Error: undefined fun 'isatty' for LibC on Windows

[Blacksmoke16]

Bug Report

Discovered as part of https://github.com/athena-framework/athena/actions/runs/4370118689/jobs/7644691322 within the framework tests:

Creating library C:\Users\runneradmin\AppData\Local\crystal\cache\D-a-_temp-crystal-nightly-true-undefined-src-ecr-process.cr\macro_run.lib and object C:\Users\runneradmin\AppData\Local\crystal\cache\D-a-_temp-crystal-nightly-true-undefined-src-ecr-process.cr\macro_run.exp
Showing last frame. Use --error-trace for full trace.
  
In D:\a\_temp\crystal-nightly-true-undefined\src\crystal\system\win32\socket.cr:338:10
  
 338 | LibC.isatty(fd) == 1

Error: Error: undefined fun 'isatty' for LibC

I haven't manually confirmed it, but I'd assume this is a resume of doing like STDIN.tty? on windows.

Seems the actual binding for that function was missed in #10784 and this is first time someone tried using IO#tty? on windows?

[Blacksmoke16]

Okay no, STDIN.tty? works just fine. But it does fail if you actually require socket:

require "socket"

pp STDIN.tty?

 339 | LibC.isatty(fd) == 1
            ^-----
Error: undefined fun 'isatty' for LibC

[HertzDevil]

Easiest to just copy Crystal::System::FileDescriptor#system_tty?, which uses LibC._isatty instead, to Crystal::System::Socket#system_tty?

[straight-shoota]

I don't think a socket on Windows can ever be a tty. Unlike on POSIX platforms, they're disparate handles. And AFAIK you cannot turn a console handle into a socket handle somehow.

So perhaps the trivial solution is for Socket#system_tty? to return false on windows?

[Blacksmoke16]

Based on what I read that's probably worth a try. It'll at least get things compiling and can always revisit if someone ever comes across a case where it could be true. Will see what CI things. I pushed up another commit.