Conversation
chrysh
force-pushed
the
service-loopback
branch
from
September 25, 2026 13:14
75a3135 to
7e79206
Compare
Every IPC service so far has defined its own copy of the same three seams: a transport trait, a server dispatch function, and the error type between them. i2c and pldm each carry one, and they have already drifted apart. This crate holds them once so a service defines its wire format and nothing else. Transport blocks until the response arrives; AsyncTransport splits the round-trip so an event-loop caller never blocks. A type implements whichever it can serve, or both. Dispatch is the server end, one request frame in, one response frame out. It returns Result rather than a byte count with 0 meaning failure: a service that merely failed an operation encodes that into the response frame and still returns Ok, and the one case with nothing to send back is a response buffer too small to hold even an error frame. Traits only, no implementations and no migrations, so there is nothing to unit test here. The generic loopback and the kernel channel transport follow, then i2c and pldm move over. Assisted-by: Claude Code
i2c and pldm each wrote their own loopback, both the same shape around a different dispatch function. Loopback<D, N> is that shape once, over the Dispatch trait: the server implementation that runs behind a kernel channel in production answers in-process here, so a host test exercises the real encode, dispatch and decode paths. It serves both transport traits because an in-process call can, and that is also its limit: the response exists the moment the request does, so poll never returns Ok(None). Testing an event loop's not-ready handling needs a real channel. N sizes the response buffer the async path holds between start and poll. The blocking path writes into the caller's buffer and does not use it. Assisted-by: Claude Opus 5
chrysh
force-pushed
the
service-loopback
branch
from
September 25, 2026 13:33
7e79206 to
842c066
Compare
This branch has not been deployed
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.
i2c and pldm each wrote their own loopback, both the same shape around a different dispatch function.
Loopback<D, N>is that shape once, over theDispatchtrait.The point is that the server implementation running behind a kernel channel in production is the one answering here, so a host test exercises the real encode, dispatch and decode paths instead of a stand-in.
It implements both transport traits, because an in-process call can serve either. That is also its limit: the response exists the moment the request does, so
pollnever returnsOk(None). Testing an event loop's not-ready handling needs a real channel, which is what the QEMU test in #497 does.Nsizes the response buffer the async path holds betweenstartandpoll. The blocking path writes into the caller's buffer and does not use it.Nine unit tests: both transport shapes, a response that does not fit on each path, double start, poll and cancel with nothing in flight, and a dispatch that cannot answer at all.
Not in here: the i2c and pldm migrations that delete their local loopbacks, and the typed clients.
Review the last commit only; below it is #485.
Progress toward 9elements#13.