test: differential replacement/split parity suite; fix DFA_UNROLLED findBoundsFrom match truncation

[jbachorik]

What does this PR do?

Adds a differential parity suite pinning the replacement/collection ops of ReggieMatcher — replaceAll(String,String), replaceFirst, functional replaceAll, split(String,int) — against java.util.regex as the oracle (curated + 400-case seeded randomized inputs), and fixes a real bug the suite caught: the DFA_UNROLLED generated findBoundsFrom greedy scan skipped its lastAcceptingPos record at "skippable" accepting states, which truncated matches and corrupted split/replaceAll results (e.g. cbb|b{0,2}b on "bb" reported [0,1) instead of [0,2)).

The optimization was statically unsound — any accepting state can be the scan's terminal state, and a terminal skippable state never records — so it is removed entirely: every accepting state records unconditionally. The scan now also delegates to the anchor-aware findLongestMatchEnd path when any state carries acceptance anchor conditions, and the unreachable assertion-handling path (dead since delegation already covers all assertion-bearing DFAs) is deleted.

Six intentional divergences from the JDK are documented in the suite and pinned on both engines (JDK drift fails the test too): no backslash-escape processing in replacement strings, $$ accepted as literal $, out-of-range/trailing $ group references appended literally instead of throwing, and functional replaceAll using the replacer result verbatim.

Motivation

The profiling-backend wave-3 migration relies on these ops per operation; the engine-level semantic contract had no differential coverage (only backend per-site parity), which is why the findBoundsFrom truncation survived.

Related Issue(s)

Draft source: .issue-drafts/2-replacement-ops-parity.md. Note the draft's premise that reggie treats $ literally in replaceAll was wrong (verified empirically) — reggie expands $n; the real divergences are the ones documented in the suite.

Change Type

• Bug fix
• Test improvement

Checklist

• I have read the CONTRIBUTING.md guidelines
• All existing tests pass (./gradlew build)
• I have added tests for my changes
• I have updated documentation (if applicable)
• My commits are signed

Performance Impact

JMH A/B (pre 967e525 vs post, 3 interleaved runs, medians): replaceAll(b{0,2}b) −31%, replaceAll(b{0,4}bc?) −16%, split(b{0,2}b) −35%; control pattern within noise. The pre-fix truncation caused more, shorter match scans; the unconditional 2-instruction record is free.

Additional Notes

Full in-depth review already performed by independent fresh-context reviewers (bytecode-trace verified removal, no sibling generator carries the pattern, all 6 divergences confirmed against JDK 26). Reviewer-flagged follow-up: none open in this diff. Gate evidence: :reggie-runtime:test + :reggie-integration-tests:test = 3281 tests, 0 failures; spotlessCheck green.

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 85.1%. Comparing base (967e525) to head (9c1c849).
⚠️ Report is 1 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff            @@
##              main    #145     +/-   ##
=========================================
+ Coverage     84.9%   85.1%   +0.2%     
  Complexity       1       1             
=========================================
  Files          163     163             
  Lines        48981   48847    -134     
  Branches      7000    6970     -30     
=========================================
- Hits         41594   41588      -6     
+ Misses        5378    5258    -120     
+ Partials      2009    2001      -8     

Files with missing lines	Coverage Δ
.../codegen/codegen/DFAUnrolledBytecodeGenerator.java	95.6% <100.0%> (+5.2%)	⬆️

... and 7 files with indirect coverage changes

---

Continue to review full report in Codecov by Harness.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 967e525...9c1c849. Read the comment docs.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-05T10:05:52.551944Z	9c1c849	Draft marked ready

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[datadog-datadog-prod-us1]

More details

Unconditional accepting-state recording and delegation of assertion/anchor-bearing DFAs to the existing longest-end path address the truncation without a reportable regression.

Was this helpful? React 👍 or 👎

_{Open Bits AI session}

_{🤖 Bits Code Review · Commit 9c1c849 · @DataDog review to ask questions}