Save on eglot-rename in otherwise-unmodified files?

[jbr]

Is there a way to have Eglot save buffers when it performs an edit to them? I'd like to be able to M-x eglot-rename RET [new name] RET and have my code compile regardless of how many files contained that symbol. Currently I switch to each buffer that was modified and manually save them, which is very unergonomic.

Alternatively, could Eglot offer a eglot-rename-and-save that prompts to save any file that previously had modifications before the Eglot-rename and saves otherwise-unmodified files without interaction?

[joaotavora]

There is no such functionality. I doubt it should be provided by Eglot though. Have you tried save-some-buffers? Maybe there should be a project-save-some-buffers: that's what you should probably run before you M-x project-compile or project-save-some-buffers. in fact doesn't M-x compile already do that for you?

[joaotavora]

Of course, maybe you don't compile your project with M-x compile. That's fine, but adding a save-some-buffers into your loop, perhaps as a "before" advice, might make sense.

CCing @dgutov might have ideas here. Personally I'm stepping away from Emacs deve as much as possible btw.

[jbr]

Yeah, I don't compile in emacs. I was hoping there was a hook/config snippet to automatically save-some-buffers within the eglot workspace (not all open buffers necessarily) after eglot-rename. I'll see if I can write something for myself like that and post it here after a bit.

Thanks for eglot, @joaotavora!

[dgutov]

It seems to me like eglot-rename should start with a forced saving of related buffers - because the server process likely only has access to file contents on disk (or operates on them anyway). Someone please correct me if I'm wrong here.

Regarding the "save after" - I'm also of the same opinion as the OP, that the rename operation should end with a new state, also saved to disk. But that's a UI-level concern which we could/should make configurable, and would be a part of the alternative renaming/refactoring UI we discussed on the emacs-devel mailing list.

If the buffers are all saved before the rename, it should be fine to save them after.

[joaotavora]

because the server process likely only has access to file contents on disk (or operates on them anyway). Someone please correct me if I'm wrong here.

You're wrong :-)

Server has a view of the buffer, doesn't need to be saved. That's why "didChange" events are exchanged.

[joaotavora]

and would be a part of the alternative renaming/refactoring UI we discussed on the emacs-devel mailing list.

Yes.

I'd probably configure it myself to "not save". Or at least, "don't save the thing that's in front of me", if that was somehow an option.

Tangent, IIRC Eglot's behavior is to visit files targeted by an LSP edit if they are not visited. I think I'd also like the option to have it change those files on disk, i.e. don't visit into buffers if there isn't one already. Lots of options here.

[dgutov]

Speaking of project-save-some-buffers, there is no such command at the moment, but Emacs 28.1 added function save-some-buffers-root that can be used as the value of save-some-buffers-default-predicate.

That alters the behavior of the command globally.

EDIT: But to use this kind of functionality temporarily in a program, you could bind the variable in a let form.

[larrasket]

This is one of the features that I really miss after migrating from LSP. It does not only affect eglot-rename but any server action that involves other buffers. Even compiling in Emacs does not solve the issue: my compile command for example might just be running some .sh that seems irrelevant to Emacs' project structure, so nothing in that root is being saved after the compile command, and I always have to save-some-buffers.