Summary:
With this, we do all the transformation and wrapping of files inside the workers, which mean faster initial builds (because parallelization and caching), and more legacy code that can be removed (see the following diff).
I've done some tests locally, and while the initial builds are slightly faster, the increase is not super substantial (the big win was in the diff were I moved the wrapping of JS files, in this diff only the assets transformation has speed up).
The most important thing that this diff enables is the possibility of doing the minification of modules also in the worker. This will mean that we can cache minified files and prod builds will get significantly faster - almost as fast as development builds (this will be relevant mainly for the opensource community).
Reviewed By: davidaurelio
Differential Revision: D6439144
fbshipit-source-id: ba2200569a668fcbe68dbfa2b3b60b3db6673326
Summary: This fixes https://github.com/facebook/metro/issues/99, which causes issues when metro tries to build files that are ignored by babel (if babel transformer ignores a file, it returns a null AST, which makes the collectDependencies logic break).
Reviewed By: mjesun
Differential Revision: D6466878
fbshipit-source-id: b5030e03775b982958a0b9204f4efccc8940ae4d
Summary: I'm working on getting CI to pass. As a first step, I'll upgrade the lerna setup to use Yarn's workspaces (when yarn is run from the Metro root) as well as upgrading Flow to the same version we use in xplat. I also copied over the Jest type definitions. This should fix all type errors for a start.
Reviewed By: davidaurelio
Differential Revision: D6361276
fbshipit-source-id: 4e8661b7d5fe4e3f6dd1e6923891bd2d23c9b4db
Summary:
<!-- Thanks for submitting a pull request! Please provide enough information so that others can review your pull request. The two fields below are mandatory. -->
**Summary**
<!-- Explain the **motivation** for making this change. What existing problem does the pull request solve? -->
It is currently not possible to resolve specific module imports (such as `react-native/Libraries/Image/AssetRegistry` using forward slashes as folder separators) using a custom mapping defined in `extraNodeModules` on Windows. This PR solves this issue by normalizing module names to use platform-specific folder separators before splitting the module name using `path.sep` as separators.
**Test plan**
<!-- Demonstrate the code is solid. Example: The exact commands you ran and their output, screenshots / videos if the pull request changes UI. -->
We use `extraNodeModules` to create a mapping between `react-native` and a forked and scoped version of react-native (i.e. `scope/react-native`). If we import a specific file from react-native (such as [`react-native/Libraries/Image/AssetRegistry`](https://github.com/facebook/react-native/blob/master/local-cli/core/Constants.js), which gets referenced when importing image assets), `ModuleResolution.js` is not able to extract the first folder name on Windows. This first folder name is the name of the module (`react-native` in the previous example) and is used to query `extraNodeModules` for possible matches. It is not able to find this folder name because the module name is split using `path.sep` as a separator, which is `'\\'` on Windows. Most module names use forward slashes as folder separators. The solution is to normalize `toModuleName` before we split on `path.sep`.
Closes https://github.com/facebook/metro-bundler/pull/89
Differential Revision: D6312391
Pulled By: jeanlauliac
fbshipit-source-id: 920c52633e8c9584ecb2bdd309dc4a8516c3199b
Summary:
Uglify@3.1.7 is causing some perf issues (more specifically, this commit: 5b4b07e9a7) that are impacting TTI on RN views and increased memory usage.
More specifically, code like:
```
function nonTrivialFn(pre) {
return pre + Math.random();
}
function hotFunctionCalledALotOfTimes(num) {
return nonTrivialFn(num + Math.random());
}
hotFunctionCalledALotOfTimes(3);
hotFunctionCalledALotOfTimes(5);
```
in v3.1.7 gets converted to:
```
function hotFunctionCalledALotOfTimes(num){
return function(pre) {
return pre + Math.random();
}(num + Math.random())
}
hotFunctionCalledALotOfTimes(3);
hotFunctionCalledALotOfTimes(5);
```
This causes a function creation each time `hotFunctionCalledALotOfTimes` is called.
By comparison, in v3.1.6, that previous code was converted to:
```
function nonTrivialFn(pre){
return pre + Math.random()
}
function hotFunctionCalledALotOfTimes(num){
return nonTrivialFn(num + Math.random())
}
hotFunctionCalledALotOfTimes(3);
hotFunctionCalledALotOfTimes(5);
```
Reviewed By: jeanlauliac, alexeylang
Differential Revision: D6296740
fbshipit-source-id: b3988d886e607103ec3ae6b9763b2f0411a8aa3c
Summary:
`metro-bundler` v0.21 contains a rewritten bundling mechanism, with simplified logic and much faster rebuild times, called delta bundler. This release contains a couple of breaking changes:
* Now, when using a custom transformer, the list of additional babel plugins to apply are passed to the `transform()` method. These are used in non-dev mode for optimization purposes (Check 367a5f5db8 (diff-40653f0c822ac59a5af13d5b4ab31d84) to see how to handle them from the transformer).
* Now, when using a custom transformer outputting `rawMappings`, the transformer does not need to call the `compactMappings` method before returning (check d74685fd1d (diff-40653f0c822ac59a5af13d5b4ab31d84) for more info).
* We've removed support for two config parameters: `postProcessModules` and `postProcessBundleSourcemap`.
Reviewed By: davidaurelio
Differential Revision: D6186035
fbshipit-source-id: 242c5c2a954c6b9b6f339d345f888eaa44704579
Summary:
**Summary**
Minification fails or minified bundle may crash due to uglify-es bugs which have been fixed recently. See https://github.com/facebook/react-native/issues/16689
**Test plan**
Try to production bundle a project using ex-navigation, which fails with:
```
Maximum call stack size exceeded
```
Use this patch and see that bundling suceeds. There are also minified runtime errors solved by this change, see https://github.com/facebook/react-native/issues/16689 for more information.
Closes https://github.com/facebook/metro-bundler/pull/85
Reviewed By: mjesun
Differential Revision: D6259177
Pulled By: rafeca
fbshipit-source-id: 55987eb338b06938181c0da74d104d23eeb135b6
Summary:
This diff migrates Metro Bundler from `worker-farm` to `jest-worker`:
* Fully removes the custom patched `worker-farm` fork.
* Removes //a lot// of non-clear Flow types used to cast functions from callbacks to promises.
* Reduces cyclomatic complexity of the `Transformer` class by using `async`/`await`.
* Cleans all additional methods inside `JSTransformer/index.js`, by moving them to a single class and properly scoping them.
**Note:** this diff does not still enable the ability to bind files to workers by using `computeWorkerKey`; this will come at a later stage.
Reviewed By: davidaurelio
Differential Revision: D6247532
fbshipit-source-id: 51259360a5c15117996777a3be74b73b583f595e
Summary: To determine whether segment boundaries are properly covered by async imports rather than requires, we need to get knowledge about it higher up in the stack. This changeset exposes which of the dependencies are async as an array of indices within the `dependencies` array (I'd prefer avoiding duplicating the strings because they could get inconsistent, and I don't want to have 2 separate arrays of names either because we'd have to modify a bunch of stuff across the stack to support that).
Reviewed By: davidaurelio
Differential Revision: D6220236
fbshipit-source-id: 1ee36bc7c59f7f27e089f7771a24c45c8bd57b5d
Summary: Having types in tests ensure that we're testing what the API is supposed to process and not something else. In that case, adding type revealed that we were mistakenly passing strings instead of `Buffer` objects to the transform and optimize functions, but it would still work because `toString` was called over it. Passing proper `Buffer` objects as expected ensures that it doesn't just work with strings, and that the integration of these files in the rest of the application works as expected. This changeset fixes these callsites and add a few invariants here and there. We use `invariant` instead of `expect()` because Flow understands only the former as a type refinement, even though `expect` would also be a form of possible refinement. I don't think it's a big deal.
Reviewed By: davidaurelio
Differential Revision: D6160019
fbshipit-source-id: cbcbb05d7bccf9e1b9f6bb3ea30b0bb2925aef1b
Summary: Here's a good example why Flow "exact types" are useful: there are a few place where we were adding unusused fields in the objects. As these fields end up in JSON files used to communicate with Buck, this can amount to a waste of resources. This changeset make the type exact and removes the unused fields.
Reviewed By: mjesun
Differential Revision: D6159350
fbshipit-source-id: 8cdf29d5729253f119778943ad961eacb8990c04
Summary: This is the last step remaining before enabling the `import()` syntax to use `require.async()` (for now that function is just doing a simple `require()` in behind).
Reviewed By: davidaurelio
Differential Revision: D6147030
fbshipit-source-id: 5cd8ee6cc550816ae3cdea0b457dc2419c99e7a7
Summary:
After checking more deeply, there were still situations where the delta bundler could generate different bundles between runs: when a module is required by two different modules (it has two or more inverse dependencies), that module would not always be inserted after the first inverse dependency in the bundle, which would cause some bundling order discrepancies.
In order to fix that, the bundler now re-traverses synchronously the dependency graph using a DFS traversing algorithm to guarantee the same order. This will add some small runtime overhead (specially when generating the initial bundle), but it's not worrying (from benchmarks in my macbook pro it'll add ~10ms of initial build for every 1000 modules traversed, being this overhead linear).
Reviewed By: mjesun
Differential Revision: D6124993
fbshipit-source-id: 9bc7cb329f01a7860c7d3b52c3376c643ea5cf3b
Summary:
The `formatBanner.js` file is requiring the `wordwrap` package, but it was not defined in the package.json. Somehow this was working before (maybe the package was being downloaded by another depdendency or maybe `formatBanner` was not used).
This fixes https://github.com/facebook/metro-bundler/issues/79
Reviewed By: mjesun
Differential Revision: D6136865
fbshipit-source-id: 722dd61cb936fca893453f4cfc3248718a329779
Summary: This diff fixes https://github.com/facebook/metro-bundler/issues/78 by adding the `runBeforeMainModule` server parameter to the bundler. This is not the cleanest solution, but will be fixed once we use the Delta Bundler for builds from the CLI
Reviewed By: davidaurelio
Differential Revision: D6074469
fbshipit-source-id: 068926ef671d9f897ad9f1bd0540036a97340c00
Summary:
Adds module size as `fb_`-prefixed node attributes to the digraph output of dependencies.
The list of edges is now followed by a list of node declarations, that use the `fb_size` attribute to expose the module size in bytes.
E.g:
```
digraph {
"a" -> "b";
"b" -> "c";
"a"[fb_size=123];
"b"[fb_size=456];
"c"[fb_size=789];
}
```
Reviewed By: fkgozali
Differential Revision: D6064861
fbshipit-source-id: 73127e08c5e38075b5112053de12f9abac1102ee
Summary:
Resolves https://github.com/facebook/metro-bundler/issues/27
**Summary**
There is an edge case where tools like react-primitives need to assign to `Platform.OS`,
but the inline transformation results in an invalid AST.
When Platform.OS is unconditionally inlined, the following scenario:
````
Platform.OS = 'ios';
````
Is transformed to:
```
"ios"="ios"
````
And results in error `Property left of AssignmentExpression expected node to be of a type ["LVal"] but instead got "StringLiteral"**`.
See issue in react-primitives: https://github.com/lelandrichardson/react-primitives/issues/79
This patch checks whether the current node is on the left hand side of an AssignmentExpression
and skips the inlining when this is the case.
It also does the same for `process.env.NODE_ENV`, which suffers from the same problem. See related closed issue https://github.com/facebook/react-native/issues/7607#issuecomment-221425153 which was resolved by using bracket notation `process.env["NODE_ENV"]` instead.
**Test plan**
Unit tests included.
**Notes**
This is my first contribution to Metro, and haven't worked with Babel AST a lot, so constructive feedback on the implementation is welcome.
Closes https://github.com/facebook/metro-bundler/pull/45
Reviewed By: cpojer
Differential Revision: D5974615
Pulled By: mjesun
fbshipit-source-id: 224e63cef24f450b33e17ff9c0e0c0cea46bc70e
Summary:
In order to aid with debugging inline requires we add a method that
will give us the modules.
**Summary**
I was looking for a way to see what modules had been loaded when I was working with inline requires. I had tried using beginEvent on Systrace, but I was worried that there were modules that I might have missed (I'm not certain if all the loaded files are polyfills before the entryFile is invoked for instance). By adding getModules() it simplifies seeing what modules are loaded.
**Test plan**
At the root of my app `index.ios.js`:
```
import App from './App';
import { AppRegistry } from 'react-native';
AppRegistry.registerComponent('App', () => App);
// after 3 seconds the initial app data should have loaded, so lets take a look:
setTimeout(() => {
console.log(require.getModules());
}, 3000);
```
**NOTE**
If there is a way to get to this easily (via the debugger or some other mechanism) I would be happy to use that instead and add documentation for it.
Closes https://github.com/facebook/metro-bundler/pull/37
Differential Revision: D6008180
Pulled By: cpojer
fbshipit-source-id: 850103bbce166bf65a0cbd710582198c66273db5
Summary:
**Summary**
This is the workaround that cpojer suggested in #65 and should resolve the biggest pain point that developers are experiencing in that issue: using moment.js in React Native. Hopefully this at least serves as a starting point to reaching a decent solution for libraries using non-static `require()` statements.
**Test plan**
- Expected error when non-static-argument-based `require()` is nested more than one level deep within a `try` statement
- Expected no error when non-static-argument-based `require()` is nested directly within a `try` statement
```bash
ip-192-168-1-10:trueflipapp richard$ react-native bundle --entry-file index.js --platform ios --bundle-output /tmp/test.js --reset-cache
Scanning folders for symlinks in /repo/trueflip/sandbox/code/trueflipapp/node_modules (8ms)
Scanning folders for symlinks in /repo/trueflip/sandbox/code/trueflipapp/node_modules (8ms)
Loading dependency graph, done.
warning: the transform cache was reset.
bundle: start
bundle: finish
bundle: Writing bundle output to: /tmp/test.js
bundle: Done writing bundle output
```
Closes https://github.com/facebook/metro-bundler/pull/69
Differential Revision: D6008567
Pulled By: cpojer
fbshipit-source-id: f9be328cf50dc47c7433ffeb5eb053b398c92122
Summary:
Hi,
**Summary**
This fixes#40 where bundler cannot resolve a file starting with `.` from the root package directory.
it just adds a check to see if the absolute path filename is the same as the localpath filename then lets it through.
**Test plan**
- Steps to reproduce have been written up in #40
Thanks!
Closes https://github.com/facebook/metro-bundler/pull/54
Reviewed By: cpojer
Differential Revision: D5974618
Pulled By: mjesun
fbshipit-source-id: 4b7113c3bed20f2c908739881d61410d651e6ed7
Summary:
Whitelist node opt `--max-old-space-size` for JSTransformer.
Bundling can choke on the default node memory settings for those of us generating large index.js files. This change allows supplied node options to make it through to the node processes spawned by the JSTransformer worker farm.
Closes https://github.com/facebook/metro-bundler/pull/70
Differential Revision: D6008141
Pulled By: cpojer
fbshipit-source-id: 1ef3b436ea30baf3f255a4fd718fe4d958d70c65
Summary:
Addresses https://github.com/facebook/metro-bundler/issues/62
Apparently with some versions of `jest-haste-map`, the nested source file returns the module wrapped with `default` instead of the exports themselves (due to some ES6 babel transform, I am guessing). I wasn't able to reproduce this myself yet, but I think a good first step is to avoid using nested library files in the first place, ie. only use what's exported at the top-level.
Reviewed By: rafeca
Differential Revision: D5890757
fbshipit-source-id: 1196264b5626eda65c4aae222db2a1620f901c55
Summary: Upgrades uglify to version 3 with (experimental) ES6 support turned on.
Reviewed By: jeanlauliac
Differential Revision: D5842410
fbshipit-source-id: 1c8ccea15785bc5bb1c68d7a83b75881432d0ce2
Summary:
This replaces `async/queue` with a hand-rolled queue that does not yield to the event loop if a unit of work can be run synchronously.
Anecdotally, this leads to a > 11x speedup for the graph traversal when all data is available synchronously, e.g. from an in-memory cache.
Reviewed By: jeanlauliac
Differential Revision: D5861763
fbshipit-source-id: f7cf5f916a13adf9ca418d7522cd2f19df596fba
Summary: Changes `GraphFn` to return a promise rather than taking a callback. This is more in line with call sites, which so far have been using `denodeify`.
Reviewed By: jeanlauliac
Differential Revision: D5857192
fbshipit-source-id: 61bbbef4fbf24b0b91c4a3008541eaa5a1af0f7a
Summary:
Changes `LoadFn` to be synchronous or return a promise, and `ResolveFn` to be synchronsous.
This makes for a nicer API, without losing the property of not yielding to the event loop for synchronous work.
Reviewed By: jeanlauliac
Differential Revision: D5855963
fbshipit-source-id: 4b3c3363f4e49a9890586462925e8e400872feb2
Summary: When running with node 8, the babel plugins `async-to-generator` and `syntax-trailing-function-commas` are unnecessary. Removing them makes runtime transpilation faster, and the resulting code is closer to the original (useful with debugging: less unmapped code areas, better breakpoint support, less renamed variables).
Reviewed By: mjesun
Differential Revision: D5842305
fbshipit-source-id: d99f719794e4a8f48fd10b0349fbb36f2994666e
Summary:
This copies the basic loading mechanism and default config from the RN local cli into Metro and exposes it under `metro-bundler`, and switches internal scripts over to it.
davidaurelio: I changed the packager-worker-for-buck to hardcode the rn.cli file, like we do in other files. I would like to pull the "find" mechanism that traverses up to find a config into Metro at some point, but for now I'd prefer to keep it lean until we need it. Let me know if that doesn't work for you.
The next diff will switch the RN cli over to these functions also.
Reviewed By: davidaurelio
Differential Revision: D5832596
fbshipit-source-id: a3e167644d96c8831e5a83378e8ba143e62426db
Summary: As a first step in defining a new public API and CLI for Metro, I'd like to pull the generally useful pieces from the RN cli into Metro. This diff makes it as a non-breaking change (so we don't have to bump the version of Metro for RN) by updating only the imports outside of the react-native-github folder.
Reviewed By: davidaurelio
Differential Revision: D5832464
fbshipit-source-id: 11b00b5517665441763c2207d577ae0e110c681b
Summary: Fixed the UNBUNDLE magic file location to match what JniJSModulesUnbundle.cpp expects.
Reviewed By: sahrens
Differential Revision: D5821637
fbshipit-source-id: 4342e4bb4d139b4eba77dd92a53b1683041fc7e9
Summary:
This diff renames all the stragglers in comments and strings from variations of "packager" to "Metro Bundler". I did one of three:
* Rename "packager" to "Metro Bundler"
* Rename "react-native-packager" to "Metro Bundler"
* Remove "packager" when code inside of Metro implies that it's about Metro
I also removed `Glossary.md` because it is unmaintained and very old. mjesun is currently starting to write documentation for Metro which will supersede whatever was there before.
Reviewed By: mjesun
Differential Revision: D5802993
fbshipit-source-id: ba99cb5ed04d84b0f7b7a8a0bf28ed99280a940a