Re: [Boost-testing] [EXTERNAL] Windows build failures

Subject: Re: [Boost-testing] [EXTERNAL] Windows build failures
From: Belcourt, Kenneth (kbelco_at_[hidden])
Date: 2014-01-28 14:59:49

On Jan 28, 2014, at 6:21 AM, Beman Dawes wrote:

On Tue, Jan 28, 2014 at 12:27 AM, Belcourt, Kenneth <kbelco_at_[hidden]<mailto:kbelco_at_[hidden]>> wrote:

On Jan 26, 2014, at 4:06 PM, Beman Dawes wrote:

On Sun, Jan 26, 2014 at 10:12 AM, Tom Kent <lists_at_[hidden]<mailto:lists_at_[hidden]>> wrote:
I'm getting several builds failing on windows with the dreaded:
...
  File "C:\local\teeks99-02\b\tools_regression_src\regression.py",
line 434, in command_test_boost_build
    "boost/bin.v2/libs/any/test/any_test.test"));
WindowsError: [Error 3] The system cannot find the path specified:
'C:\\local\\teeks99-02\\b\\results\\boost/bin.v2/libs
/any/test/any_test.test/*.*'

(We really need to make more descriptive and accurate error handling)

When I look in results/bjam.log, I'm seeing this at the end:
error: Unable to find file or target named
error:     '../libs/sync/test/'
error: referred to from project at
error:     '.'

Any ideas on that?

I was finally able to reproduce this problem.  It depends on the order you run the tests.  If you run --tag=develop first (and this creates the boost_root directory) and then run --tag=master, it works fine.  If on the other hand you run --tag=master first, and it creates the boost_root directory, and then run --tag=develop, you'll get the sync error above.

At least on my boost working directory, the sync submodule was never initialized. A git submodule --init was needed.

Perhaps that happens when a submodule is present in develop, but not master?

Yes, I believe that's the issue.  I'm working on a fix for this.

Thank you very much!

I just pushed a change to regression.py in develop branch to init the submodules as Beman noted above.  This fixes the testing order problems, tested on Linux and Darwin.

Note that there's still some inefficiency in the regression.py script in that it seems to perform more git operations than necessary (multiple submodule fetch's, etc..).  I'll try to look at this next since it can incur network and authentication traffic that, for some testers, can be expensive.

-- Noel