bpo-40521: Make bytes singletons per interpreter #21074
Conversation
Each interpreter now has its own empty bytes string and single byte character singletons. Replace STRINGLIB_EMPTY macro with STRINGLIB_GET_EMPTY() macro.
Each interpreter now has its own empty bytes string and single byte character singletons. Replace STRINGLIB_EMPTY macro with STRINGLIB_GET_EMPTY() macro.
Each interpreter now has its own empty bytes string and single byte character singletons. Replace STRINGLIB_EMPTY macro with STRINGLIB_GET_EMPTY() macro.
|
Hi @vstinner, this change introduced a segfault for us since it uses the private It seems that, with this change, PyBytes_FromStringAndSize is no longer safe to call when the GIL is not held. Is that by design? If it's by design, perhaps it would be useful to either use PyInterpreterState_Get or add in an assertion in order to catch this incorrect usage? |
This is due to python/cpython#21074, which causes a segfault in the latest Python master when creating a bytes object of size 0. readlines() has been reimplemented to use a C++ vector in order to prevent constantly re-locking and unlocking the GIL for every line.
Please open a new issue at https://bugs.python.org/ with all details. But I expect that you misuse the Python C API:
You must always hold the GIL to call any function of the Python C API. There are very few exceptions at this rule, and usually these exceptions mention that they can be called without holding the GIL, in their documentation. It's not a new requirement of Python 3.10, it's just that previous magically, it worked in some cases when the GIL was not held. |
Each interpreter now has its own empty bytes string and single byte
character singletons.
Replace STRINGLIB_EMPTY macro with STRINGLIB_GET_EMPTY() macro.
https://bugs.python.org/issue40521