bug#81294: dired-do-find-regexp doesn't find Umlauts

> Cc: [email protected], [email protected], [email protected]
> Date: Wed, 15 Jul 2026 15:19:30 +0300
> From: Eli Zaretskii <[email protected]>
> 
> > From: Johann Höchtl <[email protected]>
> > Date: Wed, 15 Jul 2026 13:33:31 +0200
> > Cc: Juri Linkov <[email protected]>, [email protected], [email protected]
> > 
> > Because I set   (set-language-environment "UTF-8") on Windows I work around grep and encoding issues like
> > so:
> > 
> > (use-package grep
> >   :defer t
> >   :init
> >   (defun my-grep-with-universal-coding (orig-fun &rest args)
> >     (let ((coding-system-for-read 'utf-8)
> >       (coding-system-for-write 'latin-1))
> >     (apply orig-fun args)))
> >   (advice-add 'rgrep :around #'my-grep-with-universal-coding))
> > 
> > This works FOR ME but is certainly brittle in many regards:
> 
> Yes, as expected.
> 
> > * works just for rgrep. Not for places like lgrep, project-find-regexp and likes, dired-do-find-regexp and likes.
> > Sure, the same logic could be used for these commands as well.
> > * makes assumptions about my input encoding and output encoding. latin-1 because of my particular windows
> > codepage (could be automatically obtained: locale-coding-system? ) and utf-8 because all my files are utf-8
> > (could as well be automatically obtained: buffer-file-coding-system? )
> > 
> > The underlying issue is that the encoded string provided as a command line parameter to find/grep set from the
> > minibuffer within emacs is utf-8 encoded, where (as it seems) MSYS2 find/grep expect this to be encoded
> > according to the system encoding. Some layer, I do not know which one, then performs an encoding translation
> > utf-8 --> (active windows codepage) and this results in a mismatch grep is no longer able to resolve.
> 
> That "some layer" is Windows APIs: internally, everything works in
> UTF-16, and is converted to/from that to the "relevant" codepage when
> a program uses non-Unicode APIs.
> 
> > Strangely, changing the codepage to chcp.com 65001, starting a new shell, starting emacs, starting a shell
> > within emacs and calling chcp, reports 1252.
> 
> That's because Emacs overrides what you do with the system's ANSI
> codepage.  To avoid that, you need to change the system-wide codepage
> setting to 65001.
> 
> It's a mess.

I think it's time to close this bug now.