Commits

Show all
Author Commit Message Labels Comments Date
Armin Rigo
Tweaks.
Branches
shadowstack-perf
Armin Rigo
Start to integrate the r15-for-exception branch...
Branches
shadowstack-perf
Armin Rigo
hg merge r15-for-exception
Branches
shadowstack-perf
Armin Rigo
Fix.
Branches
shadowstack-perf
Armin Rigo
Tentative tweak. Will measure.
Branches
shadowstack-perf
Armin Rigo
Improve the generated code somewhat.
Branches
shadowstack-perf
Armin Rigo
Found and fixed the bug.
Branches
shadowstack-perf
Armin Rigo
Temporarily add debugging aids. Should either be reverted or made more official...
Branches
shadowstack-perf
Armin Rigo
Starts to look good, but tests are still failing.
Branches
shadowstack-perf
Armin Rigo
hg merge default
Branches
shadowstack-perf
Armin Rigo
Merge 0cdaf4c98369, moving regalloc.py to a general tool in pypy/tool/algo/.
Armin Rigo
Move the ShadowStackRootWalker in its own file. No other change.
Armin Rigo
Revert the last 4 commits, up to c292d7b6630f. It should be done with a proper merge of the branch r15-for-exception, at some point.
Branches
shadowstack-perf
Antonio Cuni
(Alex_Gaynor, antocuni): special-case type.__eq__ so that type (and subclasses) are marked as compares_by_identity()
Antonio Cuni
merge heads
Antonio Cuni
make sure that we always pass the numeric address when converting to c_void_p
Antonio Cuni
test and fix for issue807
Antonio Cuni
fix this test
Antonio Cuni
a passing test
Antonio Cuni
fix test_callbacks by converting/unwrapping the arguments before calling the actual callback
Antonio Cuni
we don't need to pass through a callback to execute _cast_addr: all tests pass by just calling it
Antonio Cuni
another failing test
Antonio Cuni
this makes all these tests failing, because pypy does not correctly unwrap the parameters before passing them to the callbacks
Armin Rigo
Finally found a way to fix this: with asmgcc, across a malloc, the non-callee-save registers could not stay alive, even though in most cases (in the fast path) they would not be touched at all. Fixed by adopting the same solution as with shadowstack, but only for the registers that need to.
Armin Rigo
Clarify this by moving the "return" statement after the comment.
Armin Rigo
Hopefully fix 96e2837af4ec. It was already causing some failures, which are now fixed; at least: test_newgc.TestMiniMarkGC.test_hash_preservation test_newgc.TestMiniMarkGCMostCompact.test_hash_preservation and probably jit.backend.x86.test.test_z*.py.
Armin Rigo
Backed out changeset 9d796d6049d6, re-enabling 39a9093e603d (which I hope to fix in the next checkin).
Antonio Cuni
temporarily backout 39a9093e603d: it seems to trigger an asmgcc bug
Alex Gaynor
Run these tests on unicode as well as str.
Alex Gaynor
When a newstr is allocated in the JIT, we can propogate the strlen, even if it isn't a constant.
  1. Prev
  2. Next