Implement P3149R11 and P3815R1 #1713

ispeters · 2025-12-14T07:18:48Z

This PR will implement P3149R11 patched with P3815R1.

[exec.scope.concepts]
- concept scope_association
- concept scope_token
[exec.associate]
- associate algorithm
[exec.stop.when]
- stop-when algorithm
[exec.spawn]
- spawn algorithm
[exec.spawn.future]
- spawn_future algorithm
[exec.counting.scopes]
- simple_counting_scope
- counting_scope

copy-pr-bot · 2025-12-14T07:18:51Z

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

include/stdexec/__detail/__scope_concepts.hpp

ericniebler · 2025-12-15T16:56:18Z

/ok to test fd9d04b

ispeters · 2025-12-16T17:10:44Z

The latest push includes an implementation of stdexec::associate with some tests, but the tests are incomplete so I haven't marked associate as done in the summary.

I have also tried to address the build failures on the previous iteration, but I don't have a local test environment for the failing build types so I haven't tested the fixes locally, other than to confirm that they still build with my local Clang (i.e. Homebrew clang version 21.1.7).

ispeters · 2025-12-17T04:11:55Z

Rebased on #1717.

ericniebler · 2025-12-22T07:14:10Z

/ok to test f786c00

ericniebler · 2025-12-23T22:24:56Z

/ok to test 97d79ee

ispeters · 2025-12-24T03:35:59Z

I managed to get GCC 12 installed on my Mac and it repro'd the ICE and the build failure in test_stop_when.cpp; I think I've addressed the build failures in GCC, nvcc, and MSVC.

ispeters · 2025-12-24T08:09:46Z

simple_counting_scope still needs more tests, but I did discover that [exec.simple.counting.mem], paragraph 9 has a bug: we need to move the scope to the joined state and return true if count is zero, regardless of the value of state.

ericniebler · 2025-12-24T23:55:35Z

/ok to test cc867f4

ispeters · 2025-12-25T07:37:11Z

I think I've fixed the build breaks in old gcc, nvc++, and MSVC. Getting a repro of the gcc 11 builds was quite a pain—I can't figure out how to get gcc 11 to run properly on macOS Tahoe so I had to use container, which I learned about here. Fixing MSVC is a matter of guessing. I can't run nvc++ on my Mac, but it turns out it's available on Godbolt, and this was very helpful.

I've updated __counting_scopes.hpp rather extensively; I renamed some of the variables in simple_counting_scope to make the naming more consistent, and I've thoroughly documented both the paragraphs in the standard being implemented and my justifications for weaker-than-seq_cst memory orderings.

ispeters · 2025-12-25T20:51:37Z

Pull the guts of simple_counting_scope into a __base_scope and implement both simple_counting_scope and counting_scope in terms of it.

ispeters · 2025-12-25T21:36:34Z

Do some template hoisting and renaming, and replace a single virtual function with a function pointer to cut down on RTTI.

ispeters · 2026-01-01T08:50:33Z

Rebase on a bunch of recent commits.
Incorporate @ericniebler's "add missing include" change into 401944f and update authorship attribution
Simplify the implementation of forwarding stop requests by only invoking __abandon if the operation is never started
Still haven't addressed the ASAN error, though

ericniebler · 2026-01-01T18:28:46Z

/ok to test a7c40ee

ericniebler · 2026-01-02T20:04:34Z

/ok to test 3c160e1

ericniebler · 2026-01-03T23:22:57Z

/ok to test c5ee582

ispeters · 2026-01-19T03:57:50Z

I just force-pushed after rebasing on main.

ispeters · 2026-01-19T23:24:29Z

I've re-jiggered the implementation of stop request handling in spawn_future; I need to hand over the computer to my kid now, so I haven't had a chance to re-review my own code like I usually do, but I think this cleans up the implementation some, at the cost of another virtual function on __future_state_base.

ericniebler

this is looking amazing. minor comments only. lmk if you want a hand with merge conflicts.

include/stdexec/__detail/__associate.hpp

ericniebler · 2026-01-27T18:50:19Z

include/stdexec/__detail/__associate.hpp

+      __associate_data(__associate_data&& __other)
+        noexcept(__nothrow_move_constructible<__wrap_sender_t>)


move operations should all be noexcept unconditionally. you can then static_assert that __nothrow_move_constructible<__wrap_sender_t> is true.

What if I write just(std::list<int>{})? I believe that, at least in some STLs, std::list doesn't have a no-throw move constructor….

ericniebler · 2026-01-27T18:53:05Z

include/stdexec/__detail/__associate.hpp

+        }
+      };
+
+      using __sender_ref = std::unique_ptr<__wrap_sender_t, __deleter>;


using a unique_ptr for this seems like overkill. my preference would be to use the __scope_guard in __detail/__scope.hpp instead. something like:

__scope_guard __g{std::destroy_at<_Sender>, &__sndr};

This has been written into the spec; I think it's required unless you propose we change the wording.

include/stdexec/__detail/__associate.hpp

include/stdexec/__detail/__spawn_future.hpp

ericniebler · 2026-02-04T23:10:57Z

include/stdexec/__detail/__spawn_future.hpp

+      void __do_consume(auto& __rcvr) noexcept {
+        using __variant_t = decltype(this->__result_);
+
+        __variant_t::visit(


Suggested change

__variant_t::visit(

__visit(

include/stdexec/__detail/__spawn_future.hpp

ericniebler · 2026-02-04T23:22:57Z

include/stdexec/__detail/__stop_when.hpp

+          return __apply(
+            [&__rcvr](__ignore, auto&& token, __ignore) noexcept {
+              return __make_token_fn{}(
+                static_cast<decltype(token)>(token), get_stop_token(STDEXEC::get_env(__rcvr)));
+            },
+            static_cast<_Self&&>(__self));


i think this can now be (untested):

auto &[__tag, __token, __child] = __self; return __make_token_fn{}(__forward_like<_Self>(__token), get_stop_token(STDEXEC::get_env(__rcvr)));

ispeters

I've applied much of the suggested feedback and left a few questions.

Still to do:

uglify all the identifiers
convert virtual functions to stored function pointers
clean up the implementation of spawn_future
address any unaddressed comments

ispeters · 2026-02-10T00:02:12Z

include/stdexec/__detail/__associate.hpp

+        }
+      };
+
+      using __sender_ref = std::unique_ptr<__wrap_sender_t, __deleter>;


This has been written into the spec; I think it's required unless you propose we change the wording.

ispeters · 2026-02-10T00:02:54Z

include/stdexec/__detail/__associate.hpp

+      __associate_data(__associate_data&& __other)
+        noexcept(__nothrow_move_constructible<__wrap_sender_t>)


What if I write just(std::list<int>{})? I believe that, at least in some STLs, std::list doesn't have a no-throw move constructor….

include/stdexec/__detail/__associate.hpp

ispeters · 2026-02-10T00:07:22Z

include/stdexec/__detail/__spawn.hpp

+
+      __spawn_state_base(__spawn_state_base&&) = delete;
+
+      virtual void __complete() noexcept = 0;


Yeah, I didn't get that far, yet.

ispeters · 2026-02-10T00:18:29Z

include/stdexec/__detail/__counting_scopes.hpp

+      static consteval auto get_completion_signatures() //
+        -> __completion_signatures_of_t<__sched_sender_of_t<_Env>, _Env> {
+        return {};


The proposed change breaks a test that requires that the join-sender have a scheduler in its receiver's environment. Keeping the deduced return type (which I expect to SFINAE away this function when appropriate) and adding the call to get_completion_signatures brings the test back to passing. Is that expected?

This diff adds a definition for `concept stdexec::scope_association` plus tests confirming it accepts and rejects the expected things.

This diff adds a definition for `concept stdexec::scope_token` plus tests confirming it accepts and rejects the expected things. Co-authored-by: Eric Niebler <eniebler@nvidia.com>

This diff defines `stdexec::associate` and adds some initial tests to confirm it works properly. Still a work in progress.

This diff defines `stdexec::__stop_when` as the implementation of _`stop-when`_ and adds tests to validate the algorithm. Co-authored-by: Eric Niebler <eniebler@nvidia.com>

This diff adds `stdexec::spawn` and its tests. Co-authored-by: Eric Niebler <eniebler@nvidia.com>

This diff adds `stdexec::simple_counting_scope` and `stdexec::counting_scope` plus tests for both. The tests found a spec bug that's been filed as an LWG issue. Co-authored-by: Eric Niebler <eniebler@nvidia.com>

This diff adds `stdexec::spawn_future` and some basic tests. Co-authored-by: Eric Niebler <eniebler@nvidia.com>

This diff probably implements the intended design that stop requests sent to a started future-sender get forwarded to the spawned work. The existing tests still pass, but I don't have new tests to confirm the new behaviour. Co-authored-by: Eric Niebler <eniebler@nvidia.com>

ispeters · 2026-02-10T00:48:18Z

Ok, I think this is correct-ish on top of a very recent main. There's still work to do, but I need to share the laptop. More tomorrow.

ericniebler · 2026-02-10T02:40:07Z

/ok to test aeed749

Wrap the call to `std::numeric_limits<std::size_t>::max` in parens so that Microsoft's `max` macro doesn't get in the way.

ispeters force-pushed the p3149_plus_p3815 branch from db2f61a to 9327406 Compare December 15, 2025 00:08

ericniebler reviewed Dec 15, 2025

View reviewed changes

include/stdexec/__detail/__scope_concepts.hpp Show resolved Hide resolved

ispeters force-pushed the p3149_plus_p3815 branch from fd9d04b to 707aed3 Compare December 16, 2025 17:09

ispeters force-pushed the p3149_plus_p3815 branch from 707aed3 to 66f394c Compare December 17, 2025 04:10

ispeters force-pushed the p3149_plus_p3815 branch 2 times, most recently from 96d4661 to f786c00 Compare December 22, 2025 06:42

ispeters force-pushed the p3149_plus_p3815 branch 3 times, most recently from 67262ee to 97d79ee Compare December 23, 2025 21:08

ispeters force-pushed the p3149_plus_p3815 branch from 97d79ee to 95d10a7 Compare December 24, 2025 03:31

ispeters force-pushed the p3149_plus_p3815 branch 2 times, most recently from 52e1755 to 521d849 Compare December 24, 2025 08:06

ispeters force-pushed the p3149_plus_p3815 branch from 521d849 to cc867f4 Compare December 24, 2025 20:05

ispeters force-pushed the p3149_plus_p3815 branch 2 times, most recently from a1dd153 to 9263dab Compare December 25, 2025 07:32

ispeters force-pushed the p3149_plus_p3815 branch from 9263dab to 254887d Compare December 25, 2025 20:51

ispeters force-pushed the p3149_plus_p3815 branch from 254887d to f78a329 Compare December 25, 2025 21:35

ispeters force-pushed the p3149_plus_p3815 branch from f78a329 to e6db1ed Compare December 25, 2025 22:07

ispeters force-pushed the p3149_plus_p3815 branch from 3c160e1 to c5ee582 Compare January 3, 2026 23:19

ccotter mentioned this pull request Jan 6, 2026

async scope bug #1731

Closed

ispeters force-pushed the p3149_plus_p3815 branch from c5ee582 to 2ac3fb0 Compare January 19, 2026 03:56

ispeters force-pushed the p3149_plus_p3815 branch from 2ac3fb0 to ed8e376 Compare January 19, 2026 23:23

ericniebler reviewed Feb 4, 2026

View reviewed changes

ispeters force-pushed the p3149_plus_p3815 branch 4 times, most recently from 8b988cb to a0be249 Compare February 10, 2026 00:33

ispeters commented Feb 10, 2026

View reviewed changes

ispeters and others added 6 commits February 9, 2026 16:35

Add stdexec::scope_association

861602a

This diff adds a definition for `concept stdexec::scope_association` plus tests confirming it accepts and rejects the expected things.

Add stdexec::scope_token

003f02a

This diff adds a definition for `concept stdexec::scope_token` plus tests confirming it accepts and rejects the expected things. Co-authored-by: Eric Niebler <eniebler@nvidia.com>

Add stdexec::associate

f030a14

This diff defines `stdexec::associate` and adds some initial tests to confirm it works properly. Still a work in progress.

Add stdexec::__stop_when

ce6397c

This diff defines `stdexec::__stop_when` as the implementation of _`stop-when`_ and adds tests to validate the algorithm. Co-authored-by: Eric Niebler <eniebler@nvidia.com>

Add stdexec::spawn

cf8cad8

This diff adds `stdexec::spawn` and its tests. Co-authored-by: Eric Niebler <eniebler@nvidia.com>

Add stdexec::simple_counting_scope and stdexec::counting_scope

1b50529

This diff adds `stdexec::simple_counting_scope` and `stdexec::counting_scope` plus tests for both. The tests found a spec bug that's been filed as an LWG issue. Co-authored-by: Eric Niebler <eniebler@nvidia.com>

ispeters force-pushed the p3149_plus_p3815 branch from a0be249 to 0033e83 Compare February 10, 2026 00:35

ispeters and others added 2 commits February 9, 2026 16:46

Add stdexec::spawn_future

e614bec

This diff adds `stdexec::spawn_future` and some basic tests. Co-authored-by: Eric Niebler <eniebler@nvidia.com>

ispeters force-pushed the p3149_plus_p3815 branch from 0033e83 to 735ef43 Compare February 10, 2026 00:47

fixes

aeed749

Fix MSVC build

f090ee6

Wrap the call to `std::numeric_limits<std::size_t>::max` in parens so that Microsoft's `max` macro doesn't get in the way.

		__associate_data(__associate_data&& __other)
		noexcept(__nothrow_move_constructible<__wrap_sender_t>)


		__spawn_state_base(__spawn_state_base&&) = delete;

		virtual void __complete() noexcept = 0;

Implement P3149R11 and P3815R1 #1713

Are you sure you want to change the base?

Implement P3149R11 and P3815R1 #1713

Uh oh!

Conversation

ispeters commented Dec 14, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

copy-pr-bot bot commented Dec 14, 2025

Uh oh!

Uh oh!

ericniebler commented Dec 15, 2025

Uh oh!

ispeters commented Dec 16, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

ispeters commented Dec 17, 2025

Uh oh!

ericniebler commented Dec 22, 2025

Uh oh!

ericniebler commented Dec 23, 2025

Uh oh!

ispeters commented Dec 24, 2025

Uh oh!

ispeters commented Dec 24, 2025

Uh oh!

ericniebler commented Dec 24, 2025

Uh oh!

ispeters commented Dec 25, 2025

Uh oh!

ispeters commented Dec 25, 2025

Uh oh!

ispeters commented Dec 25, 2025

Uh oh!

ispeters commented Jan 1, 2026

Uh oh!

ericniebler commented Jan 1, 2026

Uh oh!

ericniebler commented Jan 2, 2026

Uh oh!

ericniebler commented Jan 3, 2026

Uh oh!

ispeters commented Jan 19, 2026

Uh oh!

ispeters commented Jan 19, 2026

Uh oh!

ericniebler left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

ispeters left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

ispeters commented Feb 10, 2026

ispeters commented Dec 14, 2025 •

edited

Loading

ispeters commented Dec 16, 2025 •

edited

Loading