Different workspaces produce different hashes

[erikkerber]

Let me know if "feature" or "bug"

Assume a Bazel repo in /a:

cd a
bazel run :bazel-diff --script_path=/tmp/bazel-diff
cd ..

/tmp/bazel-diff generate-hashes --workspacePath=/a /tmp/a-hash
cp -r a b
/tmp/bazel-diff generate-hashes --workspacePath=/b /tmp/b-hash
cmp /tmp/a-hash /tmp/b-hash

As a background — we're caching the hash results to avoid too many Git checkouts/fetches in builds, but finding out that different workspaces result in different results. The workspaces are a boring implementation detail of our CI system which different paths/checkouts for different jobs.

[aarontsharp]

We are also trying to accomplish similar as our CI system by default does shallow clones and we likewise want to avoid switching branches and calculating hashes for previous builds. I've found unless the hashes are generated on the exact same machine, they're always different. In our case, even different CI runner instances with otherwise identical configuration, OS, file systems, etc always produce different hashes.

[tinder-maxwellelliott]

Can you test out #182 and let me know how it works, if it works we can find a way to set the list of ignored attributes when hashing

[erikkerber]

@tinder-maxwellelliott 100% did the trick. Awesome 🙇🏼

[tinder-maxwellelliott]

Okay Ill work on making it so the list of ignored attributes can be provided by an end user

[tinder-maxwellelliott]

@erikkerber Can you do one last test with #182, I added the ability to list attributes you wish to ignore during rule target hashing

[aarontsharp]

@tinder-maxwellelliott #182 worked for me, thanks! I had to add location and path to the ignore list. In my case, I observed many of the external rules (like bazel_tools) always seemed to have a different path in our different runners. I'm testing this out in our CI environment now, but I think this likely addresses the issues we were seeing. Thanks again!

[tinder-maxwellelliott]

Awesome, I'll close this then and merge it in. Thanks Y'all

[jieyu]

I was trying to use the --ignoredRuleHashingAttributes flag to ignore certain attributes from a rule. So I have a rule that has a private attribute _go_vendor_modules_txt that is a filegroup. when compute hashes, I set --ignoredRuleHashingAttributes=_go_vendor_modules_txt. However, when trying to get impacted targets, it seems that the change of files in _go_vendor_modules_txt is still not ignored. Anything I did wrong?

[tinder-maxwellelliott]

I was trying to use the --ignoredRuleHashingAttributes flag to ignore certain attributes from a rule. So I have a rule that has a private attribute _go_vendor_modules_txt that is a filegroup. when compute hashes, I set --ignoredRuleHashingAttributes=_go_vendor_modules_txt. However, when trying to get impacted targets, it seems that the change of files in _go_vendor_modules_txt is still not ignored. Anything I did wrong?

Is this a rule attribute? Last I checked the proto I saw nothing like _go_vendor_modules_txt

[jieyu]

@tinder-maxwellelliott it's a custom rule in house.