Summary:
The first time I tried to commit this changeset, it was causing many new packages to be installed, because the dependency would depend on newer versions that what we have installed. So, I had made a diff so upgrade all the babel packages. Unfortunately this caused some problem as the newer versions of Babel are more strict on some syntaxes. Of course, these have to be addressed, but I don't want this changeset to be coupled with Babel upgrades and the issues that arise from it.
So instead, I decided to install the slightly older version of the async-to-generator module. At first I tried with just doing:
yarn add babel-plugin-transform-async-to-generator@6.16.0
But, `yarn` is stubborn: because this module depends on a caret version of `babel-helper-remap-async-to-generator`, it installs the very last version of it, that itself needs more recent versions of other Babel modules. So, instead, I add to install a slightly older version of the dependency manually, then then the plugin:
yarn add babel-helper-remap-async-to-generator@6.16.0
yarn add babel-plugin-transform-async-to-generator@6.16.0
This allows us to have a `yarn.lock` with only a minimal amount of changes, and uncouple this change from any Babel upgrades. Because we only have a few new modules, the `node_modules` folder also stays the same, 133M, and it gives us confidence this will not cause significant startup time regressions.
Reviewed By: cpojer
Differential Revision: D4578733
fbshipit-source-id: deb0f720b895b7196aaf432adec3e56f18663940
Summary:
allow-large-files
By using async/await the code is (1) less nested, (2) more compact and (3) more robust (no exceptions running away, and much less risks of forgetting to call the callback/resolve, or mistakenly calling it twice). I now tend to think we could switch to it for all the callsites that are not in a perf-critical path.
I switched from 'request' to 'node-fetch' because 'request' has an annoying callback with 2 arguments. So it's simpler to use an interface that's (1) already returning a Promise and (2) that is becoming standard.
This changeset was a way for me to start experiment with introducing async/await in packager codebase, and it looks pretty good so far.
Reviewed By: cpojer
Differential Revision: D4559167
fbshipit-source-id: 89a328c5766c2ba890e9d0e67a81a38dac6cfc73
Summary: I figured out that the uncaught rejection that happens on transform error originated from a "fork" along the chain, where we both "then" on a promise, and return that promise. In that case Node.js, legitimately I think, identifies this as an uncaught rejection case. One solution, in this changeset, is to do the side-effects as part of the promise chain instead of "forking". Another option would be to add an explicit error handler to the additional "then", but it seems we don't have to handle that case here.
Reviewed By: davidaurelio
Differential Revision: D4515592
fbshipit-source-id: 17d813cfdbc76685b6273b8d94e97d948fd68674
Summary:
This moves the `src` directory one level up and removes the `react-packager` folder. Personally, I always disliked this indirection. I'm reorganizing some things in RNP, so this seems to make sense.
Not sure if I forgot to update any paths. Can anyone advice if there are more places that need change?
Reviewed By: jeanlauliac
Differential Revision: D4487867
fbshipit-source-id: d63f9c79d6238300df9632d2e6a4e6a4196d5ccb
Summary: We can simplify as there's only one option, and it seems to me we can also make it required, as otherwise the `JSTransformer` is essentially doing nothing. I don't believe we still have use cases where no transform happens at all, even in OSS?
Reviewed By: davidaurelio
Differential Revision: D4481723
fbshipit-source-id: b2e3830b206d56242d298ff3a7b5f4587ecfd29a
Summary: It's not used from any callsite as far as I can see.
Reviewed By: davidaurelio
Differential Revision: D4481708
fbshipit-source-id: 2c503fb7ef20f9370a950c315832f3ace4709739
Summary: it's not used from any callsite as far as I can see.
Reviewed By: davidaurelio
Differential Revision: D4481699
fbshipit-source-id: 4cf63ef7953d6cfc58e7ef4f22ecb99bf51e76a0
Summary: It's not used in any callsite.
Reviewed By: cpojer
Differential Revision: D4481683
fbshipit-source-id: 3fa55693f5f56b4fb6c455ad77d7780f69be81a9
Summary: The idea is to make it easier to interact with tools consuming the packager's output. For example, Nuclide. Do you think that'd work well?
Reviewed By: davidaurelio
Differential Revision: D4482041
fbshipit-source-id: 6c64d7963195a4d786ed8902640f9e9f279f5f83
Summary:
I don't know which version is better, but it should sync with `Resolver/polyfills/require.js: 168`
```js
// keep args in sync with with defineModuleCode in
// packager/react-packager/src/Resolver/index.js
// and packager/react-packager/src/ModuleGraph/worker.js
factory(global, require, moduleObject, exports, dependencyMap);
```
Closes https://github.com/facebook/react-native/pull/12033
Differential Revision: D4461635
Pulled By: davidaurelio
fbshipit-source-id: 1a17e6cbac5525dc09609b658bef82a23574b00c
Summary: Tests for things introduced over the last 10 weeks
Reviewed By: cpojer
Differential Revision: D4442404
fbshipit-source-id: cf7a14f237360fde083b7e2ffcd9d70aafd0ced6
Summary:
This replaces a polymorphic method with three separate methods that support the different used invocations.
In isolation, this leads to a 25% speedup.
Shoutout to trueadm for teaching me about this.
Reviewed By: trueadm
Differential Revision: D4436781
fbshipit-source-id: 3dce83556debf19d5305c4566a56b9e9565e85bc
Summary: Sets `generateSourceMaps` to true when requesting a bundle in non-dev mode or minified.
Reviewed By: bestander, cpojer
Differential Revision: D4415492
fbshipit-source-id: 9b4dcbd577a535dab143bb7a067210fbd831bf1a
Summary:
This fixes issues that only surface when running with node4 / npm2.
Closes https://github.com/facebook/react-native/pull/11888
Differential Revision: D4415021
Pulled By: bestander
fbshipit-source-id: 8672cd892c9dae41cc7b5b7b3227eb24dd780d22
Summary:
Introduces a new mechanism to build source maps that allows us to use real mapping segments instead of just mapping line-by-line.
This mechanism is only used when building development bundles to improve the debugging experience in Chrome.
The new mechanism takes advantage of a new feature in babel-generator that exposes raw mapping objects. These raw mapping objects are converted to arrays with 2, 4, or 5 for the most compact representation possible.
We no longer generate a source map for the bundle that maps each line to itself in conjunction with configuring babel generator to retain lines.
Instead, we create a source map with a large mappings object produced from the mappings of each individual file in conjunction with a “carry over” – the number of preceding lines in the bundle.
The implementation makes a couple of assumptions that hold true for babel transform results, e.g. mappings being in the order of the generated code, and that a block of mappings always belongs to the same source file. In addition, the implementation avoids allocation of objects and strings at all costs. All calculations are purely numeric, and base64 vlq produces numeric ascii character codes. These are written to a preallocated buffer objects, which is turned to a string only at the end of the building process. This implementation is ~5x faster than using the source-map library.
In addition to providing development source maps that work better, we can now also produce individual high-quality source maps for production builds and combine them to an “index source map”. This approach is unfeasable for development source maps, because index source map consistently crash Chrome.
Better production source maps are useful to get precise information about source location and symbol names when symbolicating stack traces from crashes in production.
Reviewed By: jeanlauliac
Differential Revision: D4382290
fbshipit-source-id: 365a176fa142729d0a4cef43edeb81084361e54d
Summary:
Adds a high performance source map builder that has certain restrictions compared to the `'source-map'` package:
- mappings have to be in the order of the generated source
- source files have to be started/ended separately on the generator. That means building up mappings is optimized for blocks of mappings that all belong to the same source file (or no file)
The implementation avoids allocation of complex value, i.e. strings and objects as much as possible by preallocating a buffer and using numeric character values throughout. The buffer is converted to a string only at the end.
This implementation is ~5✕ faster than using `'source-map'`.
Reviewed By: jeanlauliac
Differential Revision: D4392260
fbshipit-source-id: 406381302d951b919243a2b15e8bb75981e9f979
Jean Lauliac
Christoph Pojer
Jhen
Alexey Kureev
Bhuwan Khattar
Liu Zhanhong
Jing Chen
David Aurelio
Radek Czemerys
Gabe Levi
Andrew Imm