Running several tests requires Cygwin on Windows.

[juj]

Parsing the results of the buildbot test runs on Windows, there's a recurring problem on a system that doesn't have cygwin set up. See

http://clb.demon.fi:8112/builders/win-emcc-incoming-tests/builds/35/steps/Test%20emscripten%20incoming/logs/failed%20(825)

Several tests fail with

test_poppler (main.o1) ... Error: Exception thrown when invoking Popen in configure with args: "sh ./configure"!

also some tests fail with

test_cube2hash (main.o1) ... Error: Exception thrown when invoking Popen in make with args: "make -j 2"!

This is because there no 'sh' or 'make' on a Windows system. They can be found through Cygwin.

How much does emscripten depend on cygwin in general? I'm thinking it's more like 'accidental' that certain parts of emscripten go through cygwin, and not by decision to have cygwin e.g. based on some required feature.

The 'make' tool would be best replaced with using 'mingw32-make' on Windows, since that understands windows paths directly, whereas the cygwin 'make' doesn't and one runs into a lot of other problems trying to run makefiles through cygwin. E.g. one of them being that cmake generates Makefiles that don't go well through in cygwin environment.

[kripken]

We only rely on sh, make, etc. in some of the tests, because they have build systems that use configure (which needs sh) and make. In principle I think if they used a more portable build system like scons or cmake there would be no problem (does openjpeg pass? it uses cmake).

So since this is just tests, getting cygwin on the test bots might be the simplest solution.

[juj]

openjpeg doesn't pass yet, it fails on some path lookup issue I don't yet understand. Perhaps a difference between forward and backward slashes.

I did install cygwin now on the windows bot, but it started to suffer from #680. I see you committed a fix to that already, will have to test how it works, and if other issues with tests come up when cygwin is installed. (historically, I think #411 was caused by the same thing)

[juj]

I'll close this issue, since it's slightly a meta-issue, and the individual test failures have each since been identified and reported as separate items.