Boost Testing :
Subject: Re: [Boost-testing] Adding regression tests for mingw(-64)/gcc and clang on Windows
From: Edward Diener (eldiener_at_[hidden])
Date: 2015-08-17 16:26:05
On 8/12/2015 11:25 PM, Jim Bell wrote:
> On 2015-08-11 11:47 PM, Edward Diener wrote:
>> On 8/11/2015 9:40 PM, Tom Kent wrote:
>>> Can you document it somewhere? I'd be interested in adding clang/gcc on
>>> mingw runs to the ones I do on windows, but I had never had much luck
>>> getting them up and running.
>> I thought I had figured it out but it is still not working when trying
>> to run the regression tests. I am getting messages as if the
>> particular version of mingw-64/gcc used in a regression test is not in
>> my path when some executable is being run or tested ( so and so
>> mingw-64/gcc DLL can not be found ).
>> Yet when I manually test out libraries in the latest boost trunk
>> everything is fine. I can see when I run tests that Boost build is
>> prepending the path to the mingw-64/gcc version I am using to the
>> Windows PATH so that when the test is run, everything is found.
> I have mingw tests for mingw-32, mingw-w64 gcc 4.9, 4.8, 4.7, 4.6, 4.5,
> and 4.4. Also the mingw-gcc that ships with Qt 5.4, Qt 5.2 and Qt 4.
Do you compile/link from each particular mingw's bin directory when you
test ? Or do you put each mingw's bin directory in your Windows PATH
when you test ? Because my experience is that you can't just invoke gcc
or g++ without getting compile/link errors unless the particular 'bin'
directory is in the PATH.
> I don't recall everything I did to get it working, but I'll fire one
> back up tomorrow.
Whatever information you can give would be helpful.
> As I recall, clang wouldn't build one kind of library (dynamic or
> static, I don't recall which), and that killed the results. Has that
>> Soemhow the regression tests are not doing the same thing in running
>> the tests that I do manually to test any particular library. I will be
>> trying to find out what that is.
>>> On Tue, Aug 11, 2015 at 7:20 PM, Edward Diener <eldiener_at_[hidden]
>>> <mailto:eldiener_at_[hidden]>> wrote:
>>> On 8/11/2015 4:38 PM, Edward Diener wrote:
>>> For a long time now we have had no regression tests for
>>> mingw/gcc or
>>> clang targeting gcc on Windows. The only regression tests for
>>> are for various releases of VC++.
>>> I have limited resources but I have a laptop usually sitting
>>> except for occasional travel where I take it with me, and I am
>>> set up to
>>> run tests on that machine for mingw/gcc and clang on Windows.
>>> However mingw/gcc and clang have limitations when running tests
>>> in that,
>>> both relying on various mingw/gcc implementations, the path
>>> to the
>>> implementation must be added at the beginning of the PATH before
>>> any of the tests. I have already argued with the mingw-64 people
>>> this necessity but they are adamant that this should be the case
>>> just to compile/link properly ( and have setup their directory
>>> to enforce it ), and I don't want to waste my breath further
>>> with them.
>>> Currently I have a batch file that does that for me when running
>>> for any particular library, but I don't know how that can be
>>> when running the regression tests unless possibly I run
>>> regression tests
>>> for only a single toolset at a time, and manipulate the PATH
>>> But that is really time consuming since evidently all of Boost
>>> must be
>>> downloaded to the run directory each time. Is there an easy
>>> solution to
>>> this or do I have to look at run.py and try modify it
>>> accordingly ( I
>>> know Python well ) ?
>>> I figured out what my problem was running multiple tests, and how to
>>> achieve regression testing with clang and gcc on Windows.
>>> Also I notice on the Boost web site regression test page (
>>> http://www.boost.org/development/running_regression_tests.html )
>>> the options of the regression tests are somewhat documented but
>>> none of
>>> the commands are documented. Is there further documentation
>>> about the
>>> regression test commands anywhere ?