# CanLex MCP server -- remote (streamable-HTTP) image.
#
# Builds anywhere: Hugging Face Spaces, Google Cloud Run, Fly.io, plain Docker.
# Retrieval is fully local and key-free; the optional CanLII case citator reads
# its key from the CANLII_API_KEY environment variable (supplied as a host
# secret -- the key is never copied into the image).
FROM python:3.12-slim

# libgomp1 is the OpenMP runtime that onnxruntime (the reranker) links against.
RUN apt-get update \
 && apt-get install -y --no-install-recommends libgomp1 \
 && rm -rf /var/lib/apt/lists/*

# Run as a non-root user (UID 1000) -- required by Hugging Face Spaces.
RUN useradd --create-home --home-dir /app --uid 1000 app
WORKDIR /app

# Python dependencies first, so this layer caches across code changes.
COPY requirements.txt .
RUN pip install --no-cache-dir -r requirements.txt

# Application code, the processed corpus (section-chunk JSON), and the prebuilt
# semantic embeddings (embeddings.npz, committed via git-LFS). Shipping the
# embeddings as an artifact keeps the build fast: embedding ~20k chunks on the
# build host took ~33 min and was tripping Hugging Face's job timeout as the
# corpus grew. Regenerate with 'python -m canlex.embed' after any corpus change
# and commit the result.
COPY --chown=app:app canlex/ ./canlex/
COPY --chown=app:app data/processed/*.json ./data/processed/
COPY --chown=app:app data/processed/embeddings.npz ./data/processed/
# Curated commentary datasets (the canlex_us_disposition tool reads the
# structured source file directly, not just its processed chunks).
COPY --chown=app:app data/curated/ ./data/curated/

USER app
# CANLEX_LOG_DIR turns on the opt-in query log (query text + outcome only;
# loganalyze.py mines it for corpus gaps). On the free tier this path is
# ephemeral -- lost on rebuild/sleep -- so treat it as best-effort telemetry;
# the durable log is the local server's (see .mcp.json).
ENV HOME=/app \
    HF_HOME=/app/.hf_cache \
    CANLEX_HTTP=1 \
    CANLEX_LOG_DIR=/tmp/canlex-logs \
    PORT=7860 \
    PYTHONUNBUFFERED=1

# Pre-fetch the embedder and cross-encoder models so the cache is baked into the
# image and the first request needs no network. The corpus is NOT re-embedded
# here -- the committed embeddings.npz is used as-is; only the bge query model
# (for embedding live queries) and the reranker are downloaded.
RUN python -c "from canlex.embed import Embedder; Embedder()" \
 && python -c "from canlex.rerank import Reranker; Reranker()"

# From here on, model files are served from the baked cache, never fetched.
ENV HF_HUB_OFFLINE=1

EXPOSE 7860
CMD ["python", "-m", "canlex.server"]