fix: pass toolchain env, lint_target_headers

[peakschris]

This PR has some minor updates that I've found are needed in for our codebase:

1. env-vars required by the compiler and known to the toolchain are automatically passed to clang-tidy. Avoid use of use_default_shell_env when running commands.
2. Allow the lint_target_headers regex to match header paths with mixed \ and / separators that seem to be produced by clang-tidy.
3. Set an envvar to prevent msys bash from mangling the path
4. Improve flag handling

---

Changes are visible to end-users: no

[peakschris]

FYI @alexeagle @jsharpe

[alexeagle]

Will look for a jsharpe peer review once this doesn't have unrelated changes

[peakschris]

@alexeagle @jsharpe I think this is ready for review now

[alexeagle]

Hey @jsharpe could you take a quick look at this? I'm out of my depth.

[peakschris]

Thanks jsharpe, I've applied two suggestions and had a question about the third. Much appreciated!

[peakschris]

@jsharpe please would you check two latest commits, we were finding some issues when running at scale

[jsharpe]

@jsharpe please would you check two latest commits, we were finding some issues when running at scale

Just a quick note to say this is still on my radar to look at but I've not had much time to do so! Will try to in the next week or so.

[peakschris]

@jsharpe I know you're busy but we are waiting to consume these fixes, any chance you could have a quick look?

[jsharpe]

This seems to be mostly working on linux with RBE / toolchains_llvm - 2 things I run into:

1. Failure to find some C system headers float.h, stddef.h etc.. I suspect this is a toolchain bug though as its picking up the system gcc version of some of these files instead of using the hermetic files from the toolchain_llvm
2. Headers from implementation_deps don't seem to be found.

@peakschris you mentioned you ran into some issues running at scale? What specifically were the issues there?

[jsharpe]

I'm happy with these changes as they are - the issues I've run into with running this on linux can likely be easily fixed in follow-ups so no need to block landing this with those changes.

[peakschris]

@jsharpe thanks for the approval

you mentioned you ran into some issues running at scale? What specifically were the issues there?

With large directories the command line parameter limit was being reached, so one of the changes here was to use param files as much as possible.

[peakschris]

@alexeagle please would you approve now that jsharpe has approved?

[jsharpe]

@alexeagle another ping on this as I think you were on vacation when the original ping was done so this may have slipped through your notifications.