wd-fuse is a read-only Wikidata FUSE prototype built on libfuse's high-level API.

The mount is intentionally hybrid:

• FUSE serves a read-only namespace.
• A Python materializer fetches Special:EntityData on first access.
• Fetched entities are frozen into a mount-private backing tree for snapshot stability.
• Reverse edges come from a local prebuilt index, never from remote on-demand queries.

• Default graph view: truthy/
• Full statement view: full/
• Entity-to-entity values: symlinks into /entities/<id>
• Literal values: .txt or tiny .json
• Ranks, qualifiers, references: nested under full statement directories
• Raw source views: raw.json, raw.ttl
• Large value sets: paginated under pages/<nnnn>/
• Writes: rejected

/
├── README.txt
├── snapshot.json
└── entities/
    └── Q42/
        ├── id.txt
        ├── type.txt
        ├── modified.txt
        ├── revision.txt
        ├── labels/
        ├── descriptions/
        ├── aliases/
        ├── truthy/
        │   └── by-property/
        │       └── P31/
        │           ├── property -> ../../../../../entities/P31
        │           └── values/
        ├── full/
        │   └── by-property/
        │       └── P31/
        │           ├── property -> ../../../../../entities/P31
        │           └── statements/
        ├── incoming/
        │   └── by-property/
        ├── raw.json
        └── raw.ttl

Truthy values use the best-rank Wikidata projection: preferred statements if any exist for a property, otherwise all non-deprecated statements.

Written once at mount time. Fields:

Runtime:

# Debian / Ubuntu
sudo apt install fuse3 libfuse3-3 python3

# Fedora / RHEL
sudo dnf install fuse3 fuse3-libs python3

Build:

# Debian / Ubuntu
sudo apt install cmake gcc libfuse3-dev

# Fedora / RHEL
sudo dnf install cmake gcc fuse3-devel

Python 3.10 or newer is required for the materializer (str | None union syntax).

The repo vendors the public libfuse headers because this environment only exposes the runtime library. You still need the shared library and fusermount3 on the host.

cmake -S . -B build
cmake --build build

mkdir -p /tmp/wd-mount
./build/wd-fuse \
  --incoming-index examples/incoming.jsonl \
  --page-size 256 \
  /tmp/wd-mount

mkdir -p /tmp/wd-cache /tmp/wd-mount
./build/wd-fuse \
  --cache-root /tmp/wd-cache \
  --incoming-index examples/incoming.jsonl \
  /tmp/wd-mount

Run in the foreground to keep error output visible (recommended for first-time use and debugging):

./build/wd-fuse --incoming-index examples/incoming.jsonl /tmp/wd-mount -f

Enable FUSE-level debug output:

./build/wd-fuse --incoming-index examples/incoming.jsonl /tmp/wd-mount -d

Then browse:

ls /tmp/wd-mount/entities/Q42
readlink /tmp/wd-mount/entities/Q42/truthy/by-property/P31/values/000000
cat /tmp/wd-mount/entities/Q42/full/by-property/P31/statements/count.txt

Unmount with:

fusermount3 -u /tmp/wd-mount

--incoming-index accepts either:

• a JSON Lines file with records like {"target":"Q42","property":"P50","source":"Q25169"}
• a directory of per-target .json or .jsonl shard files

A single file where every line is one reverse-edge record:

{"target":"Q42","property":"P50","source":"Q25169"}
{"target":"Q42","property":"P31","source":"Q463035"}
{"target":"Q5","property":"P31","source":"Q42"}

For large indexes, split into per-entity files under one of these layouts (tried in order):

index/
├── Q42.jsonl          # flat: index/Q42.jsonl
├── Q/
│   └── Q42.jsonl      # one-char prefix: index/Q/<id>.jsonl
└── QA/
    └── Q42.jsonl      # two-char prefix: index/QA/<id>.jsonl

Each shard file may be .jsonl (one record per line) or .json (array or {"edges":[...]} object). The "target" field may be omitted in per-entity shards; it is inferred from the filename.

The prototype reads only local index data for reverse edges — no remote queries are issued.

• Supported entity ids in v0: Q... and P...
• raw.ttl uses Special:EntityData/<id>.ttl?flavor=dump
• Without --revision-pin, each entity is fixed at the revision first fetched during the mount generation
• Entity materialization may take 1–3 seconds per entity (two HTTP requests to www.wikidata.org). Transient errors are retried up to three times with exponential back-off.

Transport endpoint is not connected — The mount process exited without cleanly unmounting. Force-unmount with:

fusermount3 -uz /tmp/wd-mount

Input/output error on a specific entity — The materializer failed (network error, unexpected API response, etc.). Check stderr output by running with -f. The entity directory will not be created, so retrying the access will attempt materialization again.

Nothing appears at the mountpoint — FUSE daemonized and may have exited immediately. Run with -f to keep output in the terminal and see any startup errors.