feat(rust): re-export what a consumer driving the connector needs - #746
Draft
wkirschenmann wants to merge 4 commits into
Draft
wkirschenmann wants to merge 4 commits into
wkirschenmann wants to merge 4 commits into
Conversation
☂️ Python Coverage
Overall Coverage
New FilesNo new covered files... Modified FilesNo covered modified files...
|
wkirschenmann
force-pushed
the
wk/feat/rust-transport-reexports
branch
2 times, most recently
from
August 9, 2026 10:40
7155d44 to
e10db52
Compare
`hyper_util::client::legacy::Client` exists only when `http1` or `http2` is on, and the crate
re-exports `hyper_util` so a dependent can build that client over the connector. `http1` was named
here; `http2`, the protocol such a dependent actually speaks, was on only because `hyper-rustls`
happens to ask for it. Depending on another crate's feature list would break the day it changes.
Writing the feature down turns nothing on, which the lockfile is the proof of:
cargo build --workspace --all-features -> Finished
git diff --exit-code -- packages/rust/Cargo.lock -> no diff
cargo test --workspace --all-features --no-fail-fast
armonik-transport lib: 106 passed; every other target passed
armonik lib: 5 passed, 110 failed, the pre-existing count for want of a running
ArmoniK.Api.Mock server, unchanged by this commit
cargo clippy --workspace --all-targets --all-features -- -D warnings -> Finished
cargo clippy --workspace --all-features --no-deps -- -Dwarnings
-Dunused-crate-dependencies -> Finished
cargo fmt --check -> clean
The crate stops at the connector and leaves the HTTP/2 engine to whoever consumes it, so it owes
that consumer the crates the engine is spoken to: `tokio` for the stream types the connector hands
back, `h2` for the reason a stream was reset, `http_body_util` for the bodies hyper is given. Taken
from crates.io separately, any of the three can resolve to a version other than the one hyper was
built against, and the mismatch shows up as a type error at the seam.
`h2` and `http_body_util` become dependencies for the re-export alone and are used nowhere in the
crate; the module documents that, so a reader does not go looking for the call site. `tokio` was
already a dependency. `tokio_stream` is not here: it used to arrive through `tonic::codegen`, and
with tonic gone nothing in the crate or its connector speaks it.
Only the versions are guaranteed, not the feature sets, which cargo unifies across the whole build.
The module and the README both say so, because a consumer reaching for `tokio::runtime` through this
re-export is the way to be surprised by it. The README names the one feature a consumer is likely to
go looking for and not find: `http-body-util`'s `channel`, left off because the `tokio` edge it adds
belongs on the dependency of whoever feeds a request body, where the cost is attached to the need.
cargo build --workspace --all-features -> Finished
cargo test --workspace --all-features --no-fail-fast
armonik-transport lib: 106 passed; every other target passed
armonik lib: 5 passed, 110 failed, the pre-existing count for want of a running
ArmoniK.Api.Mock server, unchanged by this commit
cargo clippy --workspace --all-targets --all-features -- -D warnings -> exit 0
cargo clippy --workspace --all-features --no-deps -- -Dwarnings
-Dunused-crate-dependencies -> exit 0
cargo fmt --check -> clean
`Cargo.lock` gains two edges from `armonik-transport` and no crate: both were already resolved for
the workspace, `h2` inside this crate's own tree through `hyper`'s `http2`.
…s built from
`https_connector` returns `HttpsConnector<ProxyConnector<HttpConnector>>`. That type was already
public, being the return type of a public function, but it could not be written down outside this
crate because `ProxyConnector` was not exported. Public and unnameable is the worst of both: a
consumer that wraps the connector in an engine of its own has to name the result to hold it, and
Rust cannot infer the type of a struct field, so it was pushed into a trait object erasing a
connector that has no reason to be erased.
`Connector` is that name, `ProxyConnector` is exported because it is a layer of it, and
`https_connector` loses the `#[doc(hidden)]` that was only hiding this. `ProxyConnector::new` stays
crate-private: the layer is assembled from an `HttpConfig`, never by hand.
cargo build --workspace --all-features -> Finished
cargo doc -p armonik-transport --all-features --no-deps -> Finished, no warnings
cargo test --workspace --all-features --no-fail-fast
armonik-transport lib: 106 passed; every other target passed
armonik lib: 5 passed, 110 failed, the pre-existing count for want of a running
ArmoniK.Api.Mock server, unchanged by this commit
cargo clippy --workspace --all-targets --all-features -- -D warnings -> Finished
cargo clippy --workspace --all-features --no-deps -- -Dwarnings
-Dunused-crate-dependencies -> Finished
cargo fmt --check -> clean
It awaits nothing: the certificates were loaded when the configuration was read, and what is left is
assembling a connector, which opens no socket. Being `async` all the same charged every caller a
runtime for a future that never yields, and left a caller already inside its only runtime with a
`block_on` that panics there rather than waits - a restriction that would otherwise have to be
documented and honoured forever.
Both callers are in `armonik`, and each loses an `.await`: `client::channel::connect`, and the
`get_nb_request` test helper that reaches the mock server's diagnostic endpoint.
cargo build --workspace --all-features -> Finished
cargo test --workspace --all-features --no-fail-fast
armonik-transport lib: 106 passed; every other target passed
armonik lib: 5 passed, 110 failed, the pre-existing count for want of a running
ArmoniK.Api.Mock server, unchanged by this commit
cargo clippy --workspace --all-targets --all-features -- -D warnings -> Finished
cargo clippy --workspace --all-features --no-deps -- -Dwarnings
-Dunused-crate-dependencies -> Finished
cargo fmt --check -> clean
wkirschenmann
force-pushed
the
wk/feat/rust-transport-reexports
branch
from
August 9, 2026 10:52
e10db52 to
813fb17
Compare
|
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Sits on
wk/feat/csharp-options-generator, to merge first.Motivation
armonik-transportstops at the connector and leaves the HTTP/2 engine to whoever consumes it. Ittherefore owes that consumer the crates the engine is spoken to, and a type it can write down to
hold the connector in.
Description
hyper-util'shttp2feature is declared rather than relied on transitively, since it was on onlybecause
hyper-rustlshappens to ask for it and another crate's feature list can change. Writing itdown turns nothing on, and
Cargo.lockstaying byte-identical is the proof.tokio,h2andhttp-body-utilare re-exported: the stream types the connector hands back, thereason a stream was reset, the bodies hyper is given. The last two are dependencies for the
re-export alone, used nowhere in the crate, which is the point rather than the smell - taken from
crates.io separately, either can resolve to a version other than the one hyper was built against,
and the skew shows up as a type error at the seam.
ConnectornamesHttpsConnector<ProxyConnector<HttpConnector>>, andProxyConnectoris exportedbecause it is a layer of it. That type was already public, being a public function's return type,
but unnameable outside the crate - the worst of both, since a consumer holding it in a struct field
had to erase it into a trait object for want of a name.
https_connectorbecomes synchronous: it awaits nothing, and beingasynccharged every caller aruntime for a future that never yields while leaving a caller already inside its only runtime with a
block_onthat panics there rather than waits.Testing
No test is added or removed - this declares a feature, adds three re-exports, exports a type and
drops an
async.cargo test --workspace --all-features --no-fail-fastis unmoved, and the checkthe first commit rests on is
git diff --exit-code -- packages/rust/Cargo.lockafter a build.Impact
https_connectoris no longerasync, a breaking signature change; both callers are inarmonikand lose an
.await, and the crate has no published release. A re-export guarantees versions, notfeature sets, which cargo unifies across the build; the README says so and names the case a consumer
is likeliest to meet.
Additional Information
tokio_streamis deliberately not re-exported: it used to arrive throughtonic::codegen, andnothing in the crate or its connector speaks it now.
Checklist