Allow out-of-source testsuite builds.

[dnadlinger]

Previously, "./" was prefixed to $RESULTS_DIR in several places, while $RESULTS_DIR == "test_result" was assumed in others.

This is useful for integrating the DMD test suite with the build system of other projects, e.g. LDC (via https://github.com/ldc-developers/dmd-testsuite).

[braddr]

Make this: export RESULTS_DIR=.$(DSEP)test_results

And move it below the if wi32 block just below this line and I believe that'll make it work on win32.

[braddr]

Thanks for doing this cleanup. Get it working on win32 and I'll merge it in.

[dnadlinger]

Applied your suggested change – let's see how it fares on the Win32 box this time (unfortunately, I don't have a Win32 VM around for testing right now). Also submitted a DMD issue, http://d.puremagic.com/issues/show_bug.cgi?id=8736, because this isn't the first time I've hit this issue.

[braddr]

Grf.. ok, I'll go poke at this one on a win32 box. I hate that part of the tester, it's twitchy as hell. But all in the service of not having to maintain two test drivers.

[dnadlinger]

Ah, the problem is of course that ${…} is only substituted by sh, not by the Windows shell. The easiest fix would be to perform the substitution in d_do_test instead, no?

[braddr]

There's multiple problems, and unfortunately I don't have time to drive solutions, so you're going to have to. A (potentially incomplete) list:

1. serialization of the d_do_test compilation and test_results directory creation is broken
2. variable expansion (as you noted)

I do like the fixes you've proposed here.. but they need to work on windows too.

[dnadlinger]

This is… interesting. GitHub shows commits here that are already in master. Yes, I checked the SHA-1 hashes, it's not a botched merge.

[dnadlinger]

@braddr: You might want to review this. Seems to work on Windows now.

[dnadlinger]

Fixed botched merge and adapted a test case that has been added in the meantime.

[dnadlinger]

Win32 fails right now, but that's an auto tester issue (can't connect to GitHub).

[braddr]

Would you mind undoing the change from test_results to generated? I like fixing all the call sites and appreciate that changing it helped find them, but I don't want to break anyone who's scripted against the current output directory name.

[dnadlinger]

I agree, but can we have the autotester run it with a different output directory name? The reason I left the change in is so that people don't accidentally commit changes relying on test_result.

[dnadlinger]

Removed the commit.

[braddr]

I agree that it'd probably be a good idea for the auto-tester to force a different name to help catch regressions. I'll add that to my todo list. It ought to be trivial.

[dnadlinger]

As trivial as specifying RESULTS_DIR=xyz on the invocation, hopefully.

[braddr]

yup

[dnadlinger]

Thanks!