Question: is this asymmetrical? I.e. when a new package is added which collides with an existing one, are they both moved to the longer style of path or just the new one? Either is fine, but if this is going to be path dependent anyway (i.e. the order of adding matters), then you could just as well use sequential numbering like @pfitzseb suggested. I.e. the first one in a collision class gets A/APackage/, the next one gets A/Apackage.1/, the next A/apackage.2, etc. The argument for UUIDs was primarily that it could always guarantee non-collision without the path names depending on the details of how the packages were added.

Yes, this is asymmetrical. It's more work and gives a more complicated git history to move around existing directories and files.

The advantage of using UUID over sequence numbers anyway is that you have a decent chance to navigate to the right directory if you happen to know the name and the (partial) UUID, e.g. from the Pkg.status output. With sequence number you have no choice but to look it up in Registry.toml (or do a linear search through the directories) unless you happen to know the chronology.

Fair enough. Might be good to have an operation to "normalize" the layout of the repo by moving any packages with name collisions to longer directories, but that can be done separately or not at all.