Support cygwin

[Berrysoft]

Description

This PR adds support for Cygwin & MSYS2.

• Cygwin wchar_t is UTF-16.
  • wcwidth is replaced by unicode-width. removed.
  • wcrtomb and mbrtowc are replaced by c32rtomb and mbrtoc32.
• The target is cross-compiled from Windows.
• There are some bugs on cygwin, so I have to add some workarounds.

Fixes issue #10956, though it has been closed.

Tracking:

• Add std support to cygwin target rust-lang/rust#137621
• feat: add cygwin support nix-rust/nix#2610
• Add cygwin support lambda-fairy/rust-errno#106

TODO:

• Link libintl correctly
• Link libpcre2 correctly

[faho]

Looks fine by me, I think @ridiculousfish would like to take a look.

Awesome work, btw

[ahaoboy]

Thanks for your excellent work. I tried to test it with the latest version(version 4.1.0-alpha0-g509826571) under msys2, but encountered a strange bug. Every time I type a character, a new line is created. I don't know whether I should open a new issue because cygwin does not officially support it yet.

[Berrysoft]

@ahaoboy Do you use oh-my-posh? It might be a bug of it, I think. I haven't dig into the code of oh-my-posh so not that sure.

[ahaoboy]

@ahaoboy Do you use oh-my-posh? It might be a bug of it, I think. I haven't dig into the code of oh-my-posh so not that sure.

I use starship, which works fine on fish 3.7.1

https://packages.msys2.org/base/fish

 starship --version
starship 1.22.1
branch:master
commit_hash:d605196
build_time:2025-01-11 17:09:54 +00:00
build_env:rustc 1.84.0 (9fc6b4312 2025-01-07),stable-x86_64-pc-windows-msvc

[Berrysoft]

OK, seems that the problem is a little complex. I'll look into it. You may try to disable starship temporarily and see if the problem is solved.

[faho]

This is an issue where I would suspect the terminal, not the prompt.

Fish 4.0 is more demanding on the terminal, and that can trigger bugs.

This isn't a manifestation I've seen, but I know the most frequent trigger is the kitty keyboard protocol, so try set -Ua fish_features no-keyboard-protocols to turn that off (and restart fish after).

[ahaoboy]

I removed all plugins and configurations, and only added a prompt configuration. The printf function will add a new line every 3 or 4 inputs.

I don't know if there are other possible reasons. It depends on whether other Windows users can reproduce the problem with the same configuration.

# ~/.config/fish/config.fish

set -Ua fish_features no-keyboard-protocols

function fish_prompt
    echo

    # This line of code will cause a newline to be added every 3 to 4 inputs.
    printf '%s@%s %s%s%s > ' $USER $hostname (set_color $fish_color_cwd) (prompt_pwd) (set_color normal)
end

[Berrysoft]

Well, I can reproduce such problem with oh-my-posh, both with Windows Terminal and mintty. It might be some misc problems between fish and cygwin api and terminal...

[faho]

set -Ua fish_features no-keyboard-protocols

Don't put that in config.fish. It will cause the universal variable to grow.

It's enough to run it interactively. If you must have it in config.fish, use

contains -- no-keyboard-protocols $fish_features
or set -Ua fish_features no-keyboard-protocols

[Berrysoft]

@ahaoboy I can reproduce your problem with your config.

It's not adding a new line in 3 or 4 inputs. It's adding a new line once per second you input.

[faho]

It's adding a new line each second you input.

Terminal timestamps, see

fish-shell/src/screen.rs

Lines 854 to 873 in 360cfdb

	fn check_status(&mut self) {
	let _ = std::io::stdout().flush();
	let _ = std::io::stderr().flush();
	if !has_working_tty_timestamps() {
	// We can't reliably determine if the terminal has been written to behind our back so we
	// just assume that hasn't happened and hope for the best. This is important for multi-line
	// prompts to work correctly.
	return;
	}
	if self.mtime_stdout_stderr != mtime_stdout_stderr() {
	// Ok, someone has been messing with our screen. We will want to repaint. However, we do not
	// know where the cursor is. It is our best bet that we are still on the same line, so we
	// move to the beginning of the line, reset the modeled screen contents, and then set the
	// modeled cursor y-pos to its earlier value.
	let prev_line = self.actual.cursor.y;
	self.reset_line(true /* repaint prompt */);
	self.actual.cursor.y = prev_line;
	}
	}

You might want to add Cygwin to has_working_tty_timestamps' exceptions.

[Berrysoft]

Thank you, @faho ! I have pushed the change and now the problem should be fixed.

[ahaoboy]

Thank you, @faho ! I have pushed the change and now the problem should be fixed.

Can you update https://github.com/Berrysoft/fish-shell/releases? I also want to test if there are any other issues in starship

[Berrysoft]

My custom release has been updated.

[Berrysoft]

Seems that poll doesn't return -1 on success any more. I removed the workarounds in fd_monitor.rs.

[krobelus]

which LLVM tree do I use to build the rust target?
It needs f03b100e9319 ([Cygwin] Fix global variable dll import (#121439), 2025-01-03), right? Maybe I can cherry-pick that one?
Alternatively, are there precompiled versions of the rust target available? I guess src/doc/rustc/src/platform-support/x86_64-pc-cygwin.md says no

[Berrysoft]

The patch for LLVM has been merged, so I think the latest release is OK. The latest version of nightly rustc is also OK.

Now there's no pre-built std. You can use -Zbuild-std with nightly toolchain.

[krobelus]

ah ok, cargo build -Zbuild-std --target=x86_64-pc-cygwin seems to work (on a Linux host, a Windows one can't find #include <paths.h> when compiling src/libc.c).

But it seems we need a new libc release (for libc::statfs), and update the nix crate to be compatible with that.
I wonder if anyone has done the nix changes already.

[Berrysoft]

A windows host can do that, and I even wrote a PKGBUILD for MSYS2: https://github.com/Berrysoft/fish-msys2/

But it seems we need a new libc release

Yes, we need a new libc, nix and errno release. Nix will introduce some breaking changes in the next release, so we need to change some code after that.

[krobelus]

https://github.com/Berrysoft/fish-msys2/

thanks, that's very useful

[krobelus]

merged; I squashed some of your commits (hope that's not a problem) and changed only one line I think (1isize -> 1_isize).
Will add a changelog entry once we pull the requisite dependencies

[ahaoboy]

Will update the document to add Windows installation and compilation steps?

https://github.com/fish-shell/fish-shell?tab=readme-ov-file#windows

[faho]

Once it's available upstream in a reasonable fashion - not as long as it involves installing forks, PRs, patches and special repos.