1. Andrews Medina
  2. pypy-br

Commits

Show all
Author Commit Message Date Builds
Ronan Lamy
get rid of the NonNativeFoo item types (a few translation failures)
Branches
longdouble2
Ronan Lamy
Factor out other-endian dtype creation
Branches
longdouble2
Ronan Lamy
fix translation oddity with .read_bool()
Branches
longdouble2
Ronan Lamy
start extracting low-level type metadata into pypy.module.micronumpy.typespec
Branches
longdouble2
Ronan Lamy
Store the unwrappped BoxType on the dtype.
Branches
longdouble2
Ronan Lamy
Give ComplexFloating item types a reference to their component type, instead of their component's box type.
Branches
longdouble2
Ronan Lamy
kill ComplexFloating.composite()
Branches
longdouble2
Ronan Lamy
dispatch set_imag on the dtype
Branches
longdouble2
Ronan Lamy
Don't use convert_{real|imag}_to indirection in W_ComplexFloatingBox
Branches
longdouble2
Ronan Lamy
merge implementations of W_Float96Box and W_Float128Box, etc.
Branches
longdouble2
Ronan Lamy
merge dtype declarations for longdouble
Branches
longdouble2
Ronan Lamy
merge implementations of Float96 and Float128
Branches
longdouble2
Armin Rigo
Fix an obscure issue where destructed SHADOWSTACKREFs could still be occasionally traced, leading to crashes.
Armin Rigo
Run -A tests even if they require translation options (specified without the "translation." prefix, e.g. in pypy/module/_continuation/support.py: "continuation=True")
Antonio Cuni
merge the refactor-call_release_gil branch. This fixes a nasty bug which occoured when JITting the call to a cffi function which calls a callback which causes the failure of guard_not_forced: in that case, during blackholing we got the wrong result from call_release_gil, because it was not passed to fail_args. The two tests which demonstrates the bug are - rpython/jit/metainterp/test/test_fficall.py::test_guard_not_forced_fails - pypy/module/pypyjit/test_pypy_c/test__ffi.py::test_cffi_call_guard_not_forced_fails
Antonio Cuni
close about-to-be-merged branch
Antonio Cuni
add a (no longer failing) test which is the whole point of the branch
Antonio Cuni
add a workaround to make test_fficall::test_guard_not_forced_fail passing on the x86 backend; see the comment for details
Antonio Cuni
move the logic for jit_ffi_call back to jit_libffi.py, and reuse it from the tests by (ugly) monkey-patching. Also, add comments to explain in more details why we need this strange dance
Antonio Cuni
implement jit_ffi_save_result_float
Antonio Cuni
we cannot use an oopspec-ed function to mark the jit_ffi_save_result, because the call gets removed by rtyper's backend optimizations before the codewriter has a chance to see it. Instead, we generate a brand new flowgraph op, 'jit_ffi_save_result'
Antonio Cuni
fix test_guard_not_forced_fails by refactoring the way libffi_call works. The problem is that we need to make sure that the box containing the result of call_release_gil is considered alive by the codewriter when we write it to the exchange_buffer. The solution is to create a new separate operation libffi_save_result whose job is to write the result in the exchange_buffer: during normal execution this is a no-op because the buffer is already filled by libffi, but during jitting the behavior is to actually write into the buffer. The result is that now the jitcode looks like this: %i0 = libffi_call_int(...) -live- libffi_save_result_int(...,…
Antonio Cuni
hg merge default
Brian Kearns
another arm change after 55a5e932519e
Brian Kearns
this backend_name too
Brian Kearns
also update these arm backend_names after 55a5e932519e
Brian Kearns
unskip the arm tests on arm after 55a5e932519e
Maciej Fijalkowski
fix the tests
Maciej Fijalkowski
fix race conditions in zipfile tests
Maciej Fijalkowski
this should go at best to a branch
  1. Prev
  2. Next