v0.4.0 broke asset bundling

[ignatz]

Summary

Now getting:

called `Result::unwrap()` on an `Err` value: no embedded PGlite runtime assets are available; enable the `bundled` feature or set PGLITE_OXIDE_RUNTIME_ARCHIVE

even with bundled feature enabled (it's also enabled by default).

(on a tangent, haven't looked into the details but v0.4.0 also seems to have significantly increased the transitive dependency footprint, which may make sense or may be accidental. I looked at the change but it's too much for my teeny brain to review)

Reproduction

clone run `cargo test`

Versions

v0.4.0

[ignatz]

When I pull v0.4.1 the behavior seems unchanged. Am I holding it wrong?

[f0rr0]

the transitive dep increase is not accidental. i migrated to wasmer from wasmtime. also i just tested v0.4.1 it is working fine for me. can you maybe give me a small repro?

[ignatz]

Repro remains the same:

git clone https://github.com/f0rr0/pglite-oxide.git
cd pglite-oxide
cargo test --all-features 

I'm getting:

---- pglite::base::tests::bundled_extension_archive_hash_is_validated stdout ----

thread 'pglite::base::tests::bundled_extension_archive_hash_is_validated' (921771) panicked at src/pglite/base.rs:2507:14:
embedded vector archive
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace

---- pglite::base::tests::embedded_runtime_archive_hash_is_validated stdout ----

thread 'pglite::base::tests::embedded_runtime_archive_hash_is_validated' (921774) panicked at src/pglite/base.rs:2494:14:
embedded runtime archive

---- pglite::pg_dump::tests::direct_pg_dump_public_api_round_trip stdout ----
Error: no embedded PGlite runtime assets are available; enable the `bundled` feature or set PGLITE_OXIDE_RUNTIME_ARCHIVE
...

Which is the same issue I'm seeing in my code.

My checkout is at:

commit d90355014896ac7666a890c403be71f6748f81b1 (HEAD -> main, tag: 0.4.1, origin/main, origin/HEAD)

[f0rr0]

The git repo is intentionally source-only for generated runtime/AOT assets, so cargo test --all-features enables the asset crates but they have no embedded payloads unless assets have been installed first.

For a fresh checkout, run:

host="$(rustc -vV | awk '/^host:/{print $2}')"
cargo run -p xtask -- assets download --latest-compatible --target-triple "$host"
cargo test --all-features

The published crates.io 0.4.1 packages do include the assets, so normal crate consumers should not need this step.

[ignatz]

My bad, I got hung up on the repo checkout. I can confirm that the crate works 🙏 . (I couldn't reproduce the setup for the fresh checkout, since I'm not sure I want to install the gh CLI :hide:).

(On a complete tangent and as someone who uses wasmtime themselves, what's the reason you switched to wasmer?)

[ignatz]

Another quick update, when I call a function like uuidv4(), which may not have existed in PG17, I'm getting:

[2026-05-08T08:59:27Z WARN  pglite_oxide::pglite::postgres_mod] PostgresMainLoopOnce trapped; attempting protocol recovery: RuntimeError: uncaught exception
        at <unnamed> (<module>[14146]:0xffffffff)
        at <unnamed> (<module>[13256]:0xffffffff)
        at <unnamed> (<module>[13252]:0xffffffff)
        at <unnamed> (<module>[4412]:0xffffffff)
        at <unnamed> (<module>[4395]:0xffffffff)
        at <unnamed> (<module>[3894]:0xffffffff)
        at <unnamed> (<module>[3893]:0xffffffff)
        at <unnamed> (<module>[5065]:0xffffffff)
        at <unnamed> (<module>[9040]:0xffffffff)
        at <unnamed> (<module>[9027]:0xffffffff)
        at <unnamed> (<module>[9026]:0xffffffff)
        at <unnamed> (<module>[9023]:0xffffffff)
        at <unnamed> (<module>[14221]:0xffffffff)

Seems a bit spammy for what ultimately amounts to a "function uuidv4() does not exist" parse error, or is this a consequence of uuidv4() should exist and isn't wired up in pglite? 🤷‍♀️

[f0rr0]

1. The maintainer path (using this repo instead of crates.io) currently shells out to gh cli to download the assets. This can be improved.
2. wasmer was needed for dynamic linking support. Extensions are separate artefacts and can be added on demand.
3. This is currently based on PG 17 which uses gen_random_uuid(). We can put the noisy thing behind a debug log level.