homepage

http://buildbot.python.org/all/#/builders/80/builds/244
 (Link target) -> 
 python3_d.def : error LNK2001: unresolved external symbol PyImport_GetModule [D:\buildarea3円.x.ware-win81-release.refleak\build\PCbuild\python3dll.vcxproj]
 python3_d.def : error LNK2001: unresolved external symbol Py_UTF8Mode [D:\buildarea3円.x.ware-win81-release.refleak\build\PCbuild\python3dll.vcxproj]
 D:\buildarea3円.x.ware-win81-release.refleak\build\PCbuild\amd64\python3_d.lib : fatal error LNK1120: 2 unresolved externals [D:\buildarea3円.x.ware-win81-release.refleak\build\PCbuild\python3dll.vcxproj]
Maybe it's related to this recent change:
commit 4e29f566e8821c09d8274eadcdd355e8b1284b8b
Author: Serhiy Storchaka <storchaka@gmail.com>
Date: Tue May 22 20:59:42 2018 +0300
 Add missed details of the C API introduced in 3.7. (GH-7047)
 
 * Set the limited API version for PyImport_GetModule and PyOS_*Fork
 functions.
 * Add PyImport_GetModule and Py_UTF8Mode in PC/python3.def.
 * Add several functions in Doc/data/refcounts.dat.
The two missing symbols are defined by:
#if !defined(Py_LIMITED_API) || Py_LIMITED_API+0 >= 0x03070000
PyAPI_FUNC(PyObject *) PyImport_GetModule(PyObject *name);
#endif
and
#if !defined(Py_LIMITED_API) || Py_LIMITED_API+0 >= 0x03070000
PyAPI_DATA(int) Py_UTF8Mode;
#endif
They are not part of the stable ABI.

My bad, I didn't test that change on Windows myself, and CI seems was broken at that moment.
But I have no ideas why the build is failed. These names are part of stable ABI since version 3.7. I don't see any difference between PyImport_GetModule and say PyInterpreterState_GetID.

It succeeds after removing directories amd64 and obj in PCbuild. Seems they are left from old builds and the clean up script did nor remove them.
This may be a cause of other weird bugs on Windows.

I guess that means there is an invalid date comparison going on to detect when the file has changed. This file doesn't change often, so it's easy to miss :)
All the critical builds should forcibly rebuild it, so this isn't urgent, but I'll take a look next time I get a chance to work on the build files.

If "PCbuild\build.bat -p x64" builds Python, "PCbuild\clean.bat -p x64" removes all files in PCbuild\amd64 but python3stub.exp and python3stub.lib.

> All the critical builds should forcibly rebuild it, so this isn't urgent, but I'll take a look next time I get a chance to work on the build files.
It's more than a *bit* urgent since it is causing one of the 3.7 stable buildbots to fail to build and not run tests and we normally would not allow a release with stable buildbot failures.

(for example, http://buildbot.python.org/all/#/builders/133/builds/251 )

Okay, I'll take a look now.

Steve: just a minute ago, zware out the win8.1 bot's repo and restarted a 3.7 build.

er, "cleaned out"

That Windows8.1 buildbot sure keeps busy with all those builders assigned to it :) Those refleaks builds take a *long* time!
http://buildbot.python.org/all/#/workers/11
But I see at least one builder has made it through successfully and no longer fails in this manner:
http://buildbot.python.org/all/#/builders/12/builds/892
So I think we can close this now. Thanks, everyone.

I think the underlying issue is not fixed still. Adding new symbols in python3.def will break this buildbot (as well as users worktrees) again.

Considering we didn't change anything in the sources, I wouldn't expect it to be fixed.
It doesn't have to block 3.7 if Ned's okay with it, but I'd like to keep this open and fix the build scripts.

Sorry, I *did* close the issue prematurely; thanks for noticing!

New changeset e97ba4c690613d734843db218aeedce2f0e5937f by Steve Dower in branch 'master':
bpo-33614: Ensures module definition files for the stable ABI on Windows are correctly regenerated. (GH-7165)
https://github.com/python/cpython/commit/e97ba4c690613d734843db218aeedce2f0e5937f

Steve - PR 7165 appears to have caused a buildbot failure:
http://buildbot.python.org/all/#/builders/12/builds/910 

Thanks! I guess we were getting lucky in the past, as that error could have shown up for any number of reasons...

New changeset 3757939e9c00a36f939d774ec5c79e5d6b9a77bb by Miss Islington (bot) in branch '3.7':
bpo-33614: Ensures module definition files for the stable ABI on Windows are correctly regenerated. (GH-7165)
https://github.com/python/cpython/commit/3757939e9c00a36f939d774ec5c79e5d6b9a77bb

New changeset c489a767af46f33e73d96a5746e46a7365814db2 by Miss Islington (bot) in branch '3.6':
bpo-33614: Ensures module definition files for the stable ABI on Windows are correctly regenerated. (GH-7165)
https://github.com/python/cpython/commit/c489a767af46f33e73d96a5746e46a7365814db2

I *believe* this change will prevent the issue in the future, so marking this as fixed. But I wasn't able to reproduce it locally, so there may be something funny about either the file system or the version of MSBuild on the buildbots that caused it to not rebuild properly.
In any case, my change causes it to rebuild based on contents rather than timestamp, so it should be more reliable.

It seems like the commit c489a767af46f33e73d96a5746e46a7365814db2
 broke the AMD64 Windows10 buildbot: bpo-33675. I reopen this issue.

Already fixed in GH-7169, GH-7176 and GH-7177 (which I apparently typo'd "bpo" in the commit message, so they didn't link up properly here).