Merge three versions of a text file

How to use this tool

1. Open or paste the common ancestor, local version and remote version in their separate panes.
2. Click Run to read the active sources within the complete task deadline.
3. Review the merged output and each conflicting base/local/remote region.
4. Choose each resolution or edit custom text; changing any source clears stale decisions.
5. Save merged.txt when resolved, or conflicted.txt while unresolved, plus merge-report.json.
6. Manually import the reviewed saved result into your original editor or workflow; keep the originals and merge report.

Supported inputs and limits

Provide base, local and remote independently. Each strict UTF-8 source allows 1 MiB and 4,000 exact line tokens. Base lines × (local lines + remote lines) allows 4,000,000 comparison cells.

Fixed node-diff3 3.2.1 Hunt–McIlroy line merge. CRLF, CR, LF and the missing terminal newline remain part of each line token. Open files to preserve original line endings; textarea editing may normalize endings.

Identical changes appear once. Unresolved conflicts retain LOCAL/BASE/REMOTE markers and are labeled unresolved; complete resolution downloads merged.txt. Reports and result text together allow 8 MiB.

Each conflict supports local, base, remote, local then remote, remote then local, or custom UTF-8 text up to 1 MiB. Compact resolution input allows 8 MiB before cloning; active choices also remain subject to the combined output limit. Unresolved is the default. Choose output BOM from base/local/remote or none; base is the default.

Conflict markers use LF. A separator LF is inserted when needed after an unterminated fragment and recorded by UTF-16 offset. No syntax-aware merge or automatic repository changes.

Each role uses either its pasted text or exactly one selected UTF-8 file. Empty text and genuine zero-byte files are valid merge sources; an invalid active file never falls back to hidden pasted text. Filename labels allow 512 UTF-8 bytes.

One leading BOM is removed and recorded, including pasted input. Invalid UTF-8, unpaired Unicode surrogates are refused. Inputs stay in this browser.

A fixed 10-second whole-task deadline or cancellation returns no partial export. Rerun after correcting the input; a fresh Worker starts each job.

File selection reads metadata only. The entire fixed 10-second period begins at Run and includes engine loading, input checks, reading, import and merging. Any cancelled or refused request publishes no partial copy or download.

Conflict IDs must refer to the current output; nonexistent or inactive IDs and custom text attached to another choice are refused. At most 12,000 IDs can arise from the three 4,000-line roles. Changing any source cancels active work and clears prior decisions and results.

The fourth pane is a bounded output preview. Copy retains the entire merged text; merge-report.json retains all blocks, conflicts, provenance and decisions. The complete copy and download both preserve the chosen leading BOM and have identical UTF-8 bytes; exact bytes/SHA-256 are reported.

This workbench accepts manually exported sources and produces a file for manual import. It does not invoke a Git CLI/mergetool, change a repository, validate code semantics or automatically mark a conflict resolved. Contents and names stay in bounded memory in this tab.