Bound Global Search submatch expansion (APP-5251)

[warp-agent-staging]

Description

Global Search now expands at most 100 submatches from one matched line and shares each final owned 4-KiB stored text window across retained rows. A shared 20,000-row model budget spans local and remote sources, while remote results expand and flush incrementally instead of building a large intermediate vector.

Per-line and total-budget truncation now reach the existing partial-results warning and telemetry. Column calculation, leading-whitespace trimming, stored-window offset remapping, and highlighting remain accurate.

Evidence: APP-5251.

Linked Issue

• The linked issue is labeled ready-to-spec or ready-to-implement. APP-5251 is tracked in Linear and is In Progress.
• This is a non-visual model-layer memory fix, so screenshots and video do not apply.

Testing

• cargo nextest run --manifest-path Cargo.toml -p warp -E 'test(/global_search/)' — 23 passed.

• cargo clippy --manifest-path Cargo.toml -p warp --all-targets --tests -- -D warnings — passed.

• ./script/format — completed.

• ./script/presubmit — formatting and configured Clippy stages passed; the workspace test build could not complete because the 63-GB sandbox filesystem filled while writing rustc's query cache.

• I have manually tested my changes locally with ./script/run. The deterministic model behavior is covered by focused unit tests.

Agent Mode

• Warp Agent Mode - This PR was created via Warp's AI Agent Mode

CHANGELOG-BUG-FIX: Prevented excessive Global Search memory use when a pattern has many matches on one line.

[warp-agent-staging]

This PR was generated with Warp.

Comment @warp-staging-factory on this PR to send it follow-up work.

 

[warp-agent-staging]

ticket_id: APP-5251 · ticket_source: linear

This PR is now stacked directly on #15785, so its diff contains only the bounded submatch expansion, shared stored-window ownership, and model-budget changes. #15785 remains the base dependency.

Responding as wilson: Open session · View in factory