WDYT about updating watch() instead to watch the list of directory entries if its target is a directory and the contents of a file if its target is a file? That would spare you a method on the API and I can't immediately identify any case that would be impossible with that setup.

Alternatively, WDYT about extending watch() to have arguments parallel with glob(): watch([<file name>]) would watch a list of specified files, watch(["*.txt"]) would watch every text file and watch(["**"]) (if implemented later), would be a recursive glob. It would also provide a convenient place to add exclude= and other accoutrements as needed. You might even be able to piggyback upon the implementation of glob().

(Doing my first review round of the day, I'll assume that you are still working on this change and that's why you haven't replied on this thread)

My bad -- it look like the changes were more extensive than I thought. Reviewed (mind answering the question above?)

fwiw i didn't reply to this thread yesterday because it didn't exist yesterday :P looks like you drafted it but never sent it until today.

but anyway, putting a glob pattern directly in rctx.watch is a bit awkward because you're presumably globbing from the working directory (ie. where the repo is being fetched into). so rctx.watch(['*.txt']) wouldn't make sense, but path.glob(...) probably would.

in any case i don't think we could realistically piggyback onto GlobFunction as it just works too differently: worrying about package boundaries, returning labels instead of paths, not digesting file contents, etc etc.

regarding folding watch_dir into watch: i ended up going with path.readdir instead, which i think it's a more natural api. we could actually still choose to make rctx.watch on a dir be sensitive to dirents in the future, but i don't think it's as important a question now. let me know if you disagree.

My bad it is then. I'm not the most relaxed and well rested I have ever been...

I have only one concern with the path.readdir() idea: we currently have the read() / watch() dichotomy for files and the "just use readdir() for everything" idea makes it so that we don't have it for directories. I'm not sure it matters, though. At the very least, this decision allows you to postpone the decision about whether you want the recursive watching syntax to be the same as that of globbing (I mean, that's probably like a day, but still...) and teaching readdir() to watch the file system is a pretty obvious win.