Repository navigation
rust: list the op-v2.7.0 forward fixes as carried reth patches - #24101
Conversation
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
aakoshh
left a comment
There was a problem hiding this comment.
The PR numbers and release annotations are correct: I checked all 8 against upstream, along with the UPSTREAM-MIRROR tags in proofs.rs and config.rs. The upstream refs need qualifying (inline comment), plus one more ask for this bullet and an idea for the forward-sweep guide; the guide change can be a separate PR.
1. Note the interaction with the replay backport on v2.8.1. v2.8.1 already contains paradigmxyz/reth#27462, with its is_cancelled() checks in the plain EVM replay loops. Picking the paradigmxyz/reth#28266 backport (ethereum-optimism/reth#41) onto it replaces those loops, so the conflict must move the checks into the executor loops, as the ethereum-optimism/reth#44 port of paradigmxyz/reth#27462 did. Otherwise the replays through replay_transactions_until_with_executor, trace_block_until_with_inspector and debug_intermediateRoots lose cancellation without a compile error. One sentence in the "RPC replay through the block executor" bullet would cover it.
2. Forward-sweep guide (docs/ai/reth-update-review.md). Your reply on ethereum-optimism/reth#44 says the sweep's report stayed in session notes and that its range ended before paradigmxyz/reth#28453 and paradigmxyz/reth#28454 merged. A few additions would make the next sweep reviewable:
- Record the upstream tip that the sweep ran to, and before tagging, re-sweep from that tip to the current
upstream/main. - Give every fix in range a verdict, not only the findings: picked, LOW, unreachable at the pin (and why), reverted, evm2-only, or comes with the next base. Put the table in the fork PR, so reviewers can check the exclusions. Today step 2 only says to record the fixes that apply, which is how paradigmxyz/reth#28251, paradigmxyz/reth#28312 and the others ended up with no verdict.
- Note the first release containing each fix. That decides whether the next base bump brings it or we have to carry it, and it would have caught the "left for v2.8.1" wording.
- Prefilter mechanically before reading. For each fix/perf commit in range: the first tag containing it, whether it's already picked, whether its touched files exist at the pin, and whether
git apply --checksucceeds on the pin. That took 263 commits down to a shortlist in a few seconds. It could live as a small script next to the guide; mine is below. - A reviewer checklist for a fork pick PR:
git range-diff <upstream>^! <pick>^!for each pick. Clean picks should differ only in the trailer and hunk context; for hand-ports, check that the delta matches the commit's "Adapted" note.- Search
<pick>..upstream/mainfor reverts and follow-ups that mention each picked PR. - Rebuild the candidate list independently instead of trusting the sweep's.
Prefilter script
#!/usr/bin/env bash
# Usage: classify.sh <scratch-dir> (run from a reth clone; pr-44 = fork pick branch)
# Reads <dir>/fixes.txt ("<sha> <date> <subject> || <paths>"), writes <dir>/classified.txt.
set -u
dir="$1"
picked=$(git log --format=%B fork/op-v2.7.0..pr-44 | sed -n 's/^(cherry picked from commit \([0-9a-f]*\))$/\1/p' | cut -c1-9)
export GIT_INDEX_FILE="$dir/idx"
git read-tree pr-44
release_of() {
local h="$1" t
for t in v2.8.1 v2.9.0 v2.10.0 v2.11.0; do
if git merge-base --is-ancestor "$h" "$t"; then echo "$t"; return; fi
done
echo "main-only"
}
: > "$dir/classified.txt"
while IFS= read -r line; do
h="${line%% *}"
subj="${line#* }"; subj="${subj#* }"; subj="${subj%% || *}"
rel=$(release_of "$h")
if echo "$picked" | grep -q "^${h:0:9}"; then pk=PICKED; else pk=-; fi
tot=0; ex=0
while IFS= read -r f; do
[ -z "$f" ] && continue
tot=$((tot+1))
if git cat-file -e "pr-44:$f" 2>/dev/null; then ex=$((ex+1)); fi
done < <(git show --format= --name-only "$h" | grep -vE '^(\.github|docs|book)/|Cargo\.lock$|\.mdx?$')
if git format-patch -1 --stdout "$h" | git apply --cached --check - 2>/dev/null; then ap=clean; else ap=conflict; fi
echo "$rel|$pk|files $ex/$tot|$ap|$subj|$h" >> "$dir/classified.txt"
done < "$dir/fixes.txt"The input is git log --no-merges --cherry-pick --right-only --format='%h %cs %s' <pick-branch>...upstream/main, filtered to fix|perf|security|revert subjects. Byte-identical picks drop out of --cherry-pick as patch-equivalent, so only hand-ported picks show as PICKED.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
|
🤖 Opus 5.5 says: Both points are in 9f1b3bb: refs qualified as suggested, and the "RPC replay through the block executor" entry now says that on v2.8.0 and v2.8.1 the |
Updates the "Carried patches" list in
UPDATING-RETH.mdfor the next base bump:op-v2.7.0. Eight of them are newer than v2.8.1, so the next bump has to pick them again: from the fork branch on an alloy-evm 0.39 base, from upstream on an evm2 base.paradigmxyz/reth#27462's cancellation checks into the executor loops. Otherwise replays lose cancellation with no compile error.Requested in the reth#44 review.
🤖 Co-created with Claude Opus 5.5