You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
2.0.1 file import: property and class share the ident namespace, tag becomes a property entity (should be a Class) #1402
Repro: Graph uses rolle:: as a property and tags:: [[Rolle]] on another page. find-or-create-class finds the existing :rolle ident (the property) and returns it as the class → pages tagged Rolle reference a property entity → should be a Class. Master already handles "property and class may share a title" (class-occupied-property-names); 2.0.1 does not. In our graph 14 names were affected (notiz, unternehmen, ort, projekt, quelle, …) — the biggest tags of the graph (Person 373 pages, Notiz 271) never became classes because of knock-on effects.
Adding another variant of this "should be a Class" import failure, which still reproduces on master (2693a98, 2026-10-01), unlike the property/class name collision described above.
Repro (different trigger): dangling tags
A page with page-level tags that reference pages which don't exist anywhere in the graph:
tags:: #repository #meterhub #CRNN
description:: used for meterhub CRNN
#repository and #CRNN have no page of their own anywhere in the graph (free-floating tags)
#meterhub exists as a regular content page (which itself has tags:: #[[meter reading]] #Research #[[Research Dataset]])
With the default "Import all tags" option (GUI default is ON), the file→DB import deterministically dies with:
→ renderer shows "DB write failed with invalid data", import aborts at the same point every run (~41 MB into our graph).
Environment
Windows 10, Logseq stable 2.0.1 (b09316a) and a self-built master 2693a98-dirty (2026-10-01) — both affected
Graph: 2,744 Markdown files (2,075 docs + assets), Chinese + English, Zotero literature notes
Workaround: import with the "Import all tags" input cleared (tags come in as plain pages / Page Tags property instead of classes); or remove the dangling tags from page-level tags:: properties beforehand.
So beyond the property/class ident collision, the tag→class conversion also produces non-class refs when a tag target simply has no page — worth covering both paths in a fix.
Activity
Svtter commentedon Oct 8, 2026
Adding another variant of this "should be a Class" import failure, which still reproduces on master (
2693a98, 2026-10-01), unlike the property/class name collision described above.Repro (different trigger): dangling tags
A page with page-level tags that reference pages which don't exist anywhere in the graph:
#repositoryand#CRNNhave no page of their own anywhere in the graph (free-floating tags)#meterhubexists as a regular content page (which itself hastags:: #[[meter reading]] #Research #[[Research Dataset]])With the default "Import all tags" option (GUI default is ON), the file→DB import deterministically dies with:
→ renderer shows "DB write failed with invalid data", import aborts at the same point every run (~41 MB into our graph).
Environment
b09316a) and a self-built master2693a98-dirty(2026-10-01) — both affectedWorkaround: import with the "Import all tags" input cleared (tags come in as plain pages / Page Tags property instead of classes); or remove the dangling tags from page-level
tags::properties beforehand.So beyond the property/class ident collision, the tag→class conversion also produces non-class refs when a tag target simply has no page — worth covering both paths in a fix.