [x:y] with x >= 0 but y < 0.
Armin Rigo
For now, just killing a lot of "if we_are_jitted" is enough. These changes should be consistent with the branch/jit-trace-hook.
Armin Rigo
Move the test to module/pypyjit, where it should be taken on by the nightly runners.
Armin Rigo
Make the error message clearer for the case where we make a typo in --pypy.
Armin Rigo
Cannot print while sys.settracing(). If run with py.test without the -s option, it will trace the py lib output capturing logic.
Carl Friedrich Bolz
allow the test I wrote yesterday to be properly skipped
Carl Friedrich Bolz
Merge the change-celldict2 branch: Strip the cell dict down to something radically simple. Now it is really just a dict mapping names to cells. For the non-JIT case this makes global-lookups non-optimized (but the caching seems to not have helped much anyway). For the JIT-case the effect is nearly as good as the old code, and even better, if the same key is looked up many times in the same module (which is quite common for librarie…
Amaury Forgeot d'Arc
Fix a keepalive problem in ctypes: When setting a struct member, the structure did not track keepalives if the value needs to be converted.
Carl Friedrich Bolz
a test that fails when running apptests with a pypy-c-jit
Carl Friedrich Bolz
Tiny optimization: when forcing a virtual, don't generate code for setting fields to NULL, since malloc returns zero-filled objects. This works fine with the x86 backend, but needs a fix in the llgraph backend to actually zero the memory (at least for structs, arrays have always been zeroed).
Carl Friedrich Bolz
this function is unused
Armin Rigo
Merge the branch/jit-trace, trying to be safe JIT-wise against using the profile or tracing hooks.
Armin Rigo
Another try to fix this test for Windows.
Armin Rigo
Add a gc.collect() as a tentative to avoid random occasional failure of that test.
Armin Rigo
Constant-fold some paths, e.g. "if we_are_jitted:" paths, when producing jitcodes.
Amaury Forgeot d'Arc
Fix test_runpy on Windows ('c:\temp' needs to be escaped correctly)
Armin Rigo
Merge the virtual-forcing branch, adding support for virtualrefs. Allows a big revert of the messy code in pypy/interpreter/ about chains of frames. I am merging this even though it's still giving a slow-down, notably on "pypy-c-jit richards", because it's really simplifying stuff and I don't want it to remain on a branch forever.
Armin Rigo
Improve the test.
Armin Rigo
A failing test that shows a missing keepalive in our ctypes.
Armin Rigo
Increase sleep()ing time to be greater than 1 second. Exactly 1 second might lead to corner cases. The test failed once because it started at time x+0.003, and the actual time it slept was lower than 0.997 seconds.
Samuele Pedroni
bump the copyright range, I'm sure will checkin code in 2010 :)
Amaury Forgeot d'Arc
Finally found where sys.path is replaced instead of modified in-place. Fixes reload(sys)
  1. Prev
  2. Next