feat: allow write_source_file(s) to write source files to bazel packages outside of the target's package

[gregmagolan]

We've received requests for this feature a few times now with the latest one being here https://bazelbuild.slack.com/archives/C04281DTLH0/p1704883631739649.

The glob trick to check that the out file exists must be disabled by setting check_that_out_file(s)_exists to False when writing to a different Bazel package. The trick doesn't work unless the out file is in the same Bazel package (glob's don't traverse into sub packages and can't look into sibling or parent packages).

Type of change

• New feature or functionality (change which adds functionality)

For changes visible to end-users

• Suggested release notes are provided below:

write_source_files(s) can now write source files to Bazel packages outside of the target's package when you set check_that_out_file(s)_exists to False.

Test plan

• New test cases added

[jbedard]

Will this not potentially add confusion to usage of this rule?

[gregmagolan]

Will this not potentially add confusion to usage of this rule?

The default behaviour is to not allow it and have the existence check so by default nothing changes. This allows some users to turn that check off and break the glass.

The docstring updates to out_file and check_that_out_file_exists now mention the opt-out. Do you have a suggestion for how to reduce the confusion?

[jbedard]

It's minor enough I'm fine with this. But imo docs explaining the restriction is better then expanding the api, adding docs explaining the restriction and docs explaining the api to workaround the restriction 🤷‍♂️

Can't users just move the write-source rule to the directory it's writing to?

[kormide]

It's minor enough I'm fine with this. But imo docs explaining the restriction is better then expanding the api, adding docs explaining the restriction and docs explaining the api to workaround the restriction 🤷‍♂️

Can't users just move the write-source rule to the directory it's writing to?

I'm also curious about this. Does this ever solve a problem that can't be solved by just moving the rule invocation to another package?

What is the experience like when the rule is invoked but the file doesn't exist yet?

[gregmagolan]

It's minor enough I'm fine with this. But imo docs explaining the restriction is better then expanding the api, adding docs explaining the restriction and docs explaining the api to workaround the restriction 🤷‍♂️
Can't users just move the write-source rule to the directory it's writing to?

I'm also curious about this. Does this ever solve a problem that can't be solved by just moving the rule invocation to another package?

What is the experience like when the rule is invoked but the file doesn't exist yet?

You can always add another write_source_file(s) target in the Bazel package you're writing to. With check_that_out_file_exists off, the DX if the file is not there is degraded but that is mentioned in the docs and the developer adding a write_source_files target with the explicit check_that_out_file_exists opt-out so they can write to another Bazel package would know that and create the file so other developers wouldn't hit the degraded DX.