The main change is the introduction of "Python Configuration" and
"Isolated Configuration" default configuration which are well better
defined. Replace all macros with functions.
Changes:
* PyPreConfig: no longer uses dynamically memory allocation,
allocator type becomes an enum, add parse_argv and configure_locale
field.
* PyConfig: Remove program field, rename module_search_path_env to
pythonpath_env
* New functions:
* PyPreConfig_InitIsolatedConfig()
* PyPreConfig_InitPythonConfig()
* PyConfig_InitIsolatedConfig()
* PyConfig_InitPythonConfig()
* parse_argv default becomes 0
* Replace Py_INIT_xxx() macros with functions
* Document default values
* Remove PyConfig.dll_path
* Explain how PyConfig preinitializes Python and the relationship
between command line arguments and the pre-initialization.
* Rename use_module_search_paths to module_search_paths_set
* Rename PyInitError_Err() to PyInitError_Error()
* "Pre-initialization" => "Preinitialization"
* Reject preconfig=NULL and config=NULL
* Replace "wchar_t**" with "wchar_t * const *", same for "char**".
* Thomas Wouters is now the BDFL-delegate
* Elaborate the Rationale
* New "Path configuration" section
* New "Python Issues" section
* Add PyWideStringList_Insert() function
* Add PyConfig.configure_c_stdio option.
* Add PyConfig.parse_argv field
* Rename bytes/wide functions. Add "Bytes" to functions taking char*,
remove "Wide" from functions taking wchar_t*. Rename
PyConfig_DecodeLocale() to PyConfig_SetBytesString().
* PyConfig: Rename '_frozen' to 'pathconfig_warnings'
* Re-introduce PyConfig._init_main
* Complete tables of global configuration variables; annotate
variable values which are negated.
* Add default values of all PyPreConfig and PyConfig fields
* Elaborate the Rationale
* Add a few more issues
* Add PyPreConfig.legacy_windows_fs_encoding rules
* Add "Isolate Python" section
* PyConfig_SetString() and PyConfig_SetBytesString() now requires the
configuration as the first argument.
* Rename Py_UnixMain() to Py_BytesMain()
This PR contains mostly minor wording tweaks plus a paragraph explicitly allowing class objects as implementations of protocols, previously there were questions whether it is actually allowed, see python/mypy#4536.
The previous wording could be taken as suggesting that frontends should opt
out of their PEP 517 processing entirely if build-backend was not defined,
whereas it's actually fine to just use the setuptools provided backend that
implements the legacy build semantics.
See bpo-36798.
This update specifies the interplay between f-strings and
the assignment operator. Specifically, it attempts to
clarify how to use assignment operators inside of f-strings.
There is too little time left before the 3.8 release and I don't want Antoine to feel rushed to pronounce. The implementation is mostly complete, so we could land it for 3.8. However, the experience in 3.8 wouldn't be nearly as good, in part because we're still sharing the GIL. Plus, we need more confidence that extension modules won't be severely impacted by the wide-spread use of subinterpreters. Waiting for 3.9 will allow us to address both issues, as well improve performance (especially startup).
Victor Stinner
till
Jeroen Demeyer
T. Wouters
Ethan Smith
Barry Warsaw
Nick Coghlan
larryhastings
Ivan Levkivskyi
Donald Stufft
Michael J. Sullivan
Hugo
Dominik Gabi
Petr Viktorin
Antoine Pitrou
Christian Heimes
Steve Dower
Mark Shannon
Łukasz Langa
Logan Jones
Guido van Rossum
Matthias Bussonnier
Brett Cannon
Eric Snow
Mariatta