to_obsidian: case-only filename collisions silently overwrite node notes on case-insensitive filesystems #2650

Case-only filename collisions overwrite each other

In export.py::to_obsidian(), the filename dedup map is keyed on the exact-case name (v0.8.37, lines 867-876):

node_filename: dict[str, str] = {}
seen_names: dict[str, int] = {}
for node_id, data in G.nodes(data=True):
    base = safe_name(data.get("label", node_id))
    if base in seen_names:          # <-- case-SENSITIVE membership test
        seen_names[base] += 1
        node_filename[node_id] = f"{base}_{seen_names[base]}"
    else:
        seen_names[base] = 0
        node_filename[node_id] = base

Two nodes whose labels differ only in case produce two distinct keys, so no _1 suffix is added — but on a case-insensitive filesystem they resolve to the same path, and the second write_text silently overwrites the first. One node's note is lost, and the [[wikilink]] to it from other notes points at a file whose on-disk casing came from whichever node wrote last.

Note that safe_name() strips / and :, which increases the collision rate — api/analyze Endpoint and api:analyze endpoint both collapse toward the same stem.

Observed

7 collisions in one 1,729-node export:

The output directory holds 1,899 .md files against 1,729 nodes + 177 community notes = 1,906 expected — i.e. 7 fewer files than expected. That shortfall is exactly the collision count above, which is what silent overwrite looks like from the outside.

Minimal reproduction

Four nodes, two case-only pairs, run through the exact safe_name + dedup logic above:

CURRENT   filenames=['apianalyze Endpoint', 'apianalyze endpoint',
                     'SprintSelector Component', 'SprintSelector component']
          files on disk = 2 of 4 nodes  -> 2 LOST
          unresolved links: ['apianalyze endpoint', 'SprintSelector component']

Suggested fix — verified against the same reproduction

Key the dedup on the casefolded name while keeping the original for display:

key = base.casefold()
if key in seen_names:
    seen_names[key] += 1
    node_filename[node_id] = f"{base}_{seen_names[key]}"
else:
    seen_names[key] = 0
    node_filename[node_id] = base

PROPOSED  filenames=['apianalyze Endpoint', 'apianalyze endpoint_1',
                     'SprintSelector Component', 'SprintSelector component_1']
          files on disk = 4 of 4 nodes  -> 0 LOST
          unresolved links: []

This is also correct on case-sensitive filesystems — it just adds a suffix where Linux would have allowed two distinct files. If that's undesirable, gate it behind a platform check or an --assume-case-insensitive-fs flag; defaulting to safe seems better than defaulting to silent loss.

Severity

Low in practice for Obsidian users: Obsidian resolves wikilinks case-insensitively, so the surviving file still answers links aimed at the lost one. In the export above, 0 of 1,799 distinct link targets is genuinely unresolved — the loss is completely invisible from inside the app. That is the argument for fixing it rather than against: seven nodes have no note and nothing anywhere surfaces it.

Environment

• graphifyy 0.8.37, Python 3.11
• macOS 26.6.1 (Darwin 25.6.0), APFS — case-insensitivity confirmed empirically, not assumed
• Graph: 1,729 nodes / 2,077 edges / 177 communities