Errors on Windows are different than on unix

[cadencemarseille]

Really frequently the errors are different on Windows than on unix. It would be great if the errors were the same.

See, e.g.: #10863 (comment)

[alexcrichton]

I'm unsure if this is a problem in our usage of libuv or if it's just a problem with windows in general. If it's our usage of libuv, we have work to do, but if it's windows in general, there's not much we can do about that.

[cadencemarseille]

I tried the following program using joyent/libuv@f166d6d and Visual Studio Express 2012 for Windows Desktop:

#include <stddef.h>
#include <stdio.h>
#include <stdlib.h>
#include <uv.h>
#include <uv-errno.h>

int main() {
    uv_loop_t *loop = uv_loop_new();
    int uv_err;

    uv_process_t child_req;
    uv_process_options_t spawn_options = { 0 };
    char *args[] = { "no-binary-by-this-name-should-exist", NULL };
    spawn_options.file = args[0];
    spawn_options.args = args;
    if (uv_err = uv_spawn(loop, &child_req, &spawn_options)) {
        fprintf(stderr, "%s (%d)\n", uv_strerror(uv_err), uv_err);
        return EXIT_FAILURE;
    }

    return uv_run(loop, UV_RUN_DEFAULT);
}

On Windows 8.1 Pro the correct code (ENOENT/-4058) is being returned:

no such file or directory (-4058)

But now I am confused why the test for PR #10863 required if cfg!(windows) { OtherIoError } else { FileNotFound }. Shouldn't the error code be mapped correctly?

[cadencemarseille]

Aha! joyent/libuv#1022 seems to have fixed the issue in libuv. That patch is included in libuv 0.11.16+.

[aturon]

It sounds like this ticket resolved. @cadencemarseille, do you agree?

[cadencemarseille]

Yes. It's an upstream issue that will be fixed when rust-lang/libuv upgrades to libuv 0.11.16+.