Repository navigation
docs: gemini: hard stops on review advice that silences errors - #12018
Merged
Merged
Conversation
Contributor
|
Note Gemini is unable to generate a review for this pull request due to the file types involved not being currently supported. |
The most common Gemini advice reviewers turn down is also the most harmful when followed: a guard, default or fallback that turns a loud failure into a silent wrong result. Put it first, as a hard stop, and add the other false positives that recur in the review threads: missing symbols that exist, Logger::error() treated as returning, platforms OpenROAD does not support, Bazel rule semantics, QoR-changing side remarks and advice that flips between rounds. Replace the null-handling bullet: since it landed, the share of Gemini comments mentioning null went up, not down. Refs The-OpenROAD-Project#11585 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Signed-off-by: Øyvind Harboe <oyvind.harboe@zylin.com>
oharboe
force-pushed
the
gemini-hard-stops
branch
from
November 2, 2026 05:55
56c7fbe to
ac4244c
Compare
Collaborator
Author
|
@maliberty Unrelated outage, merge? Whittles down bad advice from Gemini. |
maliberty
approved these changes
Nov 4, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Gemini review comments cost authors time to refute, and the most common advice that gets turned down is also the most harmful when followed: a null, empty or zero guard, a default or a quiet fallback that turns a loud, early failure into a silent wrong result. This makes that a hard stop at the top of
.gemini/styleguide.md, and adds the other false positives that recur.Refs #11585.
Where the rules come from
Every Gemini review thread on OpenROAD pull requests since April 2026 that got a human reply (717 of 2,905 comments) was read and the turned-down advice sorted by habit (approximate counts, a thread can count twice):
auto&on a by-value return)Logger::error()treated as returninggenrule$(@D), transitional BUILD state)The null bullet is replaced
Since the current guide landed on 2026-10-13, the share of Gemini comments that mention
null/nullptrwent up, from 20 % (487 of 2,466) to 36 % (159 of 439). That is a keyword count, not a reading of each comment, but it does not look like the bullet is working. The replacement says what we want instead: a reachable null dereference is reported with its triggering input and fixed withlogger_->error()or an assert, never a silent default. The bullet's other point, trace guards and callees before reporting, is kept in a general form.🤖 Generated with Claude Code