reloc.ml: should not use String.lowercase_ascii

[nojb]

While investigating an issue related to ocaml/ocaml#1200, I noticed that there are several uses of String.lowercase_ascii applied to filenames in the module reloc.ml. It would be preferable not to use this function since those filenames may be UTF-8-encoded in the future.

[alainfrisch]

There are some use where the result is used as a memoization key to avoid repeated lookups on the file system (which is case insensitive). Which problems do you see with lowercase_ascii being applied to utf8-encoded strings?

[nojb]

Indeed, it is true that the use here is probably safe, but in any case it won't correspond to the notion of "case-insensitive" of the file system. Luckily, it seems that one can use some Window APIs to access this precise notion:

https://blogs.msdn.microsoft.com/greggm/2005/09/21/comparing-file-names-in-native-code/

[alainfrisch]

I'm not sure we need to be so precise, at least when this is only used to avoid repeated calls to the file system. What we need is that two filenames being equal after normalization are assumed to be identical by the file system, not the other way around.

Moreover, I'd prefer to maintain the property that flexlink can be produced using only the OCaml compiler and stdlib (no Unix, no external stub).

[nojb]

You are right, it is perfectly fine and safe. I got confused and was thinking that lowercase_ascii was the Latin-1 lowercase which could potentially lead to problems when passed UTF-8 strings.

Closing!