Set ANT_VERSION in xspec-bat.cmd

[AirQuick]

To prevent spurious test failures, xspec-bat.cmd (a test script for Windows) skips Ant tests when the ANT_VERSION environment variable is not defined.

You have to set the variable manually if you want to run Ant test on your local machine.

This pull request sets a pseudo ANT_VERSION in xspec-bat.cmd, if ant is available in PATH. Now you don't have to set the variable manually.

No code change in XSpec core.

---

This pull request derives from #243. So needs to be handled after that.

[AirQuick]

• This pull request just improves the way to run tests on a local machine. It doesn't touch the XSpec code. None of the existing tests are modified.
• The automated tests are not affected. (demonstrated on AppVeyor)
• This pull request makes it easy to handle some pending pull requests (towards Is it possible to apply Saxon configuration files?  #235 in particular).

@cirulls
So I think you can merge this (after #243) safely without going into details, or I'd like to merge this by myself in a few days.

[cirulls]

I executed xspec-bat.cmd on branch master and on my local Windows environment (which doesn't have ANT_VERSION set up) and I can confirm that the ant tests are all skipped. I then executed xspec-bat.cm on the branch of this pull request and in the same local environment and I can confirm that the ant tests are executed. So, as far as the changes in this pull request are concerned, this is all good.

However, I'd like to highlight an inconsistency between xspec.bats and xspec-bat.cmd. The ant tests in xspec.bat assume that ant is installed: there is no check for the existance of ant (see this test as an example). As a result, if ant is not installed, the test is probably going to exit and fail (I haven't tested it though).

Conversely, the ant tests in xspec-bat.cmd (both in this pull request and in master) first check that ant is installed and then execute the tests if that's the case. If not, the tests are skipped.

These scenarios occur when the test suite is executed locally. In fact, on Travis and AppVeyor ant is always installed first and, if the installation is not successful, there is an error code as we saw recently.

I'm not too fuzzy about which way we lean on (i.e. adding the check in xspec.bats or removing the check in xspec-bat.cmd) but I would prefer to avoid this inconsistency and align these two scripts as close as possible in terms of behaviour. I guess it also depends on how strongly we feel that ant tests should be executed locally before submitting a pull request.

@AirQuick, what's your opinion on this?

[AirQuick]

Right. They're inconsistent.
The inconsistencies (not only Ant but also some others) are resolved in #289. Ant is a must thereafter both on *nix and Windows. See the step 3 of How to test in #289.

So this pull request (#244) should be considered as a transitory measure to ease testing locally until #289.

[cirulls]

Thanks for clarifying and for fixing the inconsistency in another pull request. I'm merging this into master.