On 4/1/2026 4:04 PM, Philipp Tomsich wrote:
Jeff,
So coming back to this...
As you are no doubt aware, there's been a couple bugfixes to fold-mem-offsets
ACK'd and pushed into the tree which make this no longer apply cleanly. One of
those committed patches just adjusts a comment and AFAICT the comment just
isn't important anymore after the rewrite. The other change was a bugfix to
make sure we distinguish between the mode of the MEM vs the mode of the
address. It appears that this new code handles that correctly.
So conceptually the easiest way to move forward is to revert the two changes,
apply your patch and generate a fresh diff against the tip of the trunk.
We have a few fixes (e.g., rejecting mismatches vector sizes) for this
in-flight that should all be ready for submission over the next week
or so.
How would you like to handle this? A single series against top-of-trunk?
Given that the first patch is significant rewrite, I'd tend to want that
separate to provide some bisectability. Beyond that, whatever works
best for your team is fine by me -- y'all know the code far better than
I do and are thus in the best position to know if/when breaking down the
changes makes sense.
jeff
