[naveenvenk17]

Fixes #11269.

place_macro updates the target location before checking overlaps. If that macro is already placed, the scan can report the instance as overlapping itself. This skips only the instance being moved and keeps checks against other placed macros.

The dedicated regression loads three already-placed R0 macros from a DEF. It moves the target successfully, then independently checks rejection against each of the other two macros. It does not pass an orientation argument or modify the existing rotation regression. The target is restored to PLACED before each rejection attempt so those calls reach overlap checking rather than the LOCKED-origin guard.

Tested on the GCP Linux VM on the exact updated commit:

• All 45 MPL targets and six repository checks passed with test-result caching disabled.
• The focused test matches its committed golden output.
• The same regression fails on the before-fix control with the expected self-overlap MPL-0041.
• The new test is registered in Bazel only.

Existing review history is preserved; no force-push was used. The older DCO failure still requires maintainer remediation. New commits have matching sign-offs. The missing upstream registry checksum is now fixed in d61ef5a.

[gemini-code-assist]

Code Review

This pull request prevents a macro from checking overlap with itself in findOverlappedMacros inside rtl_mp.cpp. It also updates Bazel test targets and introduces several new tests, including place_macro_right_angle_rotation. The review feedback points out that the golden file place_macro_right_angle_rotation.defok was placed in the wrong directory (testcases/), which will cause Bazel test failures, and should be moved to the parent test directory.

[gemini-code-assist]

The golden file place_macro_right_angle_rotation.defok is added under src/mpl/test/testcases/, but this diff_files call expects it in the current directory (src/mpl/test/). Additionally, the Bazel glob in src/mpl/test/BUILD (glob(["place_macro_right_angle_rotation.*"])) will not find it inside testcases/, which will cause the Bazel test to fail. To resolve this and align with the repository's conventions, please move the golden file from src/mpl/test/testcases/place_macro_right_angle_rotation.defok to src/mpl/test/place_macro_right_angle_rotation.defok.

[naveenvenk17]

Updated for current master. The golden output now comes from the existing parent test location, so there is no separate misplaced .defok file.

The regression now first marks the target macro as placed, then moves it. That exercises the self-overlap path directly. I also squashed the branch into one signed commit.

Validation on the GCP VM:

• //src/mpl/test:place_macro_right_angle_rotation-tcl_test passed three consecutive runs
• //src/mpl/test:all passed 42 of 42 tests
• clang-format-18 --dry-run --Werror and git diff --check passed

[AcKoucher]

@naveenvenk17 Thanks for fixing this!

[naveenvenk17]

Thanks, I split that setup out.

place_macro_right_angle_rotation is back to checking only the R90 placement. The new place_macro_already_placed_overlap test first marks the macro placed, then requests the same legal R90 placement without allowing overlap. That path only succeeds when the target macro is excluded from its own overlap search.

The new regression has its own expected log and DEF files, plus Bazel and CMake registration.

I rebased this on current master and validated it on the GCP VM:

• both focused placement tests passed three uncached runs
• the full //src/mpl/test:all suite passed 43 of 43 targets
• clang-format --dry-run --Werror and git diff --check passed

I also had the final test split reviewed separately before the last test run; there were no blocking findings.

[AcKoucher]

@naveenvenk17 Please, do not force-push after a review, because we lose history.

[AcKoucher]

Please, use an input DEF that does not require setting -orientation that way we keep the test case narrower.

[naveenvenk17]

You are right. I rebased and force-pushed to keep the branch current, but that removed review history. I will keep any further updates as additive commits and will not force-push this PR again.

[maliberty]

DCO needs fixing along with lint

[naveenvenk17]

The test now loads the placed R0 macros from a dedicated DEF and does not pass -orientation. It checks a valid move and separate overlaps with both neighbours. The actual run also caught the target becoming LOCKED after a successful move; the test now restores PLACED between independent attempts so it checks MPL-0041, not the OpenDB origin guard.

All 45 MPL targets and six repository checks passed on the final VM-tested commit. The before-fix control fails on self-overlap as expected. I kept the reviewed history intact and did not force-push.

DCO still flags the older commit. Could we handle that through an additive remediation or maintainer override rather than rewriting it?

[AcKoucher]

The test being added here is overly complicated. I opened #11907 with your fix preserved and a simpler, straight forward test.

[naveenvenk17]

Thanks, the shorter test makes sense. #11907 preserves the fix, so I'm closing this one in its favor.