Fix for saving from insert mode

[omentic]

Fixes #6513, fixes #3501, addresses #1583. This somewhat changes the granularity of the undo history: but keeps the preciseness. While adding to the undo history on a save may not seem idiomatic conceptually at first: I think it's the best way to address the issue, because 1) saving a file is somewhat of a "finishing" action and 2) comparing the undo history with the current state of a file is quite a nice way to check if something has been modified and I would prefer not to mess with it.

I also added e2a8774 because I stumbled across it while tracking down this issue and thought that having a separate history entry for a raw paste before any manipulation of it would be desirable. I'd be fine rolling it back but that also propagating to history makes more sense to me.

[archseer]

#2883 (review)

[omentic]

While committing changes from insert mode is perhaps not idiomatic, this fixes a bug. Any shortcuts to save a file from insert mode are currently broken: they do not update the file modification indicator and Helix thinks the file has not been written and will not let you close it, see #6513.

This shouldn't terribly change how anyone not binding to :write uses Helix, and makes Helix more usable for anyone binding to :write.

[Xalfer]

Looking at the changes you only add doc.append_changes_to_history(view) into write_impl but not into write_all_impl.

[David-Else]

Why not just stop the ability to save in insert mode? write could check you are in normal mode, and if not return an error. It would enforce best practice. #2883 (review) explained, and archseer amplified this sentiment by quoting it.

I admit when I first started using Helix I bound ctrl-s to save and saved all the time in insert mode. This was a mistake and led me down the wrong path for using Helix, if it were not possible I never would have wasted my time doing it.

[the-mikedavis]

I think it's worth fixing this regardless of the use-case, although I'm not a huge fan of the saving-in-insert-mode use-case. Ideally we shouldn't have to think about making commit checkpoints in typeable commands like :clipboard-paste-replace - it's pretty easy to forget to append changes to history there and accidentally introduce a bug. Plus I think it might be possible to trigger the same problems as saving in insert mode with command sequences in normal mode

[pascalkuthe]

yeah I don't care for the saving in insertmode usecase that much but I do care about the sequence of commands (in normal mode) usecase

[omentic]

@the-mikedavis Feedback addressed, ready for re-review. I'll keep test driving but works well locally.

[Solsticely]

I've been daily driving this for about a week now and it fixes a bunch of stability issues i've been having :)

[the-mikedavis]

This improves the situation but I believe we need a larger change to make sure we're treating all execute_commands roughly the same. I have a patch that I'm working on locally in the-mikedavis@f7b2f77 that I will turn into a PR if I can iron out the bugs

[omentic]

@the-mikedavis any update on your end? I'll be rebasing my fork soon and would like to try out your patch, if it's finished. (though imo i think this should just be merged as-is and improved later...)

[the-mikedavis]

I pushed up the change here: https://github.com/helix-editor/helix/tree/append-changes-ensure-cursor-automatically

I made a few more changes before I pushed it so I'll try it out for a while before I turn it into a PR