-
Notifications
You must be signed in to change notification settings - Fork 12.7k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Remove the pipes compiler #8170
Conversation
This is fine with me. I should like to see it re-implemented in a more idiomatic way post-1.0. (Though, the send_any/send_all convenience wrapper issues are still relevant without protocols.) |
I'm ok with this, at this point. It's been 10 months since we last revisited this and we haven't seen any new uses for the full proto system, just ongoing maintenance burden while trying to redo IO. Sad but IMO reasonable at this point. If we can get @eholk interested in reviving / retargeting the things that have changed since its initial construction, it might well live again as an external syntax extension. |
The pipes compiler produced data types that encoded efficient and safe bounded message passing protocols between two endpoints. It was also capable of producing unbounded protocols. It was useful research but was arguably done before its proper time. I am removing it for the following reasons: * In practice we used it only for producing the `oneshot` and `stream` unbounded protocols and all communication in Rust use those. * The interface between the proto! macro and the standard library has a large surface area and was difficult to maintain through language and library changes. * It is now written in an old dialect of Rust and generates code which would likely be considered non-idiomatic. * Both the compiler and the runtime are difficult to understand, and likewise the relationship between the generated code and the library is hard to understand. Debugging is difficult. * The new scheduler implements `stream` and `oneshot` by hand in a way that will be significantly easier to maintain. This shouldn't be taken as an indication that 'channel protocols' for Rust are not worth pursuing again in the future.
The pipes compiler produced data types that encoded efficient and safe bounded message passing protocols between two endpoints. It was also capable of producing unbounded protocols. It was useful research but was arguably done before its proper time. I am removing it for the following reasons: * In practice we used it only for producing the `oneshot` protcol and the unbounded `stream` protocol and all communication in Rust use those. * The interface between the proto! macro and the standard library has a large surface area and was difficult to maintain through language and library changes. * It is now written in an old dialect of Rust and generates code which would likely be considered non-idiomatic. * Both the compiler and the runtime are difficult to understand, and likewise the relationship between the generated code and the library is hard to understand. Debugging is difficult. * The new scheduler implements `stream` and `oneshot` by hand in a way that will be significantly easier to maintain. This shouldn't be taken as an indication that 'channel protocols' for Rust are not worth pursuing again in the future. Concerned parties may include: @graydon, @pcwalton, @eholk, @bblum The most likely candidates for closing are #7666, #3018, #3020, #7021, #7667, #7303, #3658, #3295.
new lint: `init-numbered-fields` This fixes rust-lang#7985. r? `@xFrednet` --- changelog: new lint: [`init_numbered_fields`]
Remove in_macro from clippy_utils changelog: none Previously done in rust-lang#7897 but reverted in rust-lang#8170. I'd like to keep `in_macro` out of utils because if a span is from expansion in any way (desugaring or macro), we should not proceed without understanding the nature of the expansion IMO. r? `@llogiq`
The pipes compiler produced data types that encoded efficient and safe
bounded message passing protocols between two endpoints. It was also
capable of producing unbounded protocols.
It was useful research but was arguably done before its proper time.
I am removing it for the following reasons:
oneshot
protcol andthe unbounded
stream
protocol and all communication in Rust use those.has a large surface area and was difficult to maintain through
language and library changes.
which would likely be considered non-idiomatic.
and likewise the relationship between the generated code and
the library is hard to understand. Debugging is difficult.
stream
andoneshot
by handin a way that will be significantly easier to maintain.
This shouldn't be taken as an indication that 'channel protocols'
for Rust are not worth pursuing again in the future.
Concerned parties may include: @graydon, @pcwalton, @eholk, @bblum
The most likely candidates for closing are #7666, #3018, #3020, #7021, #7667, #7303, #3658, #3295.