bug#68377: Cannot commit parts of *vc-diff* on Microsoft Windows

[Eli Zaretskii]

> From: Sam Steingold <sds@gnu.org>
> Date: Wed, 10 Jan 2024 21:07:14 -0500
> 
> https://emacs.stackexchange.com/q/80039/795
> 
> Emacs has a beautiful feature: in the `*vc-diff*` buffer (generated by 
> [`vc-diff`](https://www.gnu.org/software/emacs/manual/html_node/emacs/Old-Revisions.html)
>  either by <kbd>C-x v =</kbd> in a file buffer or by <kbd>=</kbd> in a 
> `*vc-dir*` buffer), kill unwanted hunks with <kbd>k</kbd> (i.e., 
> [`diff-hunk-kill`](https://www.gnu.org/software/emacs/manual/html_node/emacs/Diff-Mode.html))
>  and then commit the remaining hunks with <kbd>C-x v v</kbd> (i.e., 
> [`vc-next-action`](https://www.gnu.org/software/emacs/manual/html_node/emacs/Basic-VC-Editing.html)).
>  If you don't know about it -- try it, it is incredibly useful!

(I must say that formatting of this report is AWFUL!  How can a human
being be expected to read and understand it it??  PLEASE try to make
bug reports more readable in the future.)

> Alas, it does _not_ work for me on windows: after <kbd>C-c C-c</kbd> (i.e., 
> [`log-edit-done`](https://www.gnu.org/software/emacs/manual/html_node/emacs/Log-Buffer.html))
>  in `*vc-log*`, I get 
> > vc-do-command: Failed (status 1): 
> > c:/Users/ssteingold/AppData/Local/Git/bin/git.exe --no-pager apply --cached 
> > c:/msys64/tmp/git-patchPbbWgJ
> 
> and `*vc*` buffer is
> ```
> error: patch failed: fxpy/model.py:1071
> error: fxpy/model.py: patch does not apply
> ```
> 
> ***Additionally***: 
> 
> 1. Software:
>    * [`GNU Emacs 29.1 (build 2, x86_64-w64-mingw32) of 
> 2023-08-02`](https://ftp.gnu.org/gnu/emacs/windows/emacs-29/)
>    * [`git version 2.42.0.windows.2`](https://gitforwindows.org/)
> 1. *normal* commits work just **fine**, using `vc-next-action` from either 
> file buffer or `*vc-dir*`.
> 1. All my files have unix line terminators (`LF`) as indicated in the mode 
> line as `(Unix)`
> 1. I set `autocrlf` to `false` globally
> 1. I have `(prefer-coding-system 'utf-8-unix)` in `.emacs`
> 1. `vc-coding-system-inherit-eol` is `t`, I did not touch it
> 1. When I add `"-v"` to `git apply` invocation in `vc-git-checkin` 
> (`vc-git.el`), `*vc*` buffer contains `error: while searching for:.....` and 
> every line in `.....` is terminated with `?` - which seems to indicate 
> line-ending problems... (!!!)
> 
> ***Q***: how do I fix this?
> (This is a work box, I cannot send email from Emacs, so I cannot even file an 
> Emacs bug report!)
> 
> # work-around 1
> 
> Add 
> ```
> (setq coding-system-for-write 'utf-8-unix)
> ```
> inside `(with-temp-file patch-file ...)` in `vc-git-checkin`

The code in question already does:

        (let ((patch-file (make-nearby-temp-file "git-patch"))
              ;; Temporarily countermand the let-binding at the
              ;; beginning of this function.
              (coding-system-for-write
               (coding-system-change-eol-conversion
                ;; On DOS/Windows, it is important for the patch file
                ;; to have the Unix EOL format, because Git expects
                ;; that, even on Windows.
                (or pcsw vc-git-commits-coding-system) 'unix)))
          (with-temp-file patch-file
            (insert vc-git-patch-string))

So coding-system-for-write should already be -unix when with-temp-file
runs.  I guess there's more here than meets the eye, or maybe I'm
missing something.

> # work-around 2
> 
> Add
> ```
> ;; replace `undecided`:
> (setf (cadr (assoc "" file-coding-system-alist)) 'utf-8-unix)
> ```
> to `.emacs`

That can't be right.

> # Emacs bug
> 
> `vc-git-checkin` should respect `vc-coding-system-inherit-eol` and set 
> [`coding-system-for-write`](https://www.gnu.org/software/emacs/manual/html_node/elisp/Specifying-Coding-Systems.html)
>  when writing the patch.

This variable is for _reading_ diffs from a VC command, not about
_writing_ them.  So it is not appropriate in this case, and the diffs
shown in the *vc-diff* buffer have correct EOLs.

> Also, it does not seem reasonable that 
> [`prefer-coding-system`](https://www.gnu.org/software/emacs/manual/html_node/emacs/Output-Coding.html)
>  does not affect 
> [`file-coding-system-alist`](https://www.gnu.org/software/emacs/manual/html_node/elisp/Default-Coding-Systems.html).

Using prefer-coding-system to make utf-8 the preferred encoding is
wrong on MS-Windows in the first place: UTF-8 is not yet supported by
Windows well enough to allow that.  In particular, encoding program
arguments in UTF-8 will cause failures in Emacs commands that run
subprocesses.