Sandbox test repositories in TMP

[nulltoken]

No description provided.

[nulltoken]

/cc @carlosmn @ethomson

[Therzok]

Mac has a weird /private/var/tmp vs /var/tmp usage for temporary directory paths.

Also, what happens when the test suite fails? Is the directory guaranteed to be cleaned up? Windows is that weird OS which doesn't cleanup Temp dir on restart.

[ethomson]

Is the directory guaranteed to be cleaned up?

Are they ever?

[Therzok]

Is it ok to add a dependency on Mono.Unix? What about people who use CoreCLR?

[Therzok]

Rephrased: I'm totally ok with this solution, but we need to ensure coreclr as a target doesn't fail. 💃

Other than that, reflection usage there is fine by me. I'm not suggesting we depend on Mono.Unix assembly for the test run.

[carlosmn]

It's much more likely that the temp directory gets cleaned that a relative path to your bin/ directory. And the temp dir is where I would expect to look to clean up if I've been crashing tests.

We don't take a hard dependency on Mono.Unix, we just use it if it's available. If you're using CoreCLR on OSX, then some tests are going to fail, but that's OSX' fault; the only way we've found to get to the path which libgit2 returns is to do the libc realpath() work.

[ethomson]

👍

[nulltoken]

@Therzok The test repositories are cleaned in a "best effort" mode. From time to time, we forget to properly release a handle and DirectoryHelper.DeleteDirectory() goes on strike.

Ok to squash and merge?

[Therzok]

Fine by me.

[carlosmn]

👍

[nulltoken]

I've added some minor tweaks:

• Remove the ShadowCopyFixture special casing
• Add some garbage collecting prior to every test run that removes old test repos
  • That may haven't been removed because of leaks/failures
  • That cannot be removed (ie. ShadowCopyFixture)

I'd be a taker for a final review

[nulltoken]

Ok to merge?

[jamill]

I know this feedback is a bit late, but here we go:

Elsewhere we use Type type = Type.GetType("Mono.Runtime") to determine if we are running on mono - should we put this logic into a function that can be called? Or maybe this is only expected to succeed when running on Mono and *nix?

Also, might be nice to include a little more documentation around why this extra logic is necessary, where the publicKeytoken was found, (and maybe change the name to BuildTemporaryReposPath).

[Therzok]

This is a bit different. You could be running with Mono on Windows (lol :D) or embedded systems that don't have Mono.Unix. We're only doing that if we have the supplied functionality.

[carlosmn]

The bit we care about isn't that we're running on mono, but that we can run realpath(), and we only really need it for OS X because that's the one with the silly rewrites of temp paths. Anywhere else we can take it or leave it.

[jamill]

Thanks for the explanation - I think having this captured in code comments could be valuable. Also, it seems we are looking for a version of the assembly with a particular public key - This is because we only expect to work with version built by who ever controls that key?