fix: preserve the full EXIF block in exports instead of the sidecar subset (#1165)

[Serraniel]

Description

Exporting with Save with Metadata enabled silently drops most of the EXIF data. From a Fujifilm X-T50 JPEG with 78 EXIF tags, only 15 survive the export: the whole GPS block, DateTime/OffsetTime*/SubSecTime*, Flash, ExposureBiasValue, ApertureValue, ShutterSpeedValue, the MakerNote, body/lens serial numbers, LensSpecification, the Interop IFD and more are gone.

Root cause: write_image_with_metadata sources tags from three branches (.rrdata sidecar → kamadak-exif → rawler). Opening an image triggers persist_exif_if_missing, which creates the sidecar — so from then on the sidecar branch always wins. The sidecar only stores a display-formatted subset of the tags (e.g. "ApertureValue": "7.4 EV", MakerNote as a truncated hex dump), so it cannot reconstruct the original EXIF, and the kamadak/rawler branches (including their GPS handling) are never reached.

Fix: for non-RAW sources, copy the complete EXIF block (IFD0 + Exif/GPS/Interop sub-IFDs, including MakerNote and unknown tags) from the source file via little_exif, instead of reconstructing a subset:

• The sidecar is still consulted afterwards, but only for the fields that are editable in the metadata panel (Artist, Copyright, ImageDescription, UserComment) — user edits override the source values, and a cleared field stays cleared in the export.
• strip_gps=true still removes the entire GPS IFD (all GPS* tags, not just lat/lon/alt).
• Orientation is reset to 1 as before (rotation is baked into the exported pixels), and ExifImageWidth/ExifImageHeight are now set to the actual export dimensions instead of carrying the stale source values.
• IFD1 is not copied, since it contains the original's embedded thumbnail which no longer matches the exported pixels.
• RAW sources are unchanged: they keep the existing sidecar/rawler reconstruction (a full block copy is not possible there, since little_exif cannot parse RAW containers).

Fixes #1165 (the GPS part of it as a side effect, see below). The closed duplicates #503, #427 and #401 report the same symptom.

Relationship to other PRs:

• fix gps metadata being stripped from exports (#1165) #1207 fixes the GPS subset of the same root cause by unconditionally re-reading GPS from the source file. This PR is complementary: it covers all tags for non-RAW sources, while fix gps metadata being stripped from exports (#1165) #1207 additionally restores GPS for RAW-with-sidecar exports, which this PR intentionally leaves untouched. Both can land: whichever merges second just needs a trivial rebase of the (textually adjacent) changes in exif_processing.rs. The regression tests here follow the pattern introduced in fix gps metadata being stripped from exports (#1165) #1207.
• fix: preserve EXIF timezone offset in JPEG export #1246 re-attaches OffsetTime* to JPEG exports; the export half of that PR is covered by the full-block copy for non-RAW sources (its get_creation_date_from_path timezone fix is independent and still valuable).

Type of Change

• Bug fix
• New feature
• Breaking change
• Performance improvement
• Code refactoring
• Documentation update
• UI/UX improvement
• Build/CI or Dependency update

Changes Made

• exif_processing::write_image_with_metadata: copy the full EXIF block from non-RAW source files (new copy_full_exif_from_source), skipping the GPS IFD when strip_gps is set and skipping IFD1; fall back to the previous sidecar/kamadak/rawler branches when the source has no parseable EXIF or is a RAW file.
• New apply_sidecar_field_overrides: the four panel-editable fields are applied from the sidecar on top of the copied block.
• Set ExifImageWidth/ExifImageHeight to the actual dimensions of the exported image.
• Regression tests: full tag preservation (incl. MakerNote byte-identity) with a sidecar present, complete GPS removal with strip_gps=true, sidecar edits overriding/clearing source values, orientation reset + dimension tag update.
• Separate commit for the WebP half of FEATURE: Export JPEG XL & WepB with EXIF metadata #1322: little_exif converts the exporter's simple-format WebP to the extended format and attaches an EXIF chunk, so WebP EXIF only needed a file-type mapping (tested with both the lossy VP8 output of the export pipeline and lossless VP8L).

Screenshots/Videos

Before/after on a real Fujifilm X-T50 JPEG, exported with Save with Metadata ON and Remove GPS Data OFF after the image had been opened once (sidecar present). Tag counts from exiftool -a -G1 -s, excluding the ExifTool/System/File/Composite/JFIF pseudo-groups:

	tags	missing vs. original
original DSCF0563.JPG	155	—
RapidRAW 1.5.8 export	14	GPS block, MakerNote (all 70+ Fujifilm tags), DateTime, OffsetTime*, SubSecTime*, Flash, ApertureValue, ShutterSpeedValue, ExposureBiasValue, body/lens serials, Interop IFD, PrintIM, ...
this PR	141	only IFD1 (stale embedded thumbnail, intentionally dropped), APP2 FlashPix preview + XMP Rating (separate segments, also dropped by 1.5.8, out of scope)

A tag-by-tag value diff of original vs. new export shows exactly the intended differences: Orientation (reset, rotation baked in), Software (RapidRAW), ExifImageWidth/Height (now the real export dimensions). With Remove GPS Data ON the export contains zero GPS* tags. exiftool parses the copied Fujifilm MakerNote in the export without errors, so the offsets survive the copy.

Testing

• I have tested these changes locally and confirmed that they work as expected without issues

cargo fmt --check, cargo clippy --lib --tests --no-deps and cargo test pass. Verified against a real Fujifilm X-T50 JPEG (see table above): tag-by-tag exiftool diff of original vs. export matches except for the intentional differences (Orientation, pixel dimensions, Software, thumbnail).

Test Configuration:

• OS: Arch Linux (RapidRAW 1.5.8 via rapidraw-bin AUR for the "before" state)
• Hardware: AMD Ryzen 7 5700X, Radeon RX 6800 XT

Checklist

• My code follows the project's code style
• I haven't added unnecessary AI-generated code comments
• My changes generate no new warnings or errors

Additional Notes

• MakerNote is copied as an opaque blob. For self-contained MakerNotes (e.g. Fujifilm, whose offsets are relative to the MakerNote itself) this round-trips perfectly; MakerNotes that use absolute file offsets may have dangling offsets after the copy — still strictly better than dropping the MakerNote entirely, and the same trade-off most non-exiftool writers make.
• exiftool -validate reports "entries out of order" notices on the exports because little_exif appends the ExifOffset/GPSInfo/InteropOffset pointer tags after the (otherwise sorted) IFD entries at encode time. That's a pre-existing quirk of the little_exif encoder (it just wasn't visible before because so few tags survived), readers are unaffected, and it would be fixed upstream, not here.
• RAW-with-sidecar exports still only get the reconstructed subset (plus GPS once fix gps metadata being stripped from exports (#1165) #1207 lands). A follow-up could extract the full EXIF from the embedded preview JPEG.

AI Disclaimer:

Please state the involvement of AI in this PR:

• This PR is entirely AI-generated
• This PR is AI-generated but guided by a human
• This PR was handwritten with AI assistance (spell check, logic suggestions, error resolving)
• This PR contains only blood, sweat, and coffee (AI-free)

[CyberTimon]

Thanks for the PR!
Before I review it, remove all code comments and tests.

[r3nor]

I think this feature is important for JPEGXL and other formats, since currently they do not support metadata in the file...