I believe that this patch is correct (cc @nojb ?).

I have a minor question. I thought that the caml/misc.h thing was used for C functions occurring in the OCaml runtime (byterun/foo.c), here it is used for a function in otherlibs/unix (while those have their own system-distinction layer of sort in having both unix and win32unix). Is it common for misc.h to have definitions only used in otherlibs, or a new trend? (I've seen another patch, from @nojb I think, also doing that just today).

Looks OK to me as well.

@gasche Indeed, misc.h contains a list of defines protected by #if _WIN32 used to choose the right function to handle Unicode for shared Unix/Windows code. It was just a matter of convenience that we also list there some functions that are only used in otherlibs. We could move these functions to some file in otherlibs if we want to (but as far as I can see there is no header file shared between unix and win32unix currently, so a new file would have to be created).

I'll cc @xavierleroy in case he knows about whether a byterun/ vs. otherlibs/*unix distinction would matter, but I don't see this as a blocker.

Has one of you to checked whether other Unix functions have been forgotten?

I just went over the list. There are two functions that MSDN tells me they exist in Unicode flavor that I had not considered: getaddrinfo and getnameinfo. Unicode host names anyone?

I will submit a PR adapting these two functions but I think we can safely leave it for 4.07. I mean, has anyone ever seen a non-ASCII host name?? 😃

Is this an approval for 4.06 as well?

@alainfrisch yes. I am operating under the assumption that this code would have been included with the general PR if it hadn't been omitted -- it's not a new risk. Are you positively or negatively surprised?

@gasche: yes, there is probably too many things defined in byterun/caml/misc.hand not enough reasons why they are defined there. A similar phenomenon occurs with utils/misc.ml. (Mental note: never name a module "Misc".) But that's not too serious and maybe one day we will clean things up.

@nojb: concerning Unicode in host names, see https://en.wikipedia.org/wiki/Internationalized_domain_name . Standards exist but don't seem widely used, in part by fear of spoofing of ASCII domain names by similarly-looking-but-different Unicode names.

Are you positively or negatively surprised?

I'm not surprised, but it's good to get an explicit agreement to push to 4.06.

Merged and cherry-picked to 4.06 (d9db5be). I've also fixed the GPR reference in Changes in a later commit (also cherry-picked).

Just for the record, I did do a check by looking for files which validated paths (which we've been unsurprisingly thorough on) and which didn't appear to be using caml_stat_strdup_to_os

See #1454 for getaddrinfo and getnameinfo.