Martin Sebor wrote:
Andrew Black wrote:
Greetings Martin.
I think it's likely that we will need to make some changes to the
windows infrastructure.
The cause of the missing legacy tests is twofold. First, the legacy
test executables appear to fail to compile in dynamic builds because
the legacy test driver symbols aren't resolved. This seems to point
to a failure to export them correctly.
Ah. That's something we need to look into here as the legacy
part of the driver isn't in Subversion
I suspect the place you'd need to look is near the changes I made on the
14th and you made on the 15th.
Second, the runall.wsf script doesn't pass the names of non-existent
executables to the exec script. Therefore, you don't see any targets
in the COMPILE or LINK states when the exec utility runs.
IMO, this needs to change. The table must show the status
of all the tests, not just those that compiled and linked
successfully.
Unless you see a quick change that your or I could easily
make ourselves I'll ask Farid to do it since he is the
expert on the Windows scripts.
I don't see any obvious quick changes. What will probably need to
happen is the script will need to glob the source directories, rather
than filtering out the built executables from the remainder of the build
directory.
Looking at the runall.wsf and generated .stdcxx_runtests solution, I
believe that the cause of the format messages likely is coming from a
failure to pass the compatibility mode options to the test suite.
Farid, the additional switches that need to be passed to the exec
utility when running the tests are '--compat -x "--compat -O -"'
(minus the single quotes). Hopefully the shell will handle the quote
processing correctly.
That was suspicion as well. Again, unless you see a quick
fix let's wait for Farid to fix it.
The way I'd tend to fix it would be to add a /COMPAT switch to the
runall.wsf script which adds the necessary flags to the command line for
the exec utility, then include that flag on the command line when
running the runall.wsf script in the .stdcxx_runtests solution.
However, I would rather have Farid fix it in the way he feels best, as
he knows the Windows build infrastructure better than I do., and he'll
be altering related parts of the scrips for the first part.
--Andrew Black
