Core: add pants_hash() function.

[kaos]

Introduce new pants_hash() function to get the fingerprint for sources to be used as target field value.

When using a target type as glob, the default globs of its source(s) field will be used.

Example:

target(
  version_tag=pants_hash("src/*.y", "tests/*.z", some_target_type)
)

Relates to #8445

[kaos]

I didn't bother creating a dedicated type for this, as it's not part of any rule API.. but did consider doing it anyway for aesthetic reasons. So if there's any preferences in that direction... 🤷🏽

[kaos]

Given that this doesn't pick up sources from other targets or its dependencies, I'm not sure how useful this feature is...

[huonw]

Given that this doesn't pick up sources from other targets or its dependencies, I'm not sure how useful this feature is...

Hmmmm, that does make it seem rather error prone, where people might rely on it in incorrect circumstances!

Does it potentially also miss global options from pants.toml that might influence behaviour? (e.g. for a pex_binary the version of pex might change the output, so the configuration of the [pex-cli] subsystem should influence the hash.)

[kaos]

Given that this doesn't pick up sources from other targets or its dependencies, I'm not sure how useful this feature is...

Hmmmm, that does make it seem rather error prone, where people might rely on it in incorrect circumstances!

Does it potentially also miss global options from pants.toml that might influence behaviour? (e.g. for a pex_binary the version of pex might change the output, so the configuration of the [pex-cli] subsystem should influence the hash.)

Yea, it merely reports the digest fingerprint of the pointed at sources. That's all. Which at a low level is fine, but the UX is not good enough to expose the list of sources to fingerprint as input for the end user to provide.

Given the above limitations, I think we need an alternative approach at providing a hash for targets rather than loose sources.