[htsugawa]

Why

mzTab-M export names each run's file format (ms_run[n]-format) by looking at the disk at export time (MztabFormatExport.SetRawFileMetadataDic).

• A missing .raw becomes [,, Unknown .raw format, ]. This branch dates from Open and browse an MS-DIAL 5 project whose raw data files are missing #829 (browsing without raw data); before Open and browse an MS-DIAL 5 project whose raw data files are missing #829 the export threw.
• A missing .d has become [,, Unknown .d folder format, ] since Console: AIF collision energies from mzML activation, multi-CE AIF alignment, and nearest-apex AIF seed scans #825.

So once the raw data are deleted, which the GUI now supports and the public-repository reanalysis does, a re-export loses a format MS-DIAL knew perfectly well when it processed the file.

Two clarifications:

• This is a provenance gap, not a validator error: jmzTab-M rates the term MAY.
• It is not what makes CI red; see Dispose a model's items even when its collection is cleared first #833 and Make the Master Build restore pass: regenerate the Release vendor unsupported lock files #834.

What changes

Recording the format

• New RawFileFormat enum (append-only, numbered) and AnalysisFileBean.RawFileFormat at MessagePack Key 19. Keys 0–18 are taken; 19 is free on master and on every remote branch, the 5.6 line included.
• RawFileFormatClassifier names the format from the path without reading the file. It never throws.
  • .raw: a directory is Waters, a file is Thermo.
  • .d: checked in RawDataHandler's own order — AcqData (Agilent), then analysis.tdf, analysis.tsf, analysis.baf (Bruker).
  • Other extensions: decided by extension, case-insensitive.
• The format is recorded where a file is registered:
  • GUI dataset setting (DatasetFileSettingModel);
  • both Console bean factories (AnalysisFilesParser);
  • a GUI path change (AnalysisFileBeanModel).
• Older projects learn their format lazily, the first time their raw data are found (AnalysisFileBeanModel's availability check). MarkRawDataMissing never touches it.

Export. SetRawFileMetadataDic resolves the format in this order:

1. what is on disk, as before;
2. the recorded format, when the disk cannot tell and the record fits the extension;
3. today's Unknown strings.

For every file present on disk the output is byte-identical to master. The one exception is a .d that holds both AcqData and analysis.* files: it now follows RawDataHandler's order, AcqData first. id_format and spectra_ref are unchanged.

Compatibility

• Projects written before this change read as Unknown.
• A reader without Key 19 opens projects written with it (tested with a keys-0–18 mirror class).

Tests

• Classifier: every extension and its case variants; .raw as a file, a folder and missing; .d with each marker, empty, missing and AcqData+tdf.
• AnalysisFileBean: a MessagePack round trip; an old-project fixture reads Unknown; older-reader compatibility; pinned enum numbers.
• Export with the raw data gone: each recorded format gives its PSI-MS term, checked against an allow-list of MS:1000560 children. A record that does not fit the extension is ignored, and a file on disk is described by what is on disk whatever was recorded.
• End to end: register, save, delete the raw data, reload, export, then assert the terms.
• Recording: Console parser recording; GUI model backfill, path change and the missing-file case.
• Results: whole-solution build and test in "Debug vendor unsupported", as PR Check runs them, are green apart from the GUI test that Dispose a model's items even when its collection is cleared first #833 fixes.

For the reviewers

• Key 19: please reserve it on the 5.6 line, so the next merge does not collide (we hit one such clash on ChromatogramPeakFeature Keys 54/55 in Lipid RT-order check: assign Confirmed to lipid annotations whose retention times follow the reversed-phase rules #832).
• Older builds: a 5.5 build or a 5.6 beta that re-saves a project written here drops Key 19. The format is then learned again the next time the raw data are present.
• Deliberately not in this PR: PSI-MS term corrections — WIFF2 should be MS:1003448, imzML MS:1003611, and netCDF MS:1002441 rather than EDAM. Those change output for present files and are left to a separate PR.

🤖 Generated with Claude Code