Issue with closing stream in Engine.

[ciathefed]

I have tried to use the example here and another one from the zinc-examples repo, and i keep getting this error.

thread 921703 panic: reached unreachable code
/opt/homebrew/Cellar/zig/0.13.0/lib/zig/std/posix.zig:245:22: 0x10446df9b in close (XXXXX)
            .BADF => unreachable, // Always a race condition.
                     ^
/opt/homebrew/Cellar/zig/0.13.0/lib/zig/std/net.zig:1798:32: 0x104490b53 in close (XXXXX)
            else => posix.close(s.handle),
                               ^
/Users/XXXXX/.cache/zig/p/122002abcd2597bc6c6dccaa76b6d871dcd461b4a82f4c80ff9fc68b01e402dae241/src/zinc/engine.zig:186:25: 0x1044c6c43 in worker (XXXXX)
            stream.close();
                        ^
/opt/homebrew/Cellar/zig/0.13.0/lib/zig/std/Thread.zig:429:13: 0x1044b4a87 in callFn__anon_11328 (XXXXX)
            @call(.auto, f, args) catch |err| {
            ^
/opt/homebrew/Cellar/zig/0.13.0/lib/zig/std/Thread.zig:674:30: 0x104491c9f in entryFn (XXXXX)
                return callFn(f, args_ptr.*);
                             ^
???:?:?: 0x19f3bf2e3 in ??? (libsystem_pthread.dylib)
???:?:?: 0x19f3ba0fb in ??? (libsystem_pthread.dylib)

• Zig: 0.13.0
• MacOS (aarch64)

[dravenk]

Thanks for report. I'm working on the new IO handle and this error will be fixed by next release.

[dxps]

Yesterday, using Zig 0.14.0-dev.2985, a minimal sample worked fine.

Today, using Zig 0.14.0-dev.2989, it fails (pretty identical with the one from the first message):

thread 3108413 panic: reached unreachable code
/home/dxps/apps/zig/lib/std/posix.zig:272:18: 0x122e1e3 in close (zinc_json_sample1)
        .BADF => unreachable, // Always a race condition.
                 ^
/home/dxps/apps/zig/lib/std/net.zig:1796:32: 0x125221a in close (zinc_json_sample1)
            else => posix.close(s.handle),
                               ^
/home/dxps/.cache/zig/p/1220a81e9421f9185fff79baba4beace2b8d9b4ec625ae3b0e06c3f3999993e87475/src/zinc/engine.zig:186:25: 0x1292e88 in worker (zinc_json_sample1)
            stream.close();
                        ^
/home/dxps/apps/zig/lib/std/Thread.zig:507:21: 0x126f148 in callFn__anon_17221 (zinc_json_sample1)
                    @call(.auto, f, args) catch |err| {
                    ^
/home/dxps/apps/zig/lib/std/Thread.zig:1378:30: 0x12537f1 in entryFn (zinc_json_sample1)
                return callFn(f, self.fn_args);
                             ^
/home/dxps/apps/zig/lib/std/os/linux/x86_64.zig:126:5: 0x126f231 in clone (zinc_json_sample1)
    asm volatile (
    ^
???:?:?: 0x0 in ??? (???)
Aborted

@dravenk Any ETA for the next release, please? Thanks.

[dxps]

@dravenk Thanks for the quick feedback! 🙏

Of course, commenting out that stream.close() avoids the failure.
Now, I am wondering what are the side effects. Just ran a minimal test:

After that, lots of connections (actually most of them, just a few of the - such as 3-4 - are in LISTEN and ESTAB) remained in CLOSE-WAIT state:

❯ ss -an | grep 8181 | wc -l
383
❯ 

I'll play and investigate further, to see how other Zig implementations are behaving in this case.

[dravenk]

Thank you all. This is the fix I remember, but the side effects are pretty obvious. Since std.net.server is very slow, the new branch will remove the std.net.server dependency, which will have a lot of work to do. This temporary code will also be removed. Anyway, thank you very much for your support and I will release the new version as soon as possible.