git submodules and jj file list

[vcunat]

I know that submodules aren't supported, but right now jj file list will output a path that doesn't exist. That seems problematic; for example the suggested replacement for git grep is not only cumbersome to write, but it also causes errors because of this.

Maybe the submodules should get rendered as empty directories? (I haven't thought it through, though)

[yuja]

Makes sense. Perhaps, submodule path can be emitted to stderr as a warning.

[arxanas]

• IIUC, it's not guaranteed that the submodule path doesn't exist on disk, and conceivably you would want to be able to learn about them?
  • I think materializing a submodule as an empty directory if it doesn't exist at all might be okay for many workflows, but not sure. I don't know if it would impede Git materializing the submodule at a later time.
• Does this mean that we ought to generalize the warning/stderr output to cases where the path might not exist on disk?
  • I think this could include ignored-but-tracked files?
• Does jj file list include files that are tracked but not present in the sparse checkout?
  • If so, maybe the usage recommendation is really that jj file list might always return unmaterialized-but-tracked files, so you always need to filter them?
  • Alternatively, it might mean that jj file list wants an additional flag to filter to materialized files only, when specifically referring to the working-copy revision.
  • Tangential: there might be a reason that you want to list files in another revision according to the current sparse checkout rules?
• I guess, generally-speaking, jj file list -r <rev> might produce files which aren't guaranteed to exist on disk currently, so why should -r @ be different?

[yuja]

jj file list works on commit, so the paths included in the output are not (and should not be) guaranteed to exist on disk. That said, I think it makes sense to exclude submodules because submodules are kinda trees, and jj file list doesn't print tree paths.

FWIW, at some point, we might add fileset functions like tracked() to filter the output.