[only21mil]

Problem

Very large diffs can show +1000k just below one million additions, then +1m at exactly one million. Counts from 999,500 through 999,999 hit this boundary. The same rounding error produces 1000m just below one billion. Additions and deletions use the same formatter.

Change

Select the million and billion tiers at their rounding boundaries. Keep the existing suffixes and decimal precision. A focused regression checks the count immediately below each boundary and the first count that rounds up.

Scope and approval

This is a small fix for an obvious numeric display error under the focused-bug exception in CONTRIBUTING.md. No issue is linked. It changes only the existing compact diff label used by the web client and desktop renderer.

Verification

Node 26.12.1 and pnpm 12.14.0. Base 5ffe045931, candidate 8b29043505.

• pnpm --dir apps/web exec vp test run src/components/chat/DiffStatLabel.test.ts --project unit: fails with the base thresholds, reporting expected 1m, received 1000k; passes with the candidate.
• pnpm exec vp lint apps/web/src/components/chat/DiffStatLabel.tsx apps/web/src/components/chat/DiffStatLabel.test.ts: no warnings or errors.
• pnpm exec vp fmt --check apps/web/src/components/chat/DiffStatLabel.tsx apps/web/src/components/chat/DiffStatLabel.test.ts: passes.
• pnpm exec vp run --filter @t3tools/web typecheck: passes, with existing Effect suggestions in unrelated files.
• pnpm exec vp run --filter t3 build: production builds pass for base and candidate.
• Opened both production clients in Chrome 156 on Linux with separate server state. In a fresh demo repository, staged generated.txt with 999,500 added lines. Opened the right panel, then Diff. The summary and file row change from +1000k to +1m. Servers and builds ran on a separate build machine through a loopback tunnel. No provider turn was needed.

Not checked: a packaged desktop app, Windows, macOS client UI, mobile, and a billion-line repository. The billion boundary has deterministic formatter coverage. No broad suite was run.

Base 5ffe045: the 999,500-line diff reads +1000k in the summary and file row.

Candidate 8b29043: the same 999,500-line diff reads +1m in the summary and file row.

The final regression test fails with the base thresholds: expected 1m, received 1000k.

The final regression test passes; changed-file lint and formatting also pass.

Prior PRs: #17539 by luke2x fixed the same 1000k rollover in the context window formatter by ending the k branch at 999,500; this applies that change to the diff label and the billion tier. #17948 by maria-rcks introduced locale grouping for full diff counts and explicitly left compact labels unchanged. This keeps that distinction. The token-count rounding PRs #15599 by ValeraZSD, #21594 by MarMarLabs, and closed #20255 by kvnloo concern separate formatters; no code is reused.

Production fix, tests, PR description and images written with GPT-6.1 Sol in T3 Code Nightly.

🤖 Generated with GPT-6.1 Sol in T3 Code Nightly

[coderabbitai]

Warning

Review limit reached

Only developers with an assigned seat can use this organization's usage-based review budget, and seats here are assigned manually. Ask an admin to assign a seat, or change the review continuation mode in Billing.

Next included review available in 6 minutes.

Check out review usage here.

View limit details

Limit details: You’ve used all 10 included reviews currently available.

Learn how review limits work.

Review configuration:

Run configuration

• Source mode: Auto
• Repository: pingdotgg/t3code
• Branch: main
• File: .coderabbit.config.ts
• Commit: d487078
• Layers: Repository TS config
• Review profile: CHILL
• Plan: Advanced
• Run ID: f6086ece-bce5-422b-a4a5-248e237dbed8

Commits

Reviewing files that changed from the base of the PR and between 8b29043 and e76886c.

Files selected for processing: 2

• apps/web/src/components/chat/DiffStatLabel.test.ts
• apps/web/src/components/chat/DiffStatLabel.tsx

No actionable comments were generated in the recent review.

ℹ️ Recent review info

⚙️ Run configuration

• Source mode: Auto
• Repository: pingdotgg/t3code
• Branch: main
• File: .coderabbit.config.ts
• Commit: 60b35a1
• Layers: Repository TS config
• Review profile: CHILL
• Plan: Advanced
• Run ID: 8abcd874-e02d-47f5-8a3b-5c109797fe5d

📥 Commits

Reviewing files that changed from the base of the PR and between 60b35a1 and 8b29043.

📒 Files selected for processing: 2

• apps/web/src/components/chat/DiffStatLabel.test.ts
• apps/web/src/components/chat/DiffStatLabel.tsx

Included review availability: This review used your included allowance. Your plan provides up to 10 included reviews per hour; 9 remain after this review.

Walkthrough

formatCompactDiffCount is now exported. Its k and m ranges end at revised thresholds, and tests cover values just below and at each promotion point.

Changes: Compact diff counts

Layer / File(s)	Summary
Formatter thresholds and tests apps/web/src/components/chat/DiffStatLabel.tsx, apps/web/src/components/chat/DiffStatLabel.test.ts	The formatter is exported. Counts promote from k to m at 999,500 and from m to b at 999,500,000. Tests check values below and at both thresholds.

Priority: Low

Estimated code review effort: 2 (Simple) | ~8 minutes

Change: Bug fix

Merge Risk: Minimal · up to 8b290

The focused formatting change is mergeable with no unresolved correctness or availability risk.

Pre-merge checks: 4

Check name	Status	Explanation
Title check		The title is concise, uses the repository's conventional format, and clearly describes the primary fix for rounded diff counts displaying as 1000k instead of 1m.
Description check		The description includes the required Problem, Change, Scope and approval, and Verification sections. It explains the bug, implementation, focused-bug scope rationale, test results, manual UI verifica…
Linked Issues check		Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check		Check skipped because no linked issues were found for this pull request.

Full details: Description check

Explanation

The description includes the required Problem, Change, Scope and approval, and Verification sections. It explains the bug, implementation, focused-bug scope rationale, test results, manual UI verification, screenshots, and limitations.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create a new PR

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

_{Comment @coderabbitai help to get the list of available commands.}

[maria-rcks]

Note

Written by gpt-6.1-sol on behalf of Maria

thanks @only21mil. your two threshold changes and four boundary assertions are already the smallest correct fix, so i kept them intact and merged github/main (d487078) on top. no pieces were dropped or moved. final head: e76886c.

verification:

• github ci: 21 passed, 7 skipped, no failures or pending checks; mergeable and clean. human review remains. coderabbit passed with a rate-limit notice, so its last full review covers the original fix; both independent takeover reviews cover the final head.
• reproduced the existing regression against current main thresholds on blacksmith: expected 1m, received 1000k.
• blacksmith vp run check:changed passed: formatting, lint, both knip passes, affected typechecks, and the existing boundary test. the remote snapshot used the exact main commit as its comparison base.
• exercised the actual client/server git diff path in system chrome with the same staged 999,500-line file. both the summary and file row change from +1000k to +1m. before uses current main formatter; after uses the committed fix. no provider turn was needed.
• thirteen direct browser-runtime formatter checks passed, from 0 through both promotion windows and exact million/billion counts.
• two independent gpt-6.1-sol reviewers approved this exact head with no findings.

unverified: packaged desktop, mobile, macos/windows, and a billion-line repository. the billion boundary is covered by the deterministic formatter test. no new test files were added during takeover.

[maria-rcks]

Note

Written by gpt-6.1-sol on behalf of Maria

Verified in a real dev environment

verified as-is at e76886c8867638d569d1e46ce0ae51ea0be057d8 on this linux boat vm. no source changes or pushes.

• reproduced on freshly fetched github/main (36aea583bb) in /home/user/main, installed with vp i, and ran the real dev app. stopped main before starting the exact pr head, also installed with vp i, with isolated empty dev state.
• drove both clients with the repo's playwright and /usr/bin/google-chrome. added a project pointing at a real git repository through the ui. submitted real codex gpt-6.1-sol turns through the composer to generate and stage exactly 999,500 lines in generated.txt and 12 in neighbor.txt. no mocks, fixtures, replay provider, or demo data.
• main shows +1000k; the pr shows +1m for matching data in both the chat changed-files summary/file row and the right-panel diff summary/file row. the ordinary file stays +12; the full unabridged total stays +999,512.
• a further real codex turn committed the disposable verification baseline and staged both deletions. chat and diff panel show -1m and -12. expanded and collapsed both files; the ordinary file renders all 12 deleted lines, and the large file renders its partial preview. the recording shows these interactions.
• blacksmith installed dependencies and ran cd apps/web && vp test run src/components/chat/DiffStatLabel.test.ts --project unit: 1 test passed, including the million and billion promotion boundaries. latest github checks: 21 success, 7 skipped, no failures or pending checks.

runtime transcript, independently confirmed against the real repository:

$ git diff --cached --numstat  # additions turn, on both versions
999500  0       generated.txt
12      0       neighbor.txt
$ git diff --cached --shortstat
2 files changed, 999512 insertions(+)
$ git diff --cached --numstat  # deletion turn, pr head
0       999500  generated.txt
0       12      neighbor.txt

Not verified

packaged electron desktop, native ios/android, macos/windows, remote relay/tunnel connections, and other providers were not exercised. a billion-line repository was not created; that boundary was exercised by the existing deterministic formatter test. no performance claim or broad test-suite claim is made. screenshots were inspected at full resolution; the trimmed recording was inspected, and uploaded playback and seeking passed in t3's browser player.