Don't cache failed generators

[jfrimmel]

When FuseSoC runs generators, the output is cached if the generator requests so. While this is generally a good idea, the cached output must not be used in subsequent runs if the generator fails with a non-zero exit code. In that case, the generator should be re-invoked in the next run to not need to user to manually run fusesoc gen clean to fix a faulty cache entry. Therefore this pull request adds in a check, if the generator call was successful and if it isn't the caching is prevented.

The first commit adds a test for a failing generator called in two subsequent FuseSoC-runs, expecting it to be called twice. The second commit changes the code calling generators to check for errors, taking action and re-raising that error. Currently the aforementioned action is to delete the entire working directory of the generator, since this will fail the current check for a cached generator result:

fusesoc/fusesoc/edalizer.py

Line 657 in 1403437

if os.path.lexists(generator_cwd) and not os.path.isdir(generator_cwd):

As an alternative, one could store a file .success (or similar) in the generator directory and only use the cache directory if that file is present there. I'm happy to implement this instead, if deemed better.

[olofk]

Ah, you're right. I didn't think of that. Agree that this implementation makes it harder to debug failed generators, but it seems like the safest option also, so let's start with this and consider other options if we need to.

Maybe we could extend this to just move the failed output dir somewhere else, (e.g. $work_root/failed_generators) but taking it in as is to solve the original problem. Thank you for your contributions!