https://github.com/tegmentum/shim-bridge-smoke-tests
End-to-end smoke tests for DataFission shim bridges (sqlink, ducklink). Runs query suites through sqlite3 + duckdb CLIs.
https://github.com/tegmentum/shim-bridge-smoke-tests
Last synced: 9 days ago
JSON representation
End-to-end smoke tests for DataFission shim bridges (sqlink, ducklink). Runs query suites through sqlite3 + duckdb CLIs.
- Host: GitHub
- URL: https://github.com/tegmentum/shim-bridge-smoke-tests
- Owner: tegmentum
- Created: 2026-06-24T18:36:31.000Z (about 2 months ago)
- Default Branch: main
- Last Pushed: 2026-07-08T02:05:26.000Z (about 1 month ago)
- Last Synced: 2026-07-08T02:14:48.525Z (about 1 month ago)
- Language: Shell
- Size: 43 KB
- Stars: 0
- Watchers: 0
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# shim-bridge-smoke-tests
End-to-end smoke test suite for DataFission shim bridges.
Catches regressions during bridge-codegen development by
running query suites through `sqlite3`, `duckdb`, and
`ducklink` (wasm-component DuckDB host) against generated
bridges.
**Fleet index**: [SHIM-BRIDGES.md](SHIM-BRIDGES.md) —
canonical map of every generated bridge (extension × host),
per-sub compose:dynlink model documentation, and the
regeneration recipes for both the legacy `wac plug`
monolithic path and the current Phase B/9.1/9.3 dynlink
per-sub / shared-shim / monolithic paths.
## What it tests
For each (target, bridge artifact, composed shim wasm) tuple,
the runner:
1. Loads the bridge into the target's CLI.
2. Sets the `_SHIM_WASM` env var so the bridge can find
its composed wasm.
3. Runs each `.sql` file in the case directory.
4. Diffs the actual output against `.expected`.
5. Reports per-case PASS/FAIL.
## Usage
```sh
# SQLite via sqlink (wasm-component host; the current default path)
scripts/run.sh sqlite \
/path/to/postgis-sqlink-loadable.wasm \
/path/to/postgis-shim-composed.wasm \
cases/postgis
# DuckDB via ducklink (wasm-component host)
DUCKLINK=~/git/ducklink/target/release/ducklink scripts/run.sh ducklink \
/path/to/postgis-ducklink-loadable.wasm \
/path/to/postgis-shim-composed.wasm \
cases/postgis
# DuckDB via native cdylib (legacy path; the -duckdb-bridge repos
# are archived — kept runnable for regression comparison only)
DUCKDB=/opt/homebrew/bin/duckdb scripts/run.sh duckdb \
/path/to/postgis_duckdb_bridge.duckdb_extension \
/path/to/postgis-shim-composed.wasm \
cases/postgis
```
### Ducklink runtime prerequisites
The `ducklink` target requires the ducklink host binary AND
its two wasm components:
```sh
cd ~/git/ducklink
cargo build --release -p ducklink-host --bin ducklink
# Also builds ~/git/ducklink/target/wasm32-wasip2/release/{ducklink_core,ducklink_cli}.wasm
# via the same crate's wasm workspace members.
```
The runner copies the composed loadable to a scratch
`--extensions-dir` under its canonical extension name
(`.wasm` — derived from the `-ducklink-loadable.wasm`
suffix) and issues `LOAD ;` inside the guest DuckDB CLI.
No `_SHIM_WASM` env var is needed — the shim wasm is
already inlined into the composed loadable via `wac plug`.
## Case design
`cases/postgis/` holds **portable cases** that run on both
SQLite and DuckDB. Every query returns an integer (typically
0 or 1 from a `CASE WHEN THEN 1 ELSE 0 END`)
because SQL formatting differs across targets:
| | SQLite `.mode list` | DuckDB `.mode csv` |
|---|---|---|
| Boolean | `1` / `0` | `true` / `false` |
| String with commas | `LINESTRING(0 0,1 1)` | `"LINESTRING(0 0,1 1)"` |
| NULL | empty | `NULL` |
| Integer | `1` | `1` |
Integers are the lowest-common-denominator output that's
identical everywhere. Wrapping boolean and string predicates
in `CASE WHEN ... THEN 1 ELSE 0 END` keeps the cases portable.
For features that only one target supports (e.g. UDTFs which
sqlink ships but ducklink scaffolds), cases live under
`cases/--only/`. Run with the matching target.
## Adding a new case
```sh
# cases/postgis/06-clusters.sql
SELECT CASE WHEN ST_NumGeometries(
ST_ClusterIntersecting()
) = 2 THEN 1 ELSE 0 END;
# cases/postgis/06-clusters.expected
1
```
The runner normalises trailing whitespace on each line and
trailing empty lines, so `.expected` files are forgiving to
hand-edit.
## Status
Verified 2026-06-24 on:
- SQLite v3.53.1 (brew) — 5/5 cases pass
- DuckDB v1.5.2 — 4/4 portable cases pass (UDTF cases are
sqlink-only)
Adding new shims is a matter of writing one more case
directory. The runner is shim-agnostic; it just needs to know
the env var name for the shim wasm path (currently hardcoded
to `POSTGIS_SHIM_WASM` — see `scripts/run.sh` for the
extension point).