Split pylifecycle.c out from pythonrun.c

[ncoghlan]

BPO	22869
Nosy	@warsaw, @ncoghlan, @pitrou, @tjguk, @zware, @zooba
Files	split_pythonrun.diff: WIP: split pythonrun into two separate modules split_pythonrun_v2.diff: Incorporated recent changes to pythonrun split_pythonrun_v3.diff: Adjusted patch as per comments on issue

^{Note: these values reflect the state of the issue at the time it was migrated and might not reflect the current state.}

Show more details

GitHub fields:

assignee = 'https://github.com/ncoghlan'
closed_at = <Date 2014-11-22.05:37:46.372>
created_at = <Date 2014-11-14.11:18:41.869>
labels = ['type-feature']
title = 'Split pylifecycle.c out from pythonrun.c'
updated_at = <Date 2014-11-22.08:28:30.969>
user = 'https://github.com/ncoghlan'

bugs.python.org fields:

activity = <Date 2014-11-22.08:28:30.969>
actor = 'ncoghlan'
assignee = 'ncoghlan'
closed = True
closed_date = <Date 2014-11-22.05:37:46.372>
closer = 'python-dev'
components = []
creation = <Date 2014-11-14.11:18:41.869>
creator = 'ncoghlan'
dependencies = []
files = ['37194', '37195', '37204']
hgrepos = []
issue_num = 22869
keywords = ['patch']
message_count = 21.0
messages = ['231155', '231156', '231158', '231211', '231232', '231233', '231234', '231425', '231450', '231453', '231454', '231455', '231456', '231457', '231458', '231509', '231510', '231511', '231512', '231513', '231515']
nosy_count = 9.0
nosy_names = ['barry', 'ncoghlan', 'pitrou', 'tim.golden', 'Arfrever', 'BreamoreBoy', 'python-dev', 'zach.ware', 'steve.dower']
pr_nums = []
priority = 'normal'
resolution = 'fixed'
stage = 'resolved'
status = 'closed'
superseder = None
type = 'enhancement'
url = 'https://bugs.python.org/issue22869'
versions = ['Python 3.5']

[ncoghlan]

When working on the original reference implementation for PEP-432, I found it useful to split the startup & shutdown code out from the main entry points into the eval loop (see bpo-22257 for more details).

However, that split made the draft implementation rather hard to maintain - any CPython commit that touched pythonrun.c was almost certain to cause a merge conflict.

The attached patch is a preliminary split into two separate modules, with pythonrun continuing to hold the main eval loop entry points, while the startup and shutdown code moves into pylifecycle.

This is an initial WIP patch (that nonetheless passes the test suite). I started work on it a few months ago, so I need to make sure that I've retained Victor's recent pythonrun changes.

[ncoghlan]

Incorporated pythonrun changes from 668e0bf30042, 6aaa0aab1e93 and d6fb87972dee into pylifecycle

[ncoghlan]

Known issues with the current split:

• reference count printing is duplicated (lifecycle prints it at shutdown, pythonrun at the interactive prompt)

• pythonrun references PyInspect_Flag directly without an extern declaration

• This particular oddity wasn't introduced by this patch, but it turns out PyOptimize_Flag lives in compile.c, rather than with the other global flags in pythonrun.c. It's also declared as a public data API in pydebug.h

[pitrou]

+1 on the principle.

reference count printing is duplicated (lifecycle prints it at
shutdown, pythonrun at the interactive prompt)

You could make it an API in object.c.

• pythonrun references PyInspect_Flag directly without an extern
  declaration

• This particular oddity wasn't introduced by this patch, but it
  turns out PyOptimize_Flag lives in compile.c, rather than with the
  other global flags in pythonrun.c. It's also declared as a public
  data API in pydebug.h

Should be easy to fix as well :-)

[ncoghlan]

It turns out *all* the global config variations are in pydebug.h (and they're basically the only thing in there, aside from the environment variable access macro). That seems a little weird to me (perhaps historical due to the early global config variables being debugging related?), but for now I've just added a couple of comments cross referencing pydebug.h from pylifecycle.c and vice-versa.

The latest version of the patch:

• moves the reference count & block allocation printing to object.[ch] as Antoine suggested (Macro = _PY_DEBUG_PRINT_TOTAL_REFS(); Symbol = _PyDebug_PrintTotalRefs())

• moves Py_OptimizeFlag to pylifecycle.c together with the other global configuration flags (and tweaks the order of definitions to exactly match the order of declarations in pydebug.h)

Since the extern declaration is actually there in pydebug.h, the reference to Py_InspectFlag from pythonrun.c is actually fine for now (although removing that direct reference will likely end up being part of the PEP-432 implementation).

[ncoghlan]

I think this is ready to go now, unless anyone spots any major flaws I missed.

[ncoghlan]

Oops, "global config variations" above should have been "global config variable declarations". I guess my brain decided that was too much typing and jammed the last two words together :)

[python-dev]

New changeset b9775a92c1d0 by Nick Coghlan in branch 'default':
Issue bpo-22869: Split pythonrun into two modules
https://hg.python.org/cpython/rev/b9775a92c1d0

[BreamoreBoy]

I'm not absolutely sure but I think this has broken the Windows builds with the following being typical of numerous errors.

2>main.obj : error LNK2019: unresolved external symbol _Py_Finalize referenced in function _Py_Main
2>main.obj : error LNK2019: unresolved external symbol _Py_Initialize referenced in function _Py_Main
2>main.obj : error LNK2019: unresolved external symbol _Py_SetProgramName referenced in function _Py_Main
2>main.obj : error LNK2019: unresolved external symbol _Py_FdIsInteractive referenced in function _Py_Main

[pitrou]

Thanks. The Windows build files (Visual Studio project files) typically have to be updated when a new C file is added to the source tree. Probably one of our Windows experts can help with that :-)

[python-dev]

New changeset 31fd106bb68a by Steve Dower in branch 'default':
Issue bpo-22869: Add pylifecycle.c/.h files to pythoncore project.
https://hg.python.org/cpython/rev/31fd106bb68a