View Markdown source ↗

Patch file encoding and recovery

The pinned source is TortoiseGit 7338078f8ddd924b8cddee35f512f2286072136d, Git/GitPatch.cpp::PatchFile and TortoiseMerge's recorded-base PatchFile path. Git patch application retains the original patch bytes. File panes decode the recorded base, working file and patched file before comparison/merging, as the source's file loaders do.

The native review captures Use UTF-8 and Legacy text encoding when it loads. The same immutable policy now reaches initial row inspection, clean file preview, recorded-base recovery, embedded three-way merge and automatic per-file recovery. Changing preferences while that review is open does not reinterpret its captured files. Explicit Refresh captures a new policy. Ordinary automatic refresh and post-save bookkeeping retain the accepted review's policy.

Connected behavior

Legacy text no longer reaches the binary chooser merely because the generic Unicode-only decoder could not read it. Source/current/patched panes retain their individual codecs; valid Unicode input still takes precedence over a legacy fallback. Direct patch application keeps its raw-byte behavior, while automatic three-way recovery writes the working pane's retained codec.

An unrepresentable automatic merge stops before writing and reports that manual preview is required. Preview stays available, opening the combined text with its original output codec. Choosing a representable output format, such as UTF-8, allows an explicit Save. File bytes, mode, stale-file checks, index and HEAD guards remain separate from display decoding.

The legacy fallback retains the existing binary boundary for NUL-containing payloads unless the prior Unicode detector recognizes them. The source detector's permissive BOM-less UTF-16 guess cannot by itself promote those patch payloads into editable text. Binary, symbolic-link and deletion choices retain their existing explicit recovery flows.

Evidence and remaining work

The checkpoint records 35 distinct focused Core cases across retained and affected checks. New real-Git cases cover Windows-1251 and Shift-JIS recorded-base previews and saves, automatic recovery that preserves an independent local edit, an unrepresentable mixed-encoding result resolved through UTF-8, and preservation of binary classification.

Eight grouped native workflows passed with system and bundled Git: the new codec family and retained recovery, special-file and native file-panel families. They exercise the production embedded editor, direct/manual/automatic application, changed preferences, error-to-Preview continuation, explicit UTF-8 Save and exact working bytes, modes, HEAD and raw index preservation. Both unaltered native captures below were inspected. They show the existing inactive-toolbar fading; complete toolbar/physical appearance acceptance remains open.

The Original patch text tab and serial-import/unified-diff legacy display now have a separate connected migration. The original file-pane/recovery checkpoint does not certify that raw-patch presentation. Wider filename alias handling, all codepage mappings, physical gestures/accessibility and signed sandbox/App Store acceptance also remain open. No whole-app or complete hosted CI claim follows.