Implement P3149R11 and P3815R1 by ispeters · Pull Request #1713 · NVIDIA/stdexec

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.

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>

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.

* Eliminate camel case * Uglify all identifiers * `s/\bassert\b/STDEXEC_ASSERT/g` * Take the address of objects with `std::addressof`

ericniebler · 2026-02-10T19:17:12Z

/ok to test 2775f7c

@ericniebler

Take @ericniebler's suggestion and use `__mapply_q` in `__spawn_future.hpp`.

@ericniebler

Take @ericniebler's suggestion and replace a constrained overload set with a single overload with an internal cascade of constexpr ifs to control the environment selection logic in `__spawn_common.hpp` and to control the stop token construction in `__stop_when.hpp`.

Replace the pure-virtual function in `__spawn_state_base` with a function pointer.

Replace the pure-virtuals in `__try_cancelable` and `__spawn_future_state_base` with function pointers.

@ericniebler

Take @ericniebler's suggestion for a better work-around for what might be a spec bug in howthe base class of `__spawn_future_state` is computed.

Since the stdexec-specific `__variant` type has an uninitialized state, we don't need the `__monostate` variant option to represent "uninitialized". Delete it.

ispeters · 2026-02-10T20:06:54Z

There are still a few things I'd like to improve, but this is probably landable.

ericniebler · 2026-02-10T20:58:15Z

/ok to test 9199dcc

Explicitly specify the template type parameters when constructing a `__fused_token` in `__make_token_fn`.

Uncollide with MSVC's `__nothrow` attribute by renaming `__nothrow` to `__non_throwing`.

ericniebler · 2026-02-10T21:15:42Z

/ok to test 892f971

ericniebler · 2026-02-10T21:56:04Z

/ok to test bdb7f09

ericniebler · 2026-02-10T23:11:25Z

/ok to test 3303429

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 and others added 11 commits February 10, 2026 11:14

Add stdexec::scope_association

8791891

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

Add stdexec::scope_token

86e22ef

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

2c65dac

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

Add stdexec::__stop_when

08d1308

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

29c0b28

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

401f608

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>

Add stdexec::spawn_future

a63340d

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

fixes

94c4874

Fix MSVC build

21d29e7

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.

Conform to coding standard

2775f7c

* Eliminate camel case * Uglify all identifiers * `s/\bassert\b/STDEXEC_ASSERT/g` * Take the address of objects with `std::addressof`

ispeters force-pushed the p3149_plus_p3815 branch from 31ccf35 to 2775f7c Compare February 10, 2026 19:14

ispeters added 5 commits February 10, 2026 11:19

Use __mapply_q

6dbc80a

Take @ericniebler's suggestion and use `__mapply_q` in `__spawn_future.hpp`.

Devirtualize __spawn_state_base

22fd39d

Replace the pure-virtual function in `__spawn_state_base` with a function pointer.

Devirtualize spawn_future

fcb41d0

Replace the pure-virtuals in `__try_cancelable` and `__spawn_future_state_base` with function pointers.

Update the completion signature computation for __spawn_future_state

afd6e39

Take @ericniebler's suggestion for a better work-around for what might be a spec bug in howthe base class of `__spawn_future_state` is computed.

ispeters force-pushed the p3149_plus_p3815 branch from 9bd9635 to afd6e39 Compare February 10, 2026 20:02

Remove __monostate from future's completion variant

9199dcc

Since the stdexec-specific `__variant` type has an uninitialized state, we don't need the `__monostate` variant option to represent "uninitialized". Delete it.

ispeters marked this pull request as ready for review February 10, 2026 20:06

ispeters added 2 commits February 10, 2026 13:13

Fix Clang 16 build

892f971

Explicitly specify the template type parameters when constructing a `__fused_token` in `__make_token_fn`.

Fix MSVC build

bdb7f09

Uncollide with MSVC's `__nothrow` attribute by renaming `__nothrow` to `__non_throwing`.

__callback is magic on msvc

3303429

ericniebler merged commit b62ecdf into NVIDIA:main Feb 10, 2026
22 checks passed

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!

ericniebler commented Feb 10, 2026

Uh oh!

ispeters commented Feb 10, 2026

Uh oh!

ericniebler commented Feb 10, 2026

Uh oh!

ericniebler commented Feb 10, 2026

Uh oh!

ericniebler commented Feb 10, 2026

Uh oh!

ericniebler commented Feb 10, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

ispeters commented Dec 14, 2025 •

edited

Loading

ispeters commented Dec 16, 2025 •

edited

Loading