Naturally, the code is going to symlink in-repo source code if it exceeds the file threshold (but not for formatter/fixer process invocations) This is certainly sub-optimal as the same exact subset of the sources likely won't be used across sandboxes. But c'est la vie. The real "issue" though is if users expect (or more accurately should be relying on) their sandbox to be mutable. Namely should they be able to add files to directories?

Although it might be surprisng to some, I lean towards "no you shouldn't assume its writable" After all the purpose of these sandboxes are hermicity and reproducibility. Conceivably, those could be poisoned if the running code starts adding/modifying files

Furthermore, very reasonable to require that the input root won't be modified to allow for the symlink optimization. (And maybe have a flag to

Process

that can disable the symlink optimization if there are ever outlier

Process

invocations that will violate the assumption.)

Plus it would support

node_modules

and Ruby bundler gem install directories to always symlink those directories.

Tangentially, should the API for

Process

provide for a "always symlink this source dir" attribute? (Versus just doing a heuristic.)

We'll have both inlcude and exclude lists for maximum custopmizability plus the huerstic for when we (the user) don't control the process