Add a V2 builtin rule to go from FilesContent -> Snapshot

[Eric-Arellano]

We would like a nice abstraction to create a Snapshot given a set of file paths with their corresponding content (i.e. FilesContent).

This would be far better than the current approach of using touch to do the same thing, as done in #7696. Notably, the current implementation only works for empty files too.

However, going from FilesContent -> Snapshot would likely lead to a cycle so we might need a different name than this. Refer to #7696 (comment).

[illicitonion]

@stuhood Any thoughts on what this API should look like? Ideally in my mind it would be:

files_content = FilesContent((
  FileContent(path="roland", content=b"European Burmese"),
  FileContent(path="susannah", content=b"Not sure actually"),
))
snapshot = yield Get(Snapshot, FilesContent, files_content)

but that would lead to graph ambiguity through Snapshot -> FilesContent -> Snapshot cycles. I guess we could do something like:

files_content = FilesToSnapshot((
  FileContent(path="roland", content=b"European Burmese"),
  FileContent(path="susannah", content=b"Not sure actually"),
))
snapshot = yield Get(Snapshot, FilesToSnapshot, files_content)

but it feels kind of sketchy to introduce a whole new type just to differentiate between "we expect this to be an output" and "we expect this to be an input" when really the two should be interchangeable...

[stuhood]

but that would lead to graph ambiguity through Snapshot -> FilesContent -> Snapshot cycles. I guess we could do something like:

It shouldn't, I don't think? Although I have been surprised before (cf #7710).

[illicitonion]

If I add a rule to go from FilesContent -> Digest, I get a whole host of No rule was available to compute errors.

If I add a rule to go from FilesContent -> Snapshot, I get no errors. But I expect to because there is a cycle in the rule graph.

The fact that these give different results is surprising and concerning...

[stuhood]

The Digest issue is expected due to #7710 (the bug is that there is ambiguity between a Digest already in a rules' params and requesting a Digest via a Get).

If I add a rule to go from FilesContent -> Snapshot, I get no errors. But I expect to because there is a cycle in the rule graph.

Mind attaching a dot file or PDF?

[stuhood]

Oh, and I don't expect FilesContent -> Snapshot to be ambiguous in practice precisely due to the constraint that is described as "mostly working" in #7710: the parameter to a Get must be consumed in a subgraph, so it will be non-ambiguous to Get in that case.

EDIT: except the cycle would still theoretically be usable. Hm. Will think about it.

[stuhood]

Oh, and I don't expect FilesContent -> Snapshot to be ambiguous in practice precisely due to the constraint that is described as "mostly working" in #7710: the parameter to a Get must be consumed in a subgraph, so it will be non-ambiguous to Get in that case.

EDIT: except the cycle would still theoretically be usable. Hm. Will think about it.

Having said that, if it currently works, I'd rather that you take advantage of that, as it will force me to apply constraints in #7710 that continue to allow for the clearly useful case of a bidirectional conversion.