Module loading case-sensitivity windows inconsistency

[ReubenHill]

From https://discourse.julialang.org/t/module-loading-case-sensitivity-on-windows/18523/5 :

On Windows 10 (x64), I create seqtest.jl in C:/Users/MyUserName/Desktop/test. This contains

module SeqTest

println("loaded SeqTest")

end # module SeqTest

in the same folder I create procmsgseqtest.jl which contains

module ProcMsgSeqTest

println("loaded ProcMsgSeqTest")

end # module ProcMsgSeqTest

Then in the Julia v1.0.2 (x64) REPL I do the following:

julia> push!(LOAD_PATH,"C:\\Users\\MyUserName\\Desktop\\test")
4-element Array{String,1}:
 "@"
 "@v#.#"
 "@stdlib"
 "C:\\Users\\MyUserName\\Desktop\\test"

julia> using SeqTest
ERROR: ArgumentError: Package SeqTest not found in current path:
- Run `import Pkg; Pkg.add("SeqTest")` to install the SeqTest package.

Stacktrace:
 [1] require(::Module, ::Symbol) at .\loading.jl:823

julia> using ProcMsgSeqTest
[ Info: Recompiling stale cache file C:\Users\MyUserName\.julia\compiled\v1.0\ProcMsgSeqTest.ji for ProcMsgSeqTest [top-level]
loaded ProcMsgSeqTest

This is also confirmed for Julia v.1.1.0 x64.

According to the documentation for Base.require, ProcMsgSeqTest should not load due to case sensitivity not being observed.

As @GregPlowman noted in the discourse discussion, this is due to the length of the file names exceeding 8 characters.

So it turns out that case-sensitive check works for SeqTest because it is short, whereas ProcMsgSeqTest is too long.

[stevengj]

As discussed in #30588, one alternative is to use longpath(shortpath(path)) instead of just longpath in isfile_casesensitive, but that will fail on volumes that have disabled short-name generation. Another alternative is to use realname(path), but I think that will fail for symbolic links?

Fortunately, the bug here apparently only creates false positives and not false negatives, so it won't cause module loading to fail in a case where it is supposed to succeed.

A third option is to use longpath(shortpath(path)), but punt and return true if that throws an error — this might have false positives on volumes that don't support short paths, but otherwise will work.

cc @malmaud, as this is a continuation of #13542.

[StefanKarpinski]

I'm not sure if it's really worth the effort to guarantee an error in all these cases.

[vtjnash]

FWIW, it’s probably not just guaranteeing the error only, but also ensuring it keeps searching for the correct package and file (which probably won’t be found)