Summary:
Pull Request resolved: https://github.com/facebook/yoga/pull/1506
In order to build the website, you must build a Yoga binary. This usually requires installing native toolchains.
We have warning messages for these right now, but an even better solution is to just manage the dependencies ourselves. This does that, bringing in specific CMake and Ninja binaries from NPM, and caching a local copy of Emscripten during the build.
A downside is that the CMake packages are chunky, so we add 130MB to node_modules (for a repo total around 350MB). This also delays acquiring Emscripten (which is even chunkier) in CI builds until it is needed, and I added some caching for it as well.
The upside of JS users being able to run and test (inc the website) without installing and managing their own versions of toolchains is a real time-saver though, and is probably worth it.
allow-large-files
Reviewed By: yungsters
Differential Revision: D52013026
fbshipit-source-id: 3d307f751463a21c5e5d5b98b8e9e63db9d3d52e
Summary:
Fixes https://github.com/facebook/yoga/issues/1417
This dramatically simplifies the matrix of Node vs web, ASM vs WASM, sync vs async compilation, or CommonJS vs ES Modules. We have one variant, using wasm, with ESModule top-level await to do async compilation. Web/node share the same binary, and we base64 encode the WASM into a wrapper JS file for compatibility with Node and bundlers.
This has some downsides, like requiring an environment with top level await, but also has upsides, like a consistent, sync looking API compatible with older Yoga, and mitigating TypeScript issues with package exports and typings resolution.
As part of this work I also removed `ts-node` from the toolchain (at the cost of a couple of config files needing to be vanilla JS).
Pull Request resolved: https://github.com/facebook/yoga/pull/1433
Test Plan:
1. `yarn test`
2. `yarn lint`
3. `yarn tsc`
4. `yarn benchmark`
5. `yarn build` website-next
6. `yarn lint` website-next
7. Locally test website-next
8. Examine package artifact created by GitHub
9. All Automation passes
Reviewed By: yungsters
Differential Revision: D50453324
Pulled By: NickGerleman
fbshipit-source-id: fe1192acc69e57fa69a1ff056dd7b5844d2198d5
Summary:
Pull Request resolved: https://github.com/facebook/yoga/pull/1382
X-link: https://github.com/facebook/react-native/pull/39437
Have been running into places where C++ 20 makes life easier for use like `std::bit_cast` (that one is easy to polyfill), in-class member initializer support for bitfields, designated initializers, defaulted comparison operator, concepts instead of SFINAE, and probably more.
Our other infra is in the process of making this jump, or already has. This tests it out everywhere, across the various reference builds, to see if we have any issues.
This is a bit more aggressive than I had previously communicated, but n - 1 is going to be a better long term place than n - 2.
If we wanted to use `std::bit_cast` we would need one of:
1. GCC 11+ (~2.5 years old)
1. Clang 14 (~2.5 years old)
1. VS 16.11 (~2 years old)
For mobile this means:
1. NDK 26 (still in Beta 😭)
1. XCode 14.3.0 (~6 months old)
https://en.cppreference.com/w/cpp/compiler_support/20
That isn't quite doable yet, but we can start taking advantage of language features in the meantime. More of these will be supported in older toolchains.
Anyone needing support for older C++ versions can lag behind on more recent changes. E.g. Yoga 2.0 supports C++ 14.
bypass-github-export-checks
Changelog: [Internal]
Reviewed By: cortinico
Differential Revision: D49261607
fbshipit-source-id: ceb06eac20dfe93352d7b796d6847a7314069cf3
Summary:
X-link: https://github.com/facebook/react-native/pull/38303
Pull Request resolved: https://github.com/facebook/yoga/pull/1327
This bumps Yoga to C++ 17 for a few reasons:
1. New versions of C++ may introduce behavior changes (e.g. evaluation order) and deprecations. Keeping the version closer to the version of large users helps avoid that.
2. C++ 17 unblocks some new bits I have wanted to use at times, like `std::optional`, `std::variant`, `if constexpr`, `[[nodiscard]]`.
3. There are already changes in C++ 20 that would be directly useful to Yoga, like `std::bit_cast` to avoid `memcpy` style type punning.
There has been some contention around C++ versions before, but by the time the next stable version of Yoga is out, it will have been more than 6 years (~2 C++ versions) since a stable version of Clang/LLVM with C++ 17 support. I would not like to go back further than n-2.
Changelog: [Internal]
Reviewed By: christophpurrer
Differential Revision: D47383922
fbshipit-source-id: eb95d4853f2168b68d6df5fddb797236eac55870
Summary:
Pull Request resolved: https://github.com/facebook/yoga/pull/1310
Emscripten normally compiles a binary into a `.js` file and a `.wasm` file. The `.js` file contains a shim to load the WebAssembly file for the target platform, along with passing some environment information to the underlying assembly.
Under Node this would use APIs like `fs.readFile` and its WebAssembly APIs to load the binary. In a browser, APIs like `instantiateStreaming` are used to start downloading and compiling the binary at the same time.
This format creates many, many, headaches, and manual bundler configuration. E.g. we must tell Webpack to treat WASM files as auxilary files instead of WebAssembly, cannot use Emscripten's loader directly, and would need to add more variants of the binary, since (or Node polyfills in the browser) `-s ENVIRONMENT='web,node'` emits code that looks like `if (isNode) {require('fs')}`.
This change makes us instead pack the WebAssembly as base64 inline with the JS loader. This adds a size penalty, and means we cannot start async compilation until the entire file is present, but should work out of the box when using different bundlers and configurations, and the size is small enough where it likely makes sense to inline into the bundle anyway.
There is a [proposal for integration of WebAssembly and ES Modules](https://github.com/WebAssembly/esm-integration/tree/main/proposals/esm-integration) that Node has experimental support for, and bundlers are veering towards supporting. It is the eventual solution we should target, but does not seem mature enough yet. E.g. WebPack [does not support](https://github.com/webpack/webpack/issues/11893) WebAssembly import objects, and will instead try to import each of the named imports as modules.
Reviewed By: rozele
Differential Revision: D46884398
fbshipit-source-id: a1c93c122c255b913f426bfb6bdb38fd9f9dfd41
Summary:
X-link: https://github.com/facebook/react-native/pull/37349
Pull Request resolved: https://github.com/facebook/yoga/pull/1288
Fixes https://github.com/facebook/yoga/issues/1283
New versions of CMake add "policies" which control how the build system acts wrt breaking changes. By default, CMake will emulate the behavior of the version specified in `cmake_minimum_required`.
Setting a policy to true (to opt into new behavior where `cmake_minimum_required` is lower than the current version) seems actually just error out on the old versions.
Googling around, apparently the way I should be doing this is to specify `<policy_max>` as part of `cmake_minimum_required `. https://gitlab.kitware.com/cmake/cmake/-/issues/20392
This should I think use new policies introduced up to 3.26 (what we test on right now), while letting 3.13 be the minimum.
Reviewed By: cortinico
Differential Revision: D45724864
fbshipit-source-id: 120cc2015a043605e7c07ef0459667643a4284b7
Summary:
Pull Request resolved: https://github.com/facebook/yoga/pull/1284
This makes TypeScript resolution play nicely with export maps, and converts the entrypoints to TypeScript.
We remove the non-export-map fallbacks as well, so the export maps are always followed.
Tests are moved to load yoga from its external export, testing the entrypoints.
This moves the only untyped bit to the binary wrapper, which another diff will move.
Reviewed By: yungsters
Differential Revision: D45713689
fbshipit-source-id: 228e6f2c9d53520230707a81e76c3c35fcd46de5
Summary:
This change restructures the package to try to remove the JS build step from the inner loop. Instead, we have a single `src` directory that we babel transform when using, then apply the same transform inline during prepublish.
At the end, we will be publishing a source directory with Babel transformed TS, JS, and TS declarations.
We do a little spring cleaning when doing this. Fixing up some of the folder/file conventions, and removing the non-export-map fallbacks.
We cannot remove the need for a native build.
Reviewed By: yungsters
Differential Revision: D45682153
fbshipit-source-id: ea2dd75c2dd6e3529b1ef6cf6ac6a64a270049a4
Summary:
Pull Request resolved: https://github.com/facebook/yoga/pull/1217
This updates the CMake build present for being able to share options, fixing up flags, etc. A GTest build is added as well, along with a script and VSCode debug target so that OSS contributors can very easily run and debug tests on any OS.
Note that this isn't completely done (need to revise Windows, Mac, documentation), but should be finished enough otherwise for review.
Changelog: [Internal]
Reviewed By: cortinico
Differential Revision: D42406686
fbshipit-source-id: 95e7ba5e4751c496a171785490e85cf0097fa839
Summary:
Pull Request resolved: https://github.com/facebook/yoga/pull/1203
~~This sets the C++ standard to C++ 17 in the working builds and Apple. GTest will stop supporting C++ 11 soon, so we need to update. C++ 14 is more embeddable, but C++ 17 support and usage should be relatively common now and the language version adds quite a bit.~~
This bumps from C++ 11 to C++ 14 in existing places where it is specified. C++ 17 allows more, and is better aligned to infra (semantics can change in std versions in suprising ways), but C++ 14 still has broader ecosystem compatibility.
Changelog: [Internal]
Reviewed By: cortinico, dmytrorykun
Differential Revision: D42285391
fbshipit-source-id: 88d7b6b8783a80b9b2e48781a2fd3d326ecd87d0
Summary:
Yoga's JavaScript bindings do not work past Node 10, or on recent versions of Ubuntu even using it. This is due to a reliance on `nbind`, a library which is no longer maintained. `nbind` itself abstracts over `embind` running Emscripten to generate an asm.js build, along with building Node native modules. In the meantime, [yoga-layout-prebuilt](https://www.npmjs.com/package/yoga-layout-prebuilt) has been used by the community instead of the official package.
https://github.com/facebook/yoga/pull/1177 was contributed as a conversion of bindings created using `nbind` to instead use `embind` directly.
I continued building on this to add more:
1. WebAssembly support (required to be async in browsers)
2. CMake + Ninja Build for the 4 flavors
3. TypeScript typings (partially generated)
4. yarn scripts to build (working on macOS, Ubuntu, Windows)
5. A README with some usage and contribution instructions
6. Updated tests to work with Jest, and updated general infra
7. ESLint and clang-format scripts
8. More GitHub actions (and now testing Windows)
9. Probably more I kinda got carried away here lol
The plan is to eventually publish this to NPM, but there is a little bit of work after this before that happens.
Pull Request resolved: https://github.com/facebook/yoga/pull/1177
Test Plan: The bindings pass Jest tests (both manual and generated). GitHub actions added for the different yarn scripts. Did some manual checks on using the library as TS.
Reviewed By: christophpurrer
Differential Revision: D42207782
Pulled By: NickGerleman
fbshipit-source-id: 1dc5ce440f1c2b9705a005bbdcc86f952785d94e
