In console: pcbuild/win32/python_d.exe -m test -j6
...
[393/395/1] test_fileio
Traceback (most recent call last):
  File "F:\Python\dev\35\lib\runpy.py", line 170, in _run_module_as_main
    "__main__", mod_spec)
  File "F:\Python\dev\35\lib\runpy.py", line 85, in _run_code
    exec(code, run_globals)
  File "F:\Python\dev\35\lib\test\__main__.py", line 3, in <module>
    regrtest.main_in_temp_cwd()
  File "F:\Python\dev\35\lib\test\regrtest.py", line 1564, in main_in_temp_cwd
    main()
  File "F:\Python\dev\35\lib\test\regrtest.py", line 738, in main
    raise Exception("Child error on {}: {}".format(test, result[1]))
Exception: Child error on test_fileio: Exit code 2147483651

In crash box: Debug assertion failed
/dd/vctools/crt_bld/self_x86/crt/src/chsize.c
line 56  _osfile(desfile) & FOPEN

By leaving box alone, all but about 4 tests finished.  When clicked on button, fileio message appeared. On previous run, click as soon as box appeared, and console window frooze much sooner, with just half of tests run.

There were also 'Python_d.exe has stopped working' messages, but same true with 3.4 tests, which finished.

Which version of MSVC are you using?

The same '2010' used for 3.4.

That may be the problem; it may be time for you to upgrade to VS 2015,
which is now in RC.  As recently as this morning, I've had a clean test run
with 2015 Community Edition on Windows 8.1.

It's entirely possible that I missed a _PyVerify_fd call around the ftruncate implementation, since that function is now a no-op for VS 2015 builds. The only reason I didn't remove them all when adding the new handling for was because there was no trustworthy version of VC14 around then (it was still Preview).

VS 2015 RC is trustworthy, so it is time to upgrade, but I'm still not quite ready to aggressively break everyone who hasn't. Maybe we should?

I just downloaded the 3mb vc_community.exe. I intend to try it tomorrow and report.

Unless you intend to support compiling with vs2010 after 3.5 is released, and from the devguide and your comment I presume not, then with beta nearing, I would agree that we should be testing 2015 builds.

Just so you're not too surprised - that vc_community.exe is a downloader that will eventually need 6-8GB :(

I'm working with the teams involved to try and get a compiler-only release, but for now the smallest installation that should work is Express for Desktop (from https://www.visualstudio.com/downloads/visual-studio-2015-downloads-vs.aspx, under the "Visual Studio 2015" expander at the lower left).

Express for Desktop still takes about 4GB... I'm going to up the urgency of a sensibly sized package for people who just need the compiler.

Thanks for the warning. I was expecting 'expansion', but 5 GB (what site says for ...Desktop) is a big chunk of the 17 GM I have left on C: (111GB SSD). (Do you have any idea how much free space I will need for Win 10 upgrade?)  So I think I will reverse myself and wait to install and suggest holding off on forced upgrade.

I download vc_community.exe by just clicking and giving a save directory.  When started, it says 9GB with options unchecked (10 GB before I unchecked one).  (At which point I quit.)

Downloading just Express for Desktop requires an MS account. It seems odd to need more (an account) to get half. It is unclear what type of account is needed -- the old free MS Now account, or the normally paid developer account.  Skipping for now rather than proceed further.

It's just a free account - any Hotmail, live.com or outlook.com email will do. The Microsoft equivalent of having a Google or Facebook account. (An MSDN Subscription is the other usual option, though that's associated with a Microsoft account anyway. I don't think there are actually any other account types these days.)

I haven't looked too closely at Win 10, but I'd be surprised if you could install it with only 17GB. It's possible that it would clean up a lot of the existing Windows files and not take that much though, I really don't know right now.

FWIW, our "AMD64 Windows7 SP1 3.5" buildbot hits this > 50% of the time in the regression test suite.  So it's not just Terry.

http://buildbot.python.org/all/builders/AMD64%20Windows7%20SP1%203.5/builds/78/steps/test/logs/stdio

I don't see any crash in that log, though the importlib tests appeared to time out.

Assert messages appear on the buildbots because they use debug builds and we don't suppress them completely. Provided the tests keep running without failure, it's fine.

To avoid the dialog when testing a debug build, you need to pass -n to test.