Patch application fails because phab thinks reverted revision already landed

Project Infrastructure Code Review
1

When a revision is closed by a commit, which is reverted and then the revision is subsequently reopened, it seems any other revisions which depend on the first one will fail to apply.

The logic in phab seems to not detect reverted revisions properly. See for example: diff checks #125816

INFO    This diff depends on: D111509
INFO      Already landed: D111509
INFO      Will be applied:
INFO    Planning to apply in order:
INFO    https://reviews.llvm.org/D130308?id=459370
WARNING D130308#459370 commit b133adc643b4 does not exist

This is a failure applying ⚙ D130308 [clang] extend getCommonSugaredType to merge sugar nodes, which depends on ⚙ D111509 [clang] use getCommonSugar in an assortment of places, the last one having been reverted and reopened.

This is a big issue for me as I have a 10 patch deep stack right now, and it’s making pre-commit CI testing quite difficult.

I realize we are moving to GitHub PRs, and so it may seem that there is little incentive fixing the existing infrastructure. But on the other hand that is still a month away, if not more in case there are any problems rolling it out.

2

I submitted a hammer and chisel approach to fixing this: patch_diff: Stop trying to skip landed revisions by mizvekov · Pull Request #417 · google/llvm-premerge-checks · GitHub

3

Pretty sure it’s a year away, rather than a month? (see Code Review Process Update which states a tentative date of 1st October 2023).

(I don’t have the knowledge to review your proposal)

4

Oops, I stand corrected, thanks!