search

r-lib / pak

A fresh approach to package installation
https://pak.r-lib.org
684 stars 62 forks source link

Record local sha? #699

Open infotroph opened 1 month ago

infotroph commented 1 month ago

When the path to a local package is inside a version-controlled directory, would it be desirable in principle for pak to keep a record of its commit hash at install time?

When debugging dev packages I often want to know which commit I installed from, and have come to really appreciate how easy it is to get that from the RemoteSha field when a package was installed from GitHub or R-universe.

For packages installed locally this doesn't seem to be recorded, and from https://github.com/r-lib/pak/issues/621#issuecomment-2080405933 I suspect that may be intentional. I agree with that comment that I don't want pak to make upgrade decisions based on the state of a local directory, but could the mentioned "we could add it and ... use for other things in the future" include providing a sha for runtime diagnostics?

(For context, I ask because I just hand-rolled a hash-recording step in the build script for one of my internal packages and thought "Well, that works, but this entire script ought to just be pak::local_install()" 😄 )

gaborcsardi commented 1 month ago

It is not trivial to decide if a local directory has an up to date sha, we'd probably need to call git, which is not so great in itself, plus there are a lot of edge cases.

infotroph commented 1 month ago

Thinking through the process, I think what I'm picturing here is (roughly equivalent to?) the ability to pass a local path to the git:: package source, which probably implies "that would mean implementing support for the Git local transport protocol." Definitely not a small change.