Normalize $SRC_DIR

[kflansburg]

This change normalizes absolute paths for the crate manifest directory to $SRC_DIR.

This closes #198.

In that issue, it was discussed that it is difficult to determine the correct manifest directory. I'm not sure exactly the semantics of the working directory when normalize_output is executed, but it was reliably the manifest directory in all of my testing (running Cargo from many different directories in a large Cargo workspace).

[Manishearth]

This seems okay but cc @flip1995 and @camsteffen for a double check

[camsteffen]

I think this could technically be wrong if running cargo test with CARGO_MANIFEST_DIR. Don't know if that's much of a concern in practice. Would a config value be better?

[kflansburg]

Ok, I did try using CARGO_MANIFEST_DIR, but it was returning the path to the compiletest-rs crate. I now see that I was using env! which is applied at compile time. Switching to std::env::var appears to return the correct directory, so I've updated the PR to use this value.

I also considered adding a configuration parameter, but that would deviate from the behavior of the official tool.

[camsteffen]

I'm not sure it makes sense to re-use cargo's CARGO_MANIFEST_DIR in compiletest.

I also considered adding a configuration parameter, but that would deviate from the behavior of the official tool.

The official tool has the rustc project structure hard-coded in, so I wouldn't worry about deviating from that. But we can also consider making a change upstream. I think rustc would be amenable to small refactors that decouple compiletest from rustc.

[kflansburg]

What is the issue with using CARGO_MANIFEST_DIR? It seems like it will work in the most situations. Are there cases where Cargo will not be in use?

[camsteffen]

I think it would be a little unfortunate to set CARGO_MANIFEST_DIR just for the purpose of configuring compiletest. There could be a case where SRC_DIR and CARGO_MANIFEST_DIR are wanted to be different. That said, I guess using CARGO_MANIFEST_DIR is still sensible and better than nothing.

[kflansburg]

Ah, I see. Just to be clear, CARGO_MANIFEST_DIR is automatically set by Cargo to be the crate manifest path (or workspace manifest if in a workspace), so I believe this would do the right thing under most circumstances, without the user having to set anything.

Im not familiar with what situations call for overriding this variable, but I could potentially see the need for specifying an alternate SRC_DIR when doing so. That could be done with a new config parameter like you suggest.

[camsteffen]

Just to be clear, CARGO_MANIFEST_DIR is automatically set by Cargo to be the crate manifest path (or workspace manifest if in a workspace)

Oh good point. I was thinking it is set by the user. That does make it better.

[kflansburg]

@flip1995 @Manishearth @camsteffen will it be possible to review this soon?

[flip1995]

LGTM. Using the CARGO_MANIFEST_DIR sounds reasonable to me.

[Manishearth]

Fixing the test failures in #245