https://bugs.kde.org/show_bug.cgi?id=528171
--- Comment #5 from [email protected] --- Hi Andy, Good question, and yes, what you describe is a valid way to get "synchronization through the files". Let me detail how to set it up correctly and where the limits are. *The principle: if all meaningful data is written to the files' metadata (XMP/EXIF/IPTC) on PC1, then the files themselves become the synchronization medium. PC2 sees the modified file, re-reads its metadata, and updates its local database accordingly. You do not depend on a shared database at all. This works, provided a few settings are identical on both machines. * Settings you need on both PCs: write to files, not only to the database: Settings → Configure digiKam → Metadata → Editing, with "Write metadata to files" (XMP sidecar or embedded, but use the same choice on both machines). On PC1 also keep "Update metadata when files are modified", on PC2 enable "Re-read metadata when files are modified". * Monitor albums: as you assumed, keep it enabled on PC2 so the file change triggers the re-read. Same metadata template and same options on both machines (rating, tags, coordinates, face regions in the write list). Any divergence here means PC2 silently ignores fields that PC1 writes. * Caveats to be aware of: network shares and monitoring: "Monitor albums" relies on filesystem change notifications. Over SMB/NAS this is not always reliable, depending on the NAS and Windows file-sharing stack. If PC2 misses changes, a periodic "Tools → Scan for new items" on PC2 is the reliable fallback — it re-reads the metadata of changed files. * Faces: confirmed face names are stored as XMP regions in the file, so a name confirmed on PC1 will arrive on PC2 as a face region with the person tag. However, the face pipeline data (face rectangles of unconfirmed candidates, ignored/skipped statuses, recognition training) is purely local to each database and is not transported through metadata. So exactly as you say: recognition status may differ between the two PCs, but confirmed identifications, the part that matters, travel with the file. * Tag hierarchy: XMP keywords store the tag name (with the full path as option), not the tag tree. If PC1 and PC2 have diverging tag hierarchies, keywords arriving from PC1 may land at the wrong place on PC2. Try to keep the tag tree identical on both machines. * Moving folders: if you move files into different folders, PC2 will see "new items" in the new location plus the old ones disappearing. That is handled correctly by a scan, but the album root paths must be identical from both PCs' point of view (same UNC path or mapped drive letter) so that the local databases stay coherent with the collection. * Practical summary Your assessment is correct: with metadata written to files, per-PC SQLite databases, monitoring enabled and "re-read metadata on change" enabled on the receiving side, changes propagate through the files themselves. Add a periodic full scan as a safety net for missed network notifications, and keep the tag tree and metadata settings identical on both machines. The only thing that remains per-PC is the unconfirmed-face state, which, as you noted, is negligible. Best regards, Gilles Caulier -- You are receiving this mail because: You are watching all bug changes.
