Add a comment that says what this does.

Naming is unclear. I'm guessing 'seg' is short for 'segment' (just write 'segment' in that case), but then it's still unclear what's the difference between a chain and a segment.

actual is a bad name, as it doesn't actually (ha! punny) has any meaning here. selected_len would be a better name.

What is the difference between 'curve start/end point' and 'actual curve start/end point'?

The term bezt_chain seems to change semantics here. Earlier it was the 'chain of selected keys', and now it seems to mean 'the keys that can be deleted + surrounding keys'. Or maybe not. It's unclear, in any case.

Shouldn't this be #DEFINEd or const short'ed somewhere?

Add a comment explaining what the function does.

These comment don't add any information.

Sentence.

Sentence.

Comment doesn't add any information.

Add a comment explaining what the function does.

Make a sentence, start with capital and end with period.

I'll stop here with the comments about this; apply it to all comments where appliccable.

The function name says 'reset', this comment says 'restore', pick one and make it consistent.

You can do if (dgo == NULL) return; here, and unindent the rest of the function.

No need to create a new variable scene just to avoid dgo-> in this line.

Why would you allocate, do this check, and immediately deallocate? Why not just do the check before the allocation?

What is 'pop' in this case?

Comment doesn't add anything.

Comment doesn't add anything.

Why?

I only see a lot of allocation & copying. Where is the cleaning done? Also what is cleaning anyway? It thought this was a function to decimate curves?

Comment doesn't add anything.

Comment doesn't add anything.

It would be better to have this in decimate_exit() instead, as it's on-exit behaviour.

yes well, that comment doesn't add anything either.

Comment doesn't add anything.

Comment doesn't add anything.

Comment doesn't add anything.

Comment doesn't add anything.

What is the purpose of this? I'm not familiar with the xxxNumInput() functions, so that could be it, but it seems a bit like taking a step forward and then a step back again.

Then why is percentage a float?

I would put this in a function, and call that function instead of setting do_pose_update=true.

Comment doesn't add anything. Especially when you would rename the function to decimate_draw_status_header.

Don't describe where it is used, but describe what it means.

Don't name things context if they are not a Context. Just name it operator_context.

why?

Just write 'Only remove the start/end point of the segment if they are not the start/end point of the curve'. That removes the use of the somewhat vague term 'actual'.

Still unclear why 12 is such a good middle ground.

Shouldn't this be "the current keyframe's interpolation type"? Or are there other aspects of the keyframe that are relevant as well?

Well yes, that's obvious from the fact that nothing is happening. Always include the why in comments.

This describes what it's doing, but not why. As this is something that's not trivial to understand, the motivation needs to be included in the source code. It should also be obvious to the user that this is what will happen.

Missing tooltip.

Is a soft-max of FLT_MAX really useful here? It makes dragging the slider hard as the values will most likely be too big.

Also: period.

Make a switch with two case statements.

as we do not have a modal mode for it

Now that there is a reasonable soft limit, shouldn't both modes just have the same behaviour?

Now that I think of it (sorry for bringing this up this late), wouldn't a switch make more sense here? Then it's immediately clear that it's a selection on fcu->bezt[i].ipo only.

We discussed (face-to-face) squaring the error here in order to use the more common RMS metric, as this would also give a more linear 'feel' to the user when experimenting with different values.

As the code is now, remove_error_margin (so not squared) is assigned to error_sq_max (the name implying 'squared'), without squaring.

Rename to prepare_for_decimate_keyframes (or just prepare_for_decimate or something along those lines), as it does more than just checking. Also update the comment for this.

Probably fine to just return true; here.

ditto

Give the (currenty anonymous) enum a name, and use that here.

Pass initial values to avoid 'may be used uninitialized in this function' warnings for those with GCC < 9 ;-)

Comment doesn't add much. Better to document why the error is squared.