The user-facing API-level change of this commit is that `SharedChan` is gone and `Chan` now has `clone`. The major parts of this patch are the internals which have changed.
Channels are now internally upgraded from oneshots to streams to shared channels depending on the use case. I've noticed a 3x improvement in the oneshot case and very little slowdown (if any) in the stream/shared case.
This patch is mostly a reorganization of the `std::comm` module, and the large increase in code is from either dispatching to one of 3 impls or the duplication between the stream/shared impl (because they're not entirely separate).
The `comm` module is now divided into `oneshot`, `stream`, `shared`, and `select` modules. Each module contains the implementation for that flavor of channel (or the select implementation for select).
Some notable parts of this patch
* Upgrades are done through a semi-ad-hoc scheme for oneshots and messages for streams
* Upgrades are processed ASAP and have some interesting interactions with select
* send_deferred is gone because I expect the mutex to land before this
* Some of stream/shared is straight-up duplicated, but I like having the distinction between the two modules
* Select got a little worse, but it's still "basically limping along"
* This lumps in the patch of deallocating the queue backlog on packet drop
* I'll rebase this on top of the "more errors from try_recv" patch once it lands (all the infrastructure is here already)
All in all, this shouldn't be merged until the new mutexes are merged (because send_deferred wasn't implemented).
Closes#11351
This is an attempt to remove some more of Rust's dependencies on libgcc and replace it with LLVM's compiler-rt lib. I've added compiler-rt as a submodule and changed libstd to link with it.
As far as I could verify, after this change, the only symbols still imported by std from libgcc are the stack unwinding functions. Other crates, however, still picked up symbols from libgcc, not from libstd, as I had hoped. So linking definitely requires some work.
I've only tested this on windows, 32-bit linux and android and fully expect it to fail on other platforms. Patches are welcome.
This, the Nth rewrite of channels, is not a rewrite of the core logic behind
channels, but rather their API usage. In the past, we had the distinction
between oneshot, stream, and shared channels, but the most recent rewrite
dropped oneshots in favor of streams and shared channels.
This distinction of stream vs shared has shown that it's not quite what we'd
like either, and this moves the `std::comm` module in the direction of "one
channel to rule them all". There now remains only one Chan and one Port.
This new channel is actually a hybrid oneshot/stream/shared channel under the
hood in order to optimize for the use cases in question. Additionally, this also
reduces the cognitive burden of having to choose between a Chan or a SharedChan
in an API.
My simple benchmarks show no reduction in efficiency over the existing channels
today, and a 3x improvement in the oneshot case. I sadly don't have a
pre-last-rewrite compiler to test out the old old oneshots, but I would imagine
that the performance is comparable, but slightly slower (due to atomic reference
counting).
This commit also brings the bonus bugfix to channels that the pending queue of
messages are all dropped when a Port disappears rather then when both the Port
and the Chan disappear.
Beforehand, using a concurrent queue always mandated that the "shared state" be
stored internally to the queues in order to provide a safe interface. This isn't
quite as flexible as one would want in some circumstances, so instead this
commit moves the queues to not containing the shared state.
The queues no longer have a "default useful safe" interface, but rather a
"default safe" interface (minus the useful part). The queues have to be shared
manually through an Arc or some other means. This allows them to be a little
more flexible at the cost of a usability hindrance.
I plan on using this new flexibility to upgrade a channel to a shared channel
seamlessly.
I implemented an add method for the btree in progress. It is intended to be refactored later using an alternative to .clone() that passes the borrow checker, but for now, it works as intended. r? @catamorphism
I factored the commits by affected files, for the most part. The last 7 or 8 contain the meat of the PR. The rest are small changes to closures found in the codebase. Maybe interesting to read to see some of the impact of the rules.
r? @pcwalton
Fixes#6801
