A Mental Model for C++ Coroutine(uvdn7.github.io) |
A Mental Model for C++ Coroutine(uvdn7.github.io) |
This means that c++26 is getting a default coroutine task type [1] AND a default executor [2]. You can even spawn the tasks like in Tokio/async Rust. [3]
I’m not totally sure if this is a GOOD idea to add to the c++ standard but oh well.
[1] https://www.open-std.org/jtc1/sc22/wg21/docs/papers/2025/p35...
[3] https://www.open-std.org/jtc1/sc22/wg21/docs/papers/2025/p31...
What are the downsides? Naively, it seems like a good idea to both provide a coroutine spec (for power users) and a default task type & default executor.
[1] https://devblogs.microsoft.com/oldnewthing/20210504-01/?p=10...
In general this moves way too fast for the density of the grammar it's trying to introduce, lines like:
> We have seen Awaitors already - suspend_always is an empty awaiter type that has await_ready returns false always.
But we haven't "seen" suspend_always, it's mentioned in half a sentence in an earlier paragraph, with no further context or examples.
There's a reason Lewis Baker's writings about C++ coroutines are 5000 word monsters, the body of grammar which needs to be covered demands that level of careful and precise definition and exploration.
A stackful coroutine is "write the live registers to your stack, swap the stack pointer to a suspended coroutine, load the old live registers from your new stack". It's a short and boring sequence of assembly.
A C++ coroutine is a CFG transform with a bunch of logic around heap allocation elision to construct something less capable than the above, with a bunch of keywords and semantics that you can kind of derive from the work the compiler needs to do to wire things together.
FWIW, I think a useful addition would be for compilers to output the intermediate source code, so you can reason more easily about behaviour and debug into readable code.
- At the transport layer, I read in a header on a message (which may come in one byte at a time!), get a size for the serialized message, then read N bytes for the message. The simple way to do things is to use a thread per socket, but that results in a lot of wasted memory, depending on how many sockets there are. Instead I use epoll, but now I can't make the simple for loop reading in bytes for the message - I have to have a buffer + allocated size + current size + state enum, wrapped in a struct, then run a switch statement every time I get an epoll event for the socket.
- At half a level higher, there might be multiple messages or other negotiations that need to happen before we can start to stream messages to the owner of the connection. Once again - need to either use a thread or a state enum to keep track of where we are.
Even if you want the enum to be able to report state, you can still set it somewhere for debug purposes.
Also you need compiler support to correctly handle thread_local.
[1] https://github.com/gpderetta/delimited/blob/master/delimited...
attribute((naked)) on a function which has a single asm block as the implementation gives you control over argument passing and changing the stack pointer.
attribute((preserve_none)) on the same function spills most live registers to the stack in the caller. The coroutine switch doesn't need to do as many push/pop which makes it a bit more readable, but mainly this means you don't spill dead registers. That's the big thing you need compiler support for.
I believe the x64 redzone is a non-issue here as you've called the switch function, as opposed to tried to call from within inline asm (which does need to be careful about that). The magic globals are a problem though (floating point control thing, maybe signal mask, errno et al) so I guess don't use the magic globals from within fibres.
"thread_local" doesn't map very sensibly onto fibres. There have been compiler bugs in that area too. Storing some information at the start of the fibre stack works fine though, you just don't get syntactic support for allocating / dereferencing from it.
Ideally I think that a ctx_t* __builtin_context_switch(ctx_t* to) would need to be provided by the compiler.
Re thread_local, I believe at least MSVC has (had?) a fiber-safe flag that would handle thread_locals correctly by not caching addresses across function calls.