Files
dataxl/docs/development.md
Luno 2e96d174f2
All checks were successful
CI / test (pull_request) Successful in 26s
Document conversion internals and limits
2026-07-18 16:25:50 +09:00

2.5 KiB

Development

Requirements

  • Go 1.24 or later

Setup

git clone https://git.rumginger.org/agent/dataxl.git
cd dataxl
go mod download

Common Commands

Format code:

gofmt -w cmd/dataxl/*.go

Run tests:

go test ./...

Build:

go build ./cmd/dataxl

Run locally:

go run ./cmd/dataxl -from yaml -to tsv -i examples/people.yaml

Build release archives locally:

scripts/build-release.sh

Test Coverage

The current tests cover:

  • YAML -> TSV flattening
  • TSV -> JSON path restoration
  • JSON -> XLSX -> JSON round trip
  • structured -> structured conversion without CLI/file I/O
  • extension normalization and format inference
  • ragged table row padding
  • conservative cell type inference, including zero-padded identifiers
  • conflicting unflatten paths

When adding a new format or path rule, add tests around both directions where possible.

CI/CD

Gitea Actions workflows live under .gitea/workflows.

  • ci.yml: runs on pushes to main, pull requests, and manual dispatch.
  • release.yml: runs on v* tag pushes and manual dispatch with a tag input.

The CI workflow checks formatting, runs tests, builds the CLI, and performs a small YAML -> TSV -> JSON smoke test.

The release workflow runs tests, cross-builds release archives for Linux, macOS, and Windows on amd64/arm64, writes checksums.txt, creates or reuses a Gitea Release, and uploads the generated assets. It uses the built-in ${{ secrets.GITEA_TOKEN }} provided by Gitea Actions.

Release Notes

Create a release by pushing a version tag:

go test ./...
git tag v0.1.0
git push origin main --tags

The same workflow can also be started manually from Gitea Actions by providing the target tag as the tag input.

After tags exist, users can install a specific version:

go install git.rumginger.org/agent/dataxl/cmd/dataxl@v0.1.0

Design Guidelines

  • Keep stdin/stdout usable for shell pipelines.
  • Keep TSV behavior predictable because it is the main Excel clipboard format.
  • Preserve headers as the contract between spreadsheet data and structured data.
  • Prefer explicit errors over silent best-effort conversion when a format is unsupported.
  • Keep dependencies small unless a format needs a mature parser/writer.
  • Keep CLI/file I/O in main.go; conversion behavior should remain testable through convert.
  • Keep format-specific behavior in its structured or table adapter.
  • Document round-trip limitations when a representation cannot preserve a value exactly.