RFC: Support formatting only changed regions

[daniel-grumberg]

What I have in mind is using something like what highlight-changes-mode does on save to track the modified lines. We could then use the current RCS patch strategy to match diff hunks to modified regions (essentially if the diff is contained within the modified region plus some configurable context value) and only apply these hunks.

I am happy to develop this but I want to first here this would be of interest to people. Also I am relatively new to elisp so it might take some time and I might need some guidance navigating the existing source.

[raxod502]

This seems like a fine addition. I'd suggest implementing it as two different components which can then be combined together to form a complete solution:

• the ability to limit formatting to a given region (perhaps the user wants to format the active region only)
• some way of determining what regions have changed since the last save

[daniel-grumberg]

Having thought about it more (I implemented something similar to this in my own config deferring to lsp for the actual formatting), formatting just a region has to be best effort. The thing you hope for is that you just need to apply the hunks that overlap with the target region. In general this is not sound because you might need some other hunks depending on language semantics. Let me know if you are still happy with implementing the functionality in a best effort manner.

[raxod502]

I think that's perfectly reasonable---there is a failure condition, but it's the kind of thing that makes a lot of sense in hindsight (what else did you expect the software to do?), and I am okay with software that fails sometimes as long as it fails for reasons that make sense and are easy to predict.

[mohkale]

I've wanted this as well, but my desire for it is starting to wane. I'd recommend reading black#134. It provides a pretty compelling reason IMO for why this isn't supported. I'd say consistency is more valuable than any of the reasons for avoiding formatting the entire file. If you're adopting a formatter you should use it on everything.

[lassik]

In format-all, I opted to only support format-region using formatters which have official command line flags to support that feature. Otherwise we open a can of worms, as you partially sketched above.

A few formatters already do have the feature. In all of those cases, the entire file is sent to the formatter as usual. The formatter outputs the entire file with changes only in the given region.

For reference:

• Implement "format region" feature for formatters without native support lassik/emacs-format-all-the-code#17
• Implement "format region" feature for formatters with native support lassik/emacs-format-all-the-code#150

[raxod502]

I agree with that assessment. Thanks for the suggestion!

[VitalyAnkh]

Any updates on this? Wishing for this feature to be implemented 😊

[raxod502]

My current support focus is Radian, and after that I will likely be prioritizing straight.el. After that, the Apheleia backlog may be a next step, but it depends on how things develop. As always, community patches will be reviewed and merged.

[xeechou]

I implemented this feature based off format-all-region, it's a small minor mode called fmo-mode (format-modified-only). The idea there is similar to @daniel-grumberg described. Extracting diff hunks and applying format-all-region in those regions. It's hooked off before-saving-hook. Hope it will help.

[hrehfeld]

(This feature would be great)