Commits

Show all
Author Commit Message Labels Comments Date
Armin Rigo
Test fixes.
Branches
effectinfo-cleanup
Armin Rigo
Comments.
Branches
effectinfo-cleanup
Armin Rigo
Rename "releases_gil" to "random_effects_on_gcobjs", which is closer to what we use it for in the jit.
Branches
effectinfo-cleanup
Armin Rigo
Clean-up effectinfo: now all calldescrs need to have one, but its extraeffects can be EF_RANDOM_EFFECTS, in replacement of effectinfo=None. This is a clean-up because it allows us to build an effectinfo with both EF_RANDOM_EFFECTS and OS_LIBFFI_CALL.
Branches
effectinfo-cleanup
Alex Gaynor
Fix for the gdb hooks when the tag of a something you try to print is None, not sure what exactly this means, but it happened to me.
Antonio Cuni
merge heads
Antonio Cuni
make sure to pretty print *all* rpython lists, not just the one named pypy_list0
Antonio Cuni
make the pretty prints a bit shorter
Antonio Cuni
a bit of refactoring: do pretty print of *pointers* to strings and lists, so we can still inspect the underyling struct when we do p *val
Antonio Cuni
add a pretty printer for rpython strings
Armin Rigo
After some more thought (sorry), re-enable the backed-out change fc9ced8e380f and mark the test I added with an XXX. I would be fine if this is fixed by forbidding it in the codewriter, too.
Armin Rigo
Sorry, I understood now the issue in test_unerased_pointers_in_short_preamble (it fails in test_loop_unroll). This test fails too, for the same deeper reason: the short preamble cannot verify (with guard_class) that the object is of exactly the right type.
Armin Rigo
Add a test similar to test_unerased_pointers_in_short_preamble, but using only regular pointers. Note that neither test appears to be failing right now, unlike the comment in 2ca138be14a7...
Armin Rigo
Backout fc9ced8e380f: this is wrong, because the problem is not really about opaque pointers. We can have the same problem with regular pointers. More precisely, we can have a (non-opaque) pointer to a base class, and then read a field that belongs to a subclass. This can occur without an explicit guard_class, but by doing the check indirectly. In this case it's wrong too to move the getfield_gc before we did the indirect check...
Philip Jenvey
whitespace
Antonio Cuni
search typeids.txt also in the parent directory
Hakan Ardo
Make the jit optimizer aware of opaque pointers and makes sure they dont end up in the short preamble. Fixes the bug tested in 2ca138be14a7
Hakan Ardo
failing test. unerased pointers make any dereference in the short preamble unsafe
Alex Gaynor
Use intbounds information to decide if a {get,set}arrayitem_gc can alias a previous setfield which became lazy.
Alex Gaynor
a bit of cleanup to the release notes.
Alex Gaynor
set a good example for the children.
Alex Gaynor
fix these tests, the resop int_mod returns a negative value for mixed operand signs.
Antonio Cuni
write a test for the rpy_type command
Antonio Cuni
refactor the code and start to write tests
Antonio Cuni
automatically find typeids.txt based on the exe name, and cache the result of load_typeids
Antonio Cuni
fix docstring
Antonio Cuni
signature
Antonio Cuni
reflow the paragraph
Antonio Cuni
improve the release announcement
Armin Rigo
merge heads
  1. Prev
  2. Next