[Rust CLI] ubuild panics with “failed to create default configuration” on Docker (v0.6.10)

[saga-dash]

When the resource path differed between the build stage and the runner stage in a Docker multi-stage build, ubuild resulted in an error.

Environment

• Implementation: sudachi.rs (Rust)
• CLI: sudachi-cli v0.6.10 (from GitHub releases tarball)
• Docker
  • build image: rust:1.89-bullseye
  • runtime image: debian:stable-slim

Minimal repro (fails)

# dockerfile
FROM rust:1.89-bullseye AS build
ENV SUDACHI_RS_VERSION=v0.6.10
WORKDIR /opt
# WORKDIR /app
RUN apt-get update && apt-get install -y curl unzip ca-certificates && rm -rf /var/lib/apt/lists/*
RUN curl -sSL "https://github.com/WorksApplications/sudachi.rs/archive/refs/tags/${SUDACHI_RS_VERSION}.tar.gz" \
    | tar xz --strip-components=1 \
 && chmod +x ./fetch_dictionary.sh && ./fetch_dictionary.sh \
 && cargo build --release && cargo install --path sudachi-cli/

FROM debian:stable-slim
COPY --from=build /usr/local/cargo/bin/sudachi /usr/local/bin/sudachi
COPY --from=build /opt/resources /app/resources
#COPY --from=build /app/resources /app/resources
WORKDIR /work
ENTRYPOINT ["sudachi"]

# dict/user.csv
高輪ゲートウェイ駅,4786,5146,8000,高輪ゲートウェイ駅,名詞,固有名詞,一般,*,*,*,タカナワゲートウェイエキ,高輪ゲートウェイ駅,*,*,*,*,*

Run

docker build -t sudachi .
docker run --rm -v "$PWD/dict:/work" sudachi ubuild -s /app/resources/system.dic -o /work/user.dic /work/user.csv

thread 'main' panicked at sudachi-cli/src/build.rs:146:47:
failed to create default configuration: Io(Os { code: 2, kind: NotFound, message: "No such file or directory" })

日本語の補足

Dockerのマルチステージビルドでresourcesのパスをbuildとrunnerと別のものにしたところ、ubuildでエラーとなりました。 buildとrunnerのパスを合わせるとエラーは出ません

[t-yamamura]

Thank you for your report. I have confirmed that the issue can be reproduced. I will investigate the cause of the problem.

[t-yamamura]

In Docker multi-stage builds, if the path to the resources directory differs between the build stage and the runtime stage, ubuild fails because the ubuild subcommand is designed to reference a fixed path (e.g., /opt/resources/sudachi.json) at build time.

The simplest workaround is to place the resources in the same path for both the build and runtime stages, as you mentioned.
Alternatively, you can use a symbolic link to align the paths instead of copying.

< # COPY --from=build /app/resources /app/resources
< RUN ln -s /app/resources /opt/resources
---
> #COPY --from=build /app/resources /app/resources

$ docker run --rm -v "$PWD/dict:/work" sudachi ubuild -s /app/resources/system.dic -o /work/user.dic /work/user.csv
/work/user.csv        1 entries in 0.001 sec
validate              1 entries in 0.000 sec
pos_table             2 bytes in 0.000 sec
conn_matrix           6 bytes in 0.000 sec
trie                  1028 bytes in 0.000 sec
word_id table         9 bytes in 0.000 sec
word_params           10 bytes in 0.000 sec
wordinfo_offsets      4 bytes in 0.000 sec
wordinfos (copy only) 56 bytes in 0.000 sec

Currently, the ubuild subcommand does not have CLI options like --resource_dir or --config-file. Therefore, modifying the ubuild code is necessary if the resources directory needs to be specified dynamically at runtime.

[saga-dash]

Thank you for confirming the reproduction and sharing the workaround.

I’m not familiar with the milestone/roadmap for the Rust CLI, but when distributing binaries this behavior may confuse end users. It would be very helpful if you could consider adding a way to override the resources location at runtime for ubuild (e.g., CLI flags such as --resource-dir / --config-file, or an environment variable).

Again, thank you for the quick response.