Implement if_nametoindex for windows

[TheOnAndOnlyZenomat]

This implements the if_nametoindex function for windows.

I noticed zig threw a compile error when trying to compile the zigdig example code for windows.
If I read the code correctly, and looking at issue #20530, this seems also hinder all IP resolution on windows, even though if_nametoindex is only called in the ipv4 codepath.

I pieced this code together from looking at the darwin implementation and the old std.x namespace, which also contained an implementation for windows.

This seems to pass all test, except for this one

zig/lib/std/net/test.zig

Line 79 in d5db027

try testing.expectError(error.Overflow, net.Address.resolveIp6("ff01::fb%wlp3s0s0s0s0s0s0s0s0", 0));

. As far as I understand, the error.Overflow case should be handled before if_nametoindex is even called. If someone can help me fix this case, I'd be happy for any pointers.

I'm also happy about any other pointers and things I should fix.

Fixes #20530.

[andrewrk]

Thanks!

[alexrp]

CI failures look related to me.

[TheOnAndOnlyZenomat]

I'd be happy to merge any changes or fixes, if that is something that would be beneficial. Anything else I can contribute to this?

[alexrp]

Well one of the tests you enabled to run on Windows is failing there:

test
+- test-modules
   +- test-std
      +- run test std-x86_64-windows.win11_ge...win11_ge-gnu-skylake-Debug-libc 2690/2751 passed, 1 failed, 60 skipped
error: 'net.test.test.parse and render IPv6 addresses' failed: expected error.Overflow, found error.InterfaceNotFound
C:\Users\ci\george5\_work\zig\zig\lib\std\fmt.zig:1768:24: 0x7ff7ce2536c0 in charToDigit (test.exe.obj)
    if (value >= base) return error.InvalidCharacter;
                       ^
C:\Users\ci\george5\_work\zig\zig\lib\std\fmt.zig:1633:23: 0x7ff7ce231c4f in parseIntWithSign__anon_14207 (test.exe.obj)
        const digit = try charToDigit(math.cast(u8, c) orelse return error.InvalidCharacter, buf_base);
                      ^
C:\Users\ci\george5\_work\zig\zig\lib\std\fmt.zig:1521:5: 0x7ff7ce28132e in parseIntWithGenericCharacter__anon_23202 (test.exe.obj)
    return parseIntWithSign(Result, Character, buf, base, .pos);
    ^
C:\Users\ci\george5\_work\zig\zig\lib\std\fmt.zig:[150](https://github.com/ziglang/zig/actions/runs/12868010238/job/35904763031?pr=22555#step:3:151)8:5: 0x7ff7ce2719cf in parseInt__anon_21549 (test.exe.obj)
    return parseIntWithGenericCharacter(T, u8, buf, base);
    ^
C:\Users\ci\george5\_work\zig\zig\lib\std\net.zig:764:13: 0x7ff7cf446160 in if_nametoindex (test.exe.obj)
            return error.InterfaceNotFound;
            ^
C:\Users\ci\george5\_work\zig\zig\lib\std\net.zig:613:28: 0x7ff7cf44438d in resolve (test.exe.obj)
                break :blk try if_nametoindex(scope_id_str);
                           ^
C:\Users\ci\george5\_work\zig\zig\lib\std\net.zig:103:26: 0x7ff7cf442956 in resolveIp6 (test.exe.obj)
        return .{ .in6 = try Ip6Address.resolve(buf, port) };
                         ^
C:\Users\ci\george5\_work\zig\zig\lib\std\testing.zig:51:13: 0x7ff7cf441c1b in expectError__anon_1173136 (test.exe.obj)
            return error.TestUnexpectedError;
            ^
C:\Users\ci\george5\_work\zig\zig\lib\std\net\test.zig:79:9: 0x7ff7cf44262d in test.parse and render IPv6 addresses (test.exe.obj)
        try testing.expectError(error.Overflow, net.Address.resolveIp6("ff01::fb%wlp3s0s0s0s0s0s0s0s0", 0));
        ^

So that needs to be resolved.

[TheOnAndOnlyZenomat]

I also saw that. What I didn't get: To me it seems like the codepath where the error.Overflow is emitted happens before if_nametoindex is even called. And no other implementation of if_nametoindex handles the error.Overflow case. And I didn't touch any code outside that.

[tensorush]

Seems that this condition doesn't trigger:

zig/lib/std/net.zig

Lines 530 to 532 in 0679358

	if (scope_id_index >= scope_id_value.len) {
	return error.Overflow;
	}

Because when running on Windows with libc posix.IFNAMESIZE resolves to being 30, instead of being 16 like on Linux/Darwin:

zig/lib/std/c.zig

Lines 5783 to 5790 in 0679358

	pub const IFNAMESIZE = switch (native_os) {
	.linux => linux.IFNAMESIZE,
	.emscripten => emscripten.IFNAMESIZE,
	.windows => 30,
	.openbsd, .dragonfly, .netbsd, .freebsd, .macos, .ios, .tvos, .watchos, .visionos => 16,
	.solaris, .illumos => 32,
	else => void,
	};

So I'd suggest making that test string 12 characters longer (scope_id_value starts after %), covering the maximum posix.IFNAMESIZE value of 32 (Solaris/Illumos) just in case:

"ff01::fb%wlp3s0s0s0s0s0s0s0s0s0s0s0s0s0s0"

[TheOnAndOnlyZenomat]

I was also looking at that piece of code, but didn't make the IFNAMESIZE connection. I could update the test case tomorrow, rebase the latest changes and push again, if this is the accepted solution

[tensorush]

Seems that the "ff01::fb%12345678901234" test string, since its scope id was only 14 characters long, was triggering one of the different error.Overflows that happen later on:

zig/lib/std/net.zig

Lines 584 to 597 in f954764

	} else {
	const digit = try std.fmt.charToDigit(c, 16);
	{
	const ov = @mulWithOverflow(x, 16);
	if (ov[1] != 0) return error.Overflow;
	x = ov[0];
	}
	{
	const ov = @addWithOverflow(x, digit);
	if (ov[1] != 0) return error.Overflow;
	x = ov[0];
	}
	saw_any_digits = true;
	}

[jklw10]

worked for me in latest master build 0.15.0-dev.79+9f235a105 please merge 👍

[squeek502]

The if_nametoindex docs say:

The if_nametoindex function is implemented for portability of applications with Unix environments, but the ConvertInterface functions are preferred. The if_nametoindex function can be replaced by a call to the ConvertInterfaceNameToLuidA function to convert the ANSI interface name to a NET_LUID followed by a call to the ConvertInterfaceLuidToIndex to convert the NET_LUID to the local interface index.

Not sure it's worth worrying about for now, though. At some point I assume it'd be better to swap things over to that (or ConvertInterfaceNameToLuidW but I'm unsure if interface names can be non-ASCII?) and also use GetAddrInfoW instead of getaddrinfo (unrelated to this PR, was just looking at other ws2_32 calls)

(or lower-level equivalents if they exist, I didn't see any ntdll calls when running if_nametoindex with NtTrace [note, though, that it doesn't capture Rtl function calls])

[alexrp]

Does this actually fix #20530?

[squeek502]

Does this actually fix #20530?

Yes, it fixes the compile error in that issue.

> zig run 20530.zig
Started http server on port 4221

The tests that call resolveIp6 effectively cover it, since the error was from resolveIp referencing resolveIp6 which would hit the compile error.

[TheOnAndOnlyZenomat]

Are those CI failures related? I saw some similar ones in different PR's, and I don't think that the string concatenation in the test cast should cause a OOM during comptime

[alexrp]

Nope, those aren't your fault. The PR should be good to go; I'll merge it once we resolve those CI issues.

[TheOnAndOnlyZenomat]

Alright. Thanks everyone for working through this with me :D