[PATCH] D95399: [flang][fir] Upstream FIR dialect changes.

[Eric Schweitz via Phabricator via llvm-commits]

schweitz added a comment.

As explained previously, the original 4 //reviews// (now 5) were subparts of the same comprehensive change set. That comprehensive patch is a coherent, logical set of changes to upstream the code generation pass that converts FIR to LLVM IR. It is complete, buildable, testable, comes with a raft of new tests, etc. It is big. Very big. Upstreaming being blocked for months has consequences.

This large change set was broken into multiple reviews in response to the feedback on how to proceed with the upstreaming process. Somewhat smaller chunks were suggested as potentially helpful. Again, this set of changes was therefore broken into 4 (now 5) logical subgroups.

It was then pointed out here in phabricator that the mutual dependence (circular references) between these subsections/reviews meant that each review did not build because of its dependences on the other subparts of the change set. That was true.

To address this issue, it required that all circular references be eliminated and that each subpart of these changes be dependent only on itself and its ancestors (not anything in the descendant changes). That was accomplished by adding a minimal number of "noise" changes to kneecap and/or scaffold out the subparts in a way that they would build without circular dependences. Of course, these process directed changes are cosmetic and confusing, which is why this was not originally presented that way.

If the reviewers are inclined, these sub-reviews can be abandoned and a single very large review, squashing the process noise, can replace them.

In the end, this change set (the 5 linked reviews) is one big pie, whether it is reviewed as a whole pie or by the bite.


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D95399/new/

https://reviews.llvm.org/D95399