It would be nice, yeah, but also rather complicated with tons of edge cases where it won't work:

• Think of a pattern where there's an Oxx that needs to be updated on the first row, but the last played row is actually not the previous pattern's last row because there was a Cxx effect in the middle of that pattern
• Or if that previous pattern is used in more than one place with a different pattern following up each time
• Or just a pattern with an offset sweep going from O00 to OFF from the first to last row, with no empty rows inbetween where SAx could be inserted.

It's pretty complex. :\

I don't expect it to actually insert SAx into the pattern, just changing how values >FFh will be calculated. And I'd imagine the author would be knowledgeable enough to know which parts the offsets are going to be affected (like in your offset sweep example, in which case it would be pretty obvious).

I had a similar idea a while back: The ability to place cue points on your sample, which a modified offset command can use instead of raw sample numbers (cue point offset). This way even if the sample is edited or replaced, as long as the cue points are in the appropriate places you don’t need to rewrite any commands you already typed in your pattern.

This solution proposed by fuzion_mixer might be easier to implement, but I do believe cue point offsets have potential too.

Huh? I don't see how what you describe differs from the existing cue points feature. Can you elaborate?