Pkg operations should be case-sensitive on case-insensitive filesystems

[stevengj]

e.g. I shouldn't be able to do Pkg.test("pycall"). Since packages are module names, and module loading is case-sensitive, Pkg should be too.

I just noticed this because Pkg.test("NBinclude") failed for a subtle reason (the tests loaded, but with an altered path) and it took me a while to realize that it was because I had mistyped Pkg.test("NBInclude").

[stevengj]

(Should be able to use the same machinery as #13542.)

[stevengj]

cc @malmaud

[jakebolewski]

I've fixed this, PR forthcoming once I figure out how to add tests.

[stevengj]

@jakebolewski, I also have a patch forthcoming. Can we compare notes?

[jakebolewski]

Sure, I didn't do anything special. Just made sure all the path relevant path comparisons in Pkg were case sensitive to the package name. @malmaud's code for OSX works for any inodes if you strip out trailing / for directories.

[stevengj]

See the commit linked above. I moved the isfoo_casesensitive checks to stat.jl. The tricky part was finding the relevant entry points in Pkg. Feel free to steal any part of this patch that is helpful.

I didn't think to strip out a trailing /, though.

[jakebolewski]

Well the two are pretty much identical sans stripping out trailing /, and yours has docs 👍

[jakebolewski]

I would advocate that case sensitivity becomes part of the isfile, isdir, ispath API through a keyword argument. But that seems like an API change for 0.6.

[stevengj]

Before we can make it part of the public API, we need to deal with the issue that currently the case is only checked for basename(path) on MacOS.

[stevengj]

Do you want to pull my patch, add your trailing / fix and any other differences, then submit a PR? (Or merge in the other direction if you prefer.)

[stevengj]

(Anyway, the change to the public isfoo API seems like it should go in a separate PR after the Pkg fix.)

[jakebolewski]

Sure, I can add to that commit.

[simonbyrne]

This is now fixed.