Commits

Author Commit Message Labels Comments Date
Armin Rigo
Rename ffi.dlopen() into ffi.dlopen_abi_unchecked(), and make the functionality stick as close as possible to the C-level dlopen().
Maciej Fijalkowski
Since we have users, make sure we actually emit UserWarning, instead of just crashing.
Armin Rigo
Give some compatibility explanation when passing a Python file object to a FILE argument.
Armin Rigo
Kill the historical non-decorator version of ffi.callback(), as a motivation for doing "the right thing" (see new paragraph in docs).
Comments 2
Armin Rigo
Add warnings about keeping alive the library object.
Armin Rigo
Document the removal of FILE, add demos.
Armin Rigo
Kill the logic to support FILE. Found out that it seems reasonable enough to use the standard cffi way.
Armin Rigo
Cancel again these changes: we can't pass 0 or None to mean NULL. Just use NULL explicitly. (After discussion on IRC, notably with Amaury -- thanks)
Armin Rigo
merge heads
Armin Rigo
Document how to work around these limitations.
Armin Rigo
Refactor get_c_name() in model.py to use the same logic as with the backend.
Armin Rigo
Fix the test for 64-bit: alignment issues
Tom Nixon
Generate correct type name for pointers to arrays. Previously, get_c_name("a") on the type "int (*)[5]" would return "int *[5]". This caused "incompatible type" errors when verifying structs containing pointers to arrays. A partial workaround for this bug was originally made in vengine_*; this has been removed.
Armin Rigo
Test for the 0-to-ptr convertion.
Armin Rigo
Change the hack: instead of passing None for NULL pointers, we pass 0. This is more compatible with C. Also accept 0 at other places that expect a pointer.
Armin Rigo
Python 3 fixes.
Armin Rigo
More careful, even if it looks more brittle in theory
Armin Rigo
Merged in stefanor/cffi (pull request #8) Strip _d from cpython 2.x debug build extension names
Stefano Rivera
That could have been simpler
Stefano Rivera
Kill the _d added in Python 2 debug builds
Maciej Fijalkowski
Merged in alex_gaynor/cffi/library-module (pull request #9) Make the library be a ModuleType subclass so that on PyPy it is optimized for the fact that the attribtues are generally never re-assigned.
Alex Gaynor
explanatory comment
Branches
library-module
Alex Gaynor
Make the library be a ModuleType subclass so that on PyPy it is optimized for the fact that the attribtues are generally never re-assigned.
Branches
library-module
Armin Rigo
update
Armin Rigo
typo
Armin Rigo
(lazka) improve the test, and fix a typo (bad arigo)
Armin Rigo
(lazka, arigo) Change dlopen() to accept either a full path or a library name.
Armin Rigo
Skip running this file on top of PyPy.
Armin Rigo
Mark the release 0.5.
Branches
release-0.5
Armin Rigo
Small documentation fixes.
  1. Prev
  2. Next