Make all backends controlled by features (using a very ugly workaround) by Wumpf · Pull Request #6949 · gfx-rs/wgpu

Wumpf · 2025-01-19T19:08:16Z

Connections

Fixes Add a default feature for wgpu that enables Vulkan & GLES on Windows & Linux #3514
Very ugly workaround for Platform target-specific features rust-lang/cargo#1197

Description
Feature flag changes:

new vulkan feature for enabling Vulkan on Windows/Android/Linux (part of default feature set)
new gles feature for enabling Vulkan on Windows/Android/Linux (part of default feature set, debatable if that's a good idea)
fix webgl enabling GL on all platforms

This is achieved by introducing three proxy crates. Each depends on wgpu-core under a different platform condition and will "forward" all feature flags to it.

Reminder why this horrible construct is the only way:
This is necessary because it's impossible today to drag in a dependency conditionally BOTH on a platform & a feature flag.
It is possible to do all these things in code, so we can forward features all the way to wgpu-hal and not have it compile e.g. it's vulkan backend on Mac iff vulkan-portability is off. But without this workaround it's impossible to avoid building vulkan associated dependencies anyways.
For more details see

Platform target-specific features rust-lang/cargo#1197

Building with default-features=false will now not even drag in wgpu-core! cargo tree -p wgpu --edges normal for simple application depending on wgpu this way:

wgpu v24.0.0 (D:\dev\wgpu\wgpu)
├── arrayvec v0.7.6
├── bitflags v2.8.0
├── document-features v0.2.10 (proc-macro)
│   └── litrs v0.4.1
├── log v0.4.25
├── parking_lot v0.12.3
│   ├── lock_api v0.4.12
│   │   └── scopeguard v1.2.0
│   └── parking_lot_core v0.9.10
│       ├── cfg-if v1.0.0
│       ├── smallvec v1.13.2
│       └── windows-targets v0.52.6
│           └── windows_x86_64_msvc v0.52.6
├── profiling v1.0.16
├── raw-window-handle v0.6.2
├── smallvec v1.13.2
├── static_assertions v1.1.0
└── wgpu-types v24.0.0 (D:\dev\wgpu\wgpu-types)
    ├── bitflags v2.8.0
    └── log v0.4.25

Similarly, only depending on dx12 (on Windows) doesn't include ash or glow (the former is a slow building dependency!):

wgpu v24.0.0 (D:\dev\wgpu\wgpu)
├── arrayvec v0.7.6
├── bitflags v2.8.0
├── document-features v0.2.10 (proc-macro)
│   └── litrs v0.4.1
├── log v0.4.25
├── parking_lot v0.12.3
│   ├── lock_api v0.4.12
│   │   └── scopeguard v1.2.0
│   └── parking_lot_core v0.9.10
│       ├── cfg-if v1.0.0
│       ├── smallvec v1.13.2
│       └── windows-targets v0.52.6
│           └── windows_x86_64_msvc v0.52.6
├── profiling v1.0.16
├── raw-window-handle v0.6.2
├── smallvec v1.13.2
├── static_assertions v1.1.0
├── wgpu-core v24.0.0 (D:\dev\wgpu\wgpu-core)
│   ├── arrayvec v0.7.6
│   ├── bit-vec v0.8.0
│   ├── bitflags v2.8.0
│   ├── document-features v0.2.10 (proc-macro) (*)
│   ├── hashbrown v0.15.2
│   │   └── foldhash v0.1.4
│   ├── indexmap v2.7.0
│   │   ├── equivalent v1.0.1
│   │   └── hashbrown v0.15.2 (*)
│   ├── log v0.4.25
│   ├── naga v24.0.0 (D:\dev\wgpu\naga)
│   │   ├── arrayvec v0.7.6
│   │   ├── bit-set v0.8.0
│   │   │   └── bit-vec v0.8.0
│   │   ├── bitflags v2.8.0
│   │   ├── codespan-reporting v0.11.1
│   │   │   ├── termcolor v1.4.1
│   │   │   │   └── winapi-util v0.1.9
│   │   │   │       └── windows-sys v0.59.0
│   │   │   │           └── windows-targets v0.52.6 (*)
│   │   │   └── unicode-width v0.1.14
│   │   ├── indexmap v2.7.0 (*)
│   │   ├── log v0.4.25
│   │   ├── rustc-hash v1.1.0
│   │   ├── strum v0.26.3
│   │   │   └── strum_macros v0.26.4 (proc-macro)
│   │   │       ├── heck v0.5.0
│   │   │       ├── proc-macro2 v1.0.93
│   │   │       │   └── unicode-ident v1.0.14
│   │   │       ├── quote v1.0.38
│   │   │       │   └── proc-macro2 v1.0.93 (*)
│   │   │       ├── rustversion v1.0.19 (proc-macro)
│   │   │       └── syn v2.0.96
│   │   │           ├── proc-macro2 v1.0.93 (*)
│   │   │           ├── quote v1.0.38 (*)
│   │   │           └── unicode-ident v1.0.14
│   │   ├── termcolor v1.4.1 (*)
│   │   └── thiserror v2.0.11
│   │       └── thiserror-impl v2.0.11 (proc-macro)
│   │           ├── proc-macro2 v1.0.93 (*)
│   │           ├── quote v1.0.38 (*)
│   │           └── syn v2.0.96 (*)
│   ├── once_cell v1.20.2
│   ├── parking_lot v0.12.3 (*)
│   ├── profiling v1.0.16
│   ├── raw-window-handle v0.6.2
│   ├── rustc-hash v1.1.0
│   ├── smallvec v1.13.2
│   ├── thiserror v2.0.11 (*)
│   ├── wgpu-hal v24.0.0 (D:\dev\wgpu\wgpu-hal)
│   │   ├── arrayvec v0.7.6
│   │   ├── bit-set v0.8.0 (*)
│   │   ├── bitflags v2.8.0
│   │   ├── gpu-allocator v0.27.0
│   │   │   ├── log v0.4.25
│   │   │   ├── presser v0.3.1
│   │   │   ├── thiserror v1.0.69
│   │   │   │   └── thiserror-impl v1.0.69 (proc-macro)
│   │   │   │       ├── proc-macro2 v1.0.93 (*)
│   │   │   │       ├── quote v1.0.38 (*)
│   │   │   │       └── syn v2.0.96 (*)
│   │   │   └── windows v0.58.0
│   │   │       ├── windows-core v0.58.0
│   │   │       │   ├── windows-implement v0.58.0 (proc-macro)
│   │   │       │   │   ├── proc-macro2 v1.0.93 (*)
│   │   │       │   │   ├── quote v1.0.38 (*)
│   │   │       │   │   └── syn v2.0.96 (*)
│   │   │       │   ├── windows-interface v0.58.0 (proc-macro)
│   │   │       │   │   ├── proc-macro2 v1.0.93 (*)
│   │   │       │   │   ├── quote v1.0.38 (*)
│   │   │       │   │   └── syn v2.0.96 (*)
│   │   │       │   ├── windows-result v0.2.0
│   │   │       │   │   └── windows-targets v0.52.6 (*)
│   │   │       │   ├── windows-strings v0.1.0
│   │   │       │   │   ├── windows-result v0.2.0 (*)
│   │   │       │   │   └── windows-targets v0.52.6 (*)
│   │   │       │   └── windows-targets v0.52.6 (*)
│   │   │       └── windows-targets v0.52.6 (*)
│   │   ├── hashbrown v0.15.2 (*)
│   │   ├── libloading v0.8.6
│   │   │   └── windows-targets v0.52.6 (*)
│   │   ├── log v0.4.25
│   │   ├── naga v24.0.0 (D:\dev\wgpu\naga) (*)
│   │   ├── once_cell v1.20.2
│   │   ├── ordered-float v4.6.0
│   │   │   └── num-traits v0.2.19
│   │   ├── parking_lot v0.12.3 (*)
│   │   ├── profiling v1.0.16
│   │   ├── range-alloc v0.1.4
│   │   ├── raw-window-handle v0.6.2
│   │   ├── rustc-hash v1.1.0
│   │   ├── thiserror v2.0.11 (*)
│   │   ├── wgpu-types v24.0.0 (D:\dev\wgpu\wgpu-types)
│   │   │   ├── bitflags v2.8.0
│   │   │   └── log v0.4.25
│   │   ├── windows v0.58.0 (*)
│   │   └── windows-core v0.58.0 (*)
│   └── wgpu-types v24.0.0 (D:\dev\wgpu\wgpu-types) (*)
└── wgpu-types v24.0.0 (D:\dev\wgpu\wgpu-types) (*)

Testing

separate test app with various config options
- TODO: try some more

Checklist

Run cargo fmt.
Run taplo format.
Run cargo clippy. If applicable, add:
- --target wasm32-unknown-unknown
- --target wasm32-unknown-emscripten
Run cargo xtask test to run tests.
Add change to CHANGELOG.md. See simple instructions inside file.

Draft todo:

do we even want this abomination
fix lotsa unused warnings
document proxy crates
should GL really be in the default feature set
test this on Mac & Linux by hand
check on emscripten and other corner cases
surely ci fails ;-)
write changelog

cwfitzgerald

I think this is broadly fine, I will have some comments on specifics once this is ready for review, but we definitely shouldn't be scared of crates. Might be worthwhile looking into cargo-release or release-plz or something if the count grows.

sagudev · 2025-01-19T20:47:44Z

Building with default-features=false will now not even drag in wgpu-core! cargo tree -p wgpu --edges normal for simple application depending on wgpu this way

This is important for #6330, I even impl it as faafd37.

brody2consult

quick comments from an outsider

Cargo.toml

wgpu/src/api/instance.rs

jimblandy · 2025-01-21T16:52:57Z

@Wumpf My impassioned request is that the explanation in the top of this PR make it into the code as a comment. I agree that this a good PR, but people are going to be utterly WTH when they see this.

jimblandy · 2025-01-21T16:53:33Z

Also: spectacular

cwfitzgerald · 2025-01-23T23:00:22Z

Sorry, kinda stepped on your toes a bit with #6980 - lmk if you need help managing the migration

TODO: we should test this on ci with cargo tree, it's too easy to sneak back in

Wumpf · 2025-01-26T17:17:08Z

hit a bit of a snag: any feature that is supposed to be forwarded to wgpu-core needs to be forwarded now to all four (yeah it's four by now) proxy crates. Doable, but extra annoying

The problem is that if I put wgpu-core = {workspace=true, optional=true} but then always address it with some-feature = ["wgpu-core?/some-feature"] it has no effect since after all nothing actually enables wgpu-core.

Wumpf changed the title ~~Make all backends controlled by features~~ Make all backends controlled by features (using a very very very ugly workaround) Jan 19, 2025

cwfitzgerald reviewed Jan 19, 2025

View reviewed changes

Wumpf changed the title ~~Make all backends controlled by features (using a very very very ugly workaround)~~ Make all backends controlled by features (using a very ugly workaround) Jan 19, 2025

brody2consult reviewed Jan 19, 2025

View reviewed changes

Cargo.toml Outdated Show resolved Hide resolved

wgpu/src/api/instance.rs Show resolved Hide resolved

cwfitzgerald mentioned this pull request Jan 23, 2025

CI/xtask for Validating State of Dependency Tree #6981

Closed

Wumpf force-pushed the make-remaining-backends-opt-out branch from d14c3a4 to df354a9 Compare January 25, 2025 17:20

Wumpf added 2 commits January 25, 2025 18:21

Use crate indirections to enable feature depending on platforms

f4e0d26

don't build with serde unless serde is actually enabled

0456cc6

TODO: we should test this on ci with cargo tree, it's too easy to sneak back in

Wumpf force-pushed the make-remaining-backends-opt-out branch from df354a9 to 0456cc6 Compare January 25, 2025 17:26

Wumpf added 3 commits January 26, 2025 17:09

fix wgpu_core cfg in wgpu not enabled with gles

9b35692

Fix enabled features for wgpu within crate

39ac474

fix dx12 enabling wgpu-core on linux - add another workaround crate

d24dbc9

Wumpf force-pushed the make-remaining-backends-opt-out branch from 31f1298 to d24dbc9 Compare January 26, 2025 16:24

... iterating

f0f47c4

This was referenced Feb 6, 2025

Alternative Solution to Dependency Fiasco #7064

Closed

Target Specific Compilation #7076

Merged

cwfitzgerald closed this Feb 22, 2025

Wumpf deleted the make-remaining-backends-opt-out branch February 22, 2025 11:35

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Make all backends controlled by features (using a very ugly workaround)#6949

Make all backends controlled by features (using a very ugly workaround)#6949
Wumpf wants to merge 6 commits intogfx-rs:trunkfrom
Wumpf:make-remaining-backends-opt-out

Wumpf commented Jan 19, 2025 •

edited

Loading

Uh oh!

cwfitzgerald left a comment

Uh oh!

sagudev commented Jan 19, 2025

Uh oh!

brody2consult left a comment

Uh oh!

Uh oh!

Uh oh!

jimblandy commented Jan 21, 2025 •

edited

Loading

Uh oh!

jimblandy commented Jan 21, 2025

Uh oh!

cwfitzgerald commented Jan 23, 2025

Uh oh!

Wumpf commented Jan 26, 2025

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

Conversation

Wumpf commented Jan 19, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

cwfitzgerald left a comment

Choose a reason for hiding this comment

Uh oh!

sagudev commented Jan 19, 2025

Uh oh!

brody2consult left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

jimblandy commented Jan 21, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

jimblandy commented Jan 21, 2025

Uh oh!

cwfitzgerald commented Jan 23, 2025

Uh oh!

Wumpf commented Jan 26, 2025

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

Wumpf commented Jan 19, 2025 •

edited

Loading

jimblandy commented Jan 21, 2025 •

edited

Loading