add cwd to cache prefix

[chrisirhc]

Caveat: I don't know what I'm doing with zig. The zig code was mostly code assistant-generated.

I don't think this is a breaking change, but if somebody does rely on the cache location, they can set the HERMETIC_CC_TOOLCHAIN_CACHE_PREFIX to one of the prior values (/tmp, /var/tmp, C:\Temp) to maintain same behavior.

Test Plan

I added a debug print statement for the zig_cache_dir variable to see its value.
I tried the following test cases:

Input:HERMETIC_CC_TOOLCHAIN_CACHE_PREFIX	Output: zig_cache_dir (Printed value)
unset	<workspace root>/zig-cache
/tmp/my-test-cache	/tmp/my-test-cache
my-test-cache	<workspace root>/my-test-cache

Where <workspace root> for me looked like /private/var/tmp/_bazel_…/…/sandbox/darwin-sandbox/…/execroot/_main.

Closes #83
Closes #183
Closes #187

[chrisirhc]

Ready for review

[chrisirhc]

Not sure why https://github.com/uber/hermetic_cc_toolchain/actions/runs/11867448986/job/33075522540 passed yet https://github.com/uber/hermetic_cc_toolchain/actions/runs/11914055410/job/33201139834?pr=200 fails.

The error is:
error: unable to create compilation: Unexpected

Maybe related to: ziglang/zig#19973 , but I don't have a windows machine to test on.

Next things to try:

• to run using a dev/debug build of Zig to see a full trace
• to fiddle around on a windows machine
• try upgrading to zig 0.13 (Dead end, more issues, see upgrade to zig 0.13.0 #203 (comment) )

[chrisirhc]

Not sure why https://github.com/uber/hermetic_cc_toolchain/actions/runs/11867448986/job/33075522540 passed yet https://github.com/uber/hermetic_cc_toolchain/actions/runs/11914055410/job/33201139834?pr=200 fails.

The error is: error: unable to create compilation: Unexpected

Rerunning the CI seems to have passed (after I rebased back to a non-upgraded zig 0.13/14 attempt).
I'm guessing this is an intermittent error. 🤷 🫤

Moving this back to "Ready for review"

[linzhp]

@sluongng @laurentlb can you verify if this works for you?

[linzhp]

Can we put it in either bazel-zig-cache or bazel-out/zig-cache? Most bazel project have either /bazel-* or /bazel-out in their .gitignore. They don't need to ignore additional directory if we reuse the existing directories.

[cerisier]

Unless mistaken, wouldn't this store the cache inside the action sandbox root, unique to each cc target?
Also, wouldn't it be destroyed between each build invocation ?

[chrisirhc]

Unless mistaken, wouldn't this store the cache inside the action sandbox root, unique to each cc target? Also, wouldn't it be destroyed between each build invocation ?

This is probably right. I think I misinterpreted/understood the prior requirement mistaking the "workspaceRoot" for "action sandbox root".

After re-reading https://bazel.build/remote/output-directories , perhaps it should store this cache directory in either the outputRoot (equivalent to bazel-out) or outputBaseRoot, or in the workspaceRoot/bazel-zig-cache as mentioned above .

To get the outputRoot, I'm guessing I'd need to get the parent directory of the execRoot or use execRoot/_main/_bin/bazel-out ?

[chrisirhc]

After doing some more exploration, some findings:

• Sandboxing changes the path, so it looks something like /private/var/tmp/_bazel_chris/6c858e132b9297eb6238c2b551e7fa73/sandbox/darwin-sandbox/19/execroot/_main/../../../_zig_cache (or <outputBase>/sandbox/darwin-sandbox/19/execroot/_main/../../../_zig_cache). So caches will no longer be shared across actions. This is by-design, per documentation Sandboxing effectively disables any cache the tool may have. .
• It's possible to pass in the workspace-root (repo root) and add a .bazel-zig-cache directory if we want to persist the cache. But the user would then need to manage this cache themselves since bazel clean would not remove it. Though this assumes the workspace-root is writeable.