Summary:
Pull Request resolved: https://github.com/facebook/react-native/pull/36384
## Changelog:
[Internal] - Adds more test coverage to PerformanceEntryObserver functionality related to buffering of the marks/measures, as well as making fixes for certain corner cases handled incorrectly (which are now covered by tests).
Reviewed By: rubennorte
Differential Revision: D43835928
fbshipit-source-id: fc8adbd6c2a2b6b201f1f14bdf51245268c5eabd
Summary:
## Changelog:
[Internal] - Add unit tests for the WebPerformance library (PerformanceEntryReporter)
This sets up a C++ unit test suite for `PerformanceEntryReporter` (the core part of the native side of WebPerformance) and adds test coverage for its core functionality.
Reviewed By: sammy-SC
Differential Revision: D43771370
fbshipit-source-id: ad3e0f3f206701c2ea6a5c9386458a76699e7c80
Summary:
This diff adds the `performance.reactNativeStartupTiming` API to the performance global object for RN. This property does not exist in web, and we are free to make up our own list of properties in the startup metrics to track RN app startup process. In our case, we have the following six properties to begin with (we may extend and add more to this list in the future):
```
- `(start|end)Time`: The time ‘zero’ for the startup timing and the end of app startup. RN has no knowledge of app start time, which will be provided by the platform. The `endTime` will be the time when the first JS bundle finishes executing (Note that RN supports multiple JS bundles, which can be loaded async)
- `executeJavaScriptBundleEntryPoint(Start|End)`: The time for RN to execute the JS entry point (and finish all sync job)
```
Changelog:
[General][Added] - Add new JS performance API to support getting RN app startup timings
Reviewed By: rshest
Differential Revision: D43326564
fbshipit-source-id: 7b4c7cae70ff64ba1714a1630cd5e183df6c06b0
Summary:
Pull Request resolved: https://github.com/facebook/react-native/pull/36314
## Changelog:
[Internal] - Added support for W3C Performance API extension (Performance,getEntries*)
See [here for the details](https://www.w3.org/TR/performance-timeline/#extensions-to-the-performance-interface).
This only supports `mark` and `measure` performance entry types (not `events`, as those are not supported by browsers either, they recommend using the `PerformanceObserver` API instead).
From the implementation perspective, we already maintained persistent buffer for `mark` entry types, which was required in order to look up measures.
The buffer is circular, limited to 1K entries by default.
I basically mimic the same behavior for `measure` entry types as well, since it's the simplest way of doing it at this point,
If we happen to want adding some other entry types that would be available via `Performance.getEntries*` API in the future, we may want to iterate on the internal data structures representation inside `PerformanceEntryReporter`, but for now I believe this approach should work just fine, and whatever changes we may want to do will be purely internal implementation detail.
Reviewed By: rubennorte
Differential Revision: D43625488
fbshipit-source-id: dd315b3f8488e910749a8e2a4158246e94d76f99
Summary:
## Changelog:
[Internal] -
Noticed this when setting up C++ unit tests for WebPerformance - the change in this diff both makes sure there is no "global constructor" warning generated when `-Wglobal-constructors` is enabled, and also makes this bit potentially a little bit more efficient, as we don't pay for the registry construction if we don't use the WebPerformance API.
Reviewed By: christophpurrer
Differential Revision: D43663521
fbshipit-source-id: 59952f2415f49bb455a3443b3bfd8971108ac72b
Summary:
Pull Request resolved: https://github.com/facebook/react-native/pull/36312
## Changelog:
[Internal] -
`clearMarks` and `clearMeasures` methods are incidental to the `NativePerformance` TurboModule functionality, as in reality this responsibility belongs more on the `NativePerformanceObserver` and `PerformanceEntryReporter` side.
This is something that [the standard indirectly suggests](https://www.w3.org/TR/user-timing/#clearmarks-method) as well (referencing [performance entry buffer](https://www.w3.org/TR/performance-timeline/#dfn-performance-entry-buffer)).
The new implementation should be also a little bit more efficient, as it avoids calling the predicate for each entry.
Finally (and frankly, the main reason for this change, from my perspective), it will simplify mocking/testing the JS part of the PerfAPI code.
Reviewed By: rubennorte
Differential Revision: D43621174
fbshipit-source-id: c4217a0da1d8ecbce797240627f7b4f057d85b97
Summary:
Pull Request resolved: https://github.com/facebook/react-native/pull/36197
[Changelog][Internal]
This turns NativePerformance* module dependencies into "TurboModule plugins", which allows for more streamlined client integration (as it makes them register automatically once the dependency is there).
Reviewed By: rubennorte
Differential Revision: D43353204
fbshipit-source-id: 01d0089750a4873088dc4aefe34fd48693ee9791
Summary:
Pull Request resolved: https://github.com/facebook/react-native/pull/36152
[Changelog][Internal]
By [the W3C standard](https://developer.mozilla.org/en-US/docs/Web/API/PerformanceObserver/observe), `PerformanceObserver.observer` can optionally take a `durationThreshold` option, so that only entries with duration larger than the threshold are reported.
This diff adds support for this on the RN side, as well as unit tests for this feature on the JS side.
NOTE: The standard suggests that default value for this is 104s. I left it at 0 for now, as for the RN use cases t may be to too high (needs discussion).
Reviewed By: rubennorte
Differential Revision: D43154319
fbshipit-source-id: 0f9d435506f48d8e8521e408211347e8391d22fc
Summary:
Pull Request resolved: https://github.com/facebook/react-native/pull/36181
[Changelog][Internal]
Implements EventCounts API (`Performance.eventCounts`) for Web Performance, according to the W3C standard, see the specs here: https://www.w3.org/TR/event-timing/#eventcounts
The rationale for why we need it is to support some advanced metrics computations, such as a ratio of "slow events" to total event count, per event type.
Reviewed By: rubennorte
Differential Revision: D43285073
fbshipit-source-id: 2c53d04d9a57c1301e37f2a5879072c8d33efbbf
Summary:
This diff adds new performance API `memory`, which is a read-only property that gets the current JS heap size from native side.
Note that the JSI API returns an unordered map with unknown list of memory information, which is different from the [web spec](https://fburl.com/p0vpbt33). We may enforce specific memory info type on the JSI API so that it can be properly translate to the web spec.
- Update the JS spec
- Update Native implementation and return memory information with JSI API `jsi::instrumentation()::getHeapInfo()`
- Add native performance module to catalyst package
Changelog:
[General][Added] - Add performance memory API with native memory Info
Reviewed By: rubennorte
Differential Revision: D43137071
fbshipit-source-id: 319f1a6ba78fce61e665b00849ecf2579094af83
Summary:
Pull Request resolved: https://github.com/facebook/react-native/pull/36116
[Changelog][Internal]
Add a minimal/reference JavaScript implementation for NativePerformanceObserver - the purpose is both unit testing (JS and native sides separately) and potentially shimming the part of functionality that is not dependent on native side.
This is both a setup for adding general unit tests for the Performance* APIs, but also to be able to do non-trivial changes on JS side for WebPerformance (such as in (D43154319).
Reviewed By: rubennorte
Differential Revision: D43167392
fbshipit-source-id: 213d9534d810dece1dd464f910e92e08dbf39508
Summary:
[Changelog][Internal]
`NativePerformanceObserver` TurboModule API would get the type for performance entries as strings in one direction (`start/stopReporting`) and as integers in another direction (inside `RawPerformanceEntry`, for optimization on the native side).
This makes is symmetrical and consistent, all the conversions are now handled on the JS side.
Reviewed By: christophpurrer
Differential Revision: D43236466
fbshipit-source-id: 08e1b62df90e6d26a11577d6b6b1d91a6bce8339
Summary:
I encountered build error when using performance API in catalyst android mobile app. The error message P617433618 points at using non-const `std::strlen` API in a `constexpr`.
```
$ buck install catalyst-android
...
stderr: xplat/js/react-native-github/Libraries/WebPerformance/PerformanceEntryReporter.cpp:208:13: error: constexpr constructor never produces a constant expression [-Winvalid-constexpr]
constexpr StrKey(const char *s)
^
xplat/js/react-native-github/Libraries/WebPerformance/PerformanceEntryReporter.cpp:209:39: note: non-constexpr function 'strlen' cannot be used in a constant expression
: key(folly::hash::fnv32_buf(s, std::strlen(s))) {}
```
Changelog:
[General][Fixed] - Fixed string key calculation in constexpr from Performance C++ native module.
Reviewed By: javache
Differential Revision: D43136624
fbshipit-source-id: c691671b157b507745c67a505c91f75cf6b878d1
Summary:
[Changelog][Internal]
Use the [list of supported events](https://www.w3.org/TR/event-timing/#sec-events-exposed) explicitly to both filter and transform reported event types, which makes this both more robust and less ambiguous.
The mapping is using a compile-time hash/lookup, similarly to how we do it with the ViewProps in the RN core.
Reviewed By: rubennorte
Differential Revision: D42342655
fbshipit-source-id: 3d0b2465fd5611db110c4792913e0a92e76cd067
Summary:
Pull Request resolved: https://github.com/facebook/react-native/pull/35771
[Changelog][Internal]
Based on the internal discussion, we don't want to report `first-input` event types for RN (just use plain `event` instead), in the way that [Event Timing API standard suggests](https://www.w3.org/TR/event-timing), as this is doesn't have that clear semantics in the context of RN, also to keep it simpler.
Reviewed By: rubennorte
Differential Revision: D42341923
fbshipit-source-id: eff2487dee17ef082604e4c807b4d41485328114
Summary:
Pull Request resolved: https://github.com/facebook/react-native/pull/35768
Changelog: [Internal]
This implements native side mechanics for reporting user events timing to JS (PerformanceObserver API).
See the standard for more details: https://www.w3.org/TR/event-timing/
The events are only logged when there are any active subscriptions (via `PerformanceObserver.observe`), also we only log "discrete events" (i.e. no likes of mouse move), so the overhead is non-existing.
There are two main metrics of interest for an event lifecycle:
* Time the event is spent in the queue, i.e. the time between it's created and dispatched
* Time that is spend in the event handler on the JS side (event dispatch), or processing time
Both of these are measured, and the corresponding fields are populated.
Reviewed By: sammy-SC
Differential Revision: D42294947
fbshipit-source-id: 4fd7938c04b942400befa4057d4929fb2763cee1
Summary:
Changelog: [Internal]
This adds definition of `PerformanceEventTiming` interface, according to the W3C standard, so that [event timing](https://www.w3.org/TR/event-timing) data can be reported from native (the C++ part is in the next diff).
See here: https://www.w3.org/TR/event-timing/#performanceeventtiming
Reviewed By: christophpurrer
Differential Revision: D42279486
fbshipit-source-id: 0dfbcd6e5a08fc1b89651bd35b24fb4e731f8b05
Summary:
Extends the WebPerformance API with ability to report events, [according to the standard](https://www.w3.org/TR/event-timing/#sec-performance-event-timing).
This is an API-only change, the actual reporting comes in a separate diff, to simplify reviewing.
Changelog: [Internal]
Reviewed By: christophpurrer
Differential Revision: D42097695
fbshipit-source-id: d8b468ffed50c1c3d889151df5e8ca644d6e1a68
Summary:
Prevents scenarios when internal performance buffer may grow indefinitely (e.g. due to a broken logging), communicating back to `PerformanceObserver` the corresponding amount of dropped entries, `droppedEntriesCount`, [according to the standard](https://w3c.github.io/performance-timeline/#dom-performanceobservercallbackoptions-droppedentriescount).
NOTE: The backwards compatibility check is failing, which is an orthogonal issue. I am looking into it and won't land this one before it is sorted.
Changelog: [Internal]
Reviewed By: christophpurrer
Differential Revision: D42008409
fbshipit-source-id: 40d30e44d39e643bfb58a6254572823cb2b3b8df
Summary:
Now that D42097191 allows it, get rid of unused API in the related WebPerformance native modules (these were kept there only to pacify the backwards compatibility checks).
Changelog: [Internal]
Reviewed By: christophpurrer
Differential Revision: D42037757
fbshipit-source-id: 5f5e7e76722cade9e730aba037f9e8ab51fc16d9
Summary:
Makes sure that the global (standard) `performance` instance is initialized to still both provide `Performance.now()` as before, but also allows for the newly implemented functionality from the WebPerf API, such as marks/measures.
Changelog: [Internal]
Reviewed By: christophpurrer
Differential Revision: D42017812
fbshipit-source-id: ddbe79e91b45a84871de94018305f2a4536ada4b
Summary:
Makes sure that we don't spam too often the JS performance entry reporting callback, which further dispatches entries to `PerformanceObserver` instances.
The logic is as following:
* ~~~If internal buffer of entries reaches the limit, we schedule the callback (with background priority)~~~
*~~~ Once the callback is processed, we schedule the next flush after a timeout of 500ms, this will also be scheduled from native with background priority~~~
* ~~~Whenever new performance type starts to be observed, we also schedule the callback, in order to prime the above~~~
* Schedule the flush with low priority, whenever there is the first entry coming into an empty buffer, and rely on the Scheduler to "do the right thing" when asked to flush it with background priority and not doo it exceedingly often (see the prolonged discussion)
Changelog: [internal]
Reviewed By: rubennorte
Differential Revision: D41875085
fbshipit-source-id: 368b525203215350ceabb43d5e9e8e3bd5242aca
Summary:
Our implementation of `PerformanceObserver.prototype.observe` diverges a little bit from the [spec](https://w3c.github.io/performance-timeline/#dom-performanceobserver-observe) regarding the behavior when calling it multiple times. E.g.:
1. We don't throw errors if both `type ` and `entryTypes` are passed.
2. We don't throw errors if neither `type ` nor `entryTypes` are passed.
3. We don't throw errors if we call observe with `type` and we call again later with `entryTypes`, and vice versa.
4. The logic to consolidate entry types in subsequent observe calls is incorrect.
This fixes those issues and also applies some minor naming improvements.
Changelog: [internal]
Reviewed By: rshest
Differential Revision: D41872269
fbshipit-source-id: 9a6b485f4ef2f479c8b6f0307012322f65a11879
Summary:
We're currently showing warnings when we call `performanceObserver.observe` and the native module for performance observers isn't available, but we don't do the same for `performance.mark`, `performance.measure`, etc.
This adds the warning in those cases.
Changelog: [internal]
Reviewed By: rshest
Differential Revision: D41872270
fbshipit-source-id: d720580b930550f27c827a58243579c42a4f6da9
Summary:
Changelog: [Internal]
This implements the C++ side logic of handling `Performance.measure` calls.
Since measures may refer to earlier logged marks by name, we need to keep track of the former. I also use a fixed size (circular) buffer to prevent the marks from piling forever.
Also adds implementation of clearing marks/measures, as per standard.
Reviewed By: mdvacca
Differential Revision: D41756928
fbshipit-source-id: 19dce28d6af4c5646274e6d5db20b41869282780
Summary:
Changelog: [Internal]
This adds JS side implementation (including the API) for the `Performance.measure` functionality, [as described in the corresponding standard](https://www.w3.org/TR/user-timing/#measure-method).
The JS part is separated from the C++ implementation (further down the stack) to help the review being more focused.
Reviewed By: mdvacca
Differential Revision: D41733190
fbshipit-source-id: 72b69f6bb332aed4b9477a186b0e818b62009220
Summary:
[Changelog][Internal]
The NativePerformance module functionality corresponds to the [timing extensions of the Performance API standard interface](https://www.w3.org/TR/user-timing/#extensions-performance-interface).
As this is logically separate from `PerformanceObserver` (which may exist without it), it makes sense to have it as a different native module, so there is no coupling between both.
Reviewed By: christophpurrer
Differential Revision: D41690145
fbshipit-source-id: 7443f4c51f54cc2fdddbdb2e89f9a1fa457ab280
Summary:
[Changelog][Internal]
Addresses the corresponding TODO in code, to make sure that PerformanceObserver.observe may be called multiple times with different entry types to observe.
Reviewed By: christophpurrer
Differential Revision: D41686237
fbshipit-source-id: 4287f63f7e71e5b1056d30ed616149841ba4f892
Summary:
[Changelog][Internal]
The name corresponds more precisely to what the method does.
Reviewed By: christophpurrer
Differential Revision: D41686205
fbshipit-source-id: 36c47b57fdeb757515cd14b890f38247f7fe8d02
Summary:
Pull Request resolved: https://github.com/facebook/react-native/pull/35526
[Changelog][Internal]
This closes the full loop according to the [technical design](https://fb.quip.com/MdqgAk1Eb2dV) of the WebPerf API implementation, with the main components and the working central data flow in place.
The next step is to add some buffering/throttling, as in this diff we just spawn an idle-priority task after every performance entry coming (even though they still naturally do come in batches, because they manage to accumulate before the task is executed).
Reviewed By: christophpurrer
Differential Revision: D41496082
fbshipit-source-id: 5fd4cf22e75806f7bc98d1d1b6691596ccadf8b9
Summary:
[Changelog][Internal]
Adds API definition for [Performance.mark](https://www.w3.org/TR/user-timing/#mark-method) support.
This is a bare bone implementation, that just logs events on the native side. The next step is the native logic for queuing, flushing etc.
Note that here I route both JS and native marks to native for now, for simplicity sake - ultimately this may not be what we want, as it may be more efficient to process marks, logged from JS, on the JS side.
Reviewed By: rubennorte
Differential Revision: D41472148
fbshipit-source-id: bdf2b182b8472a71a5500235849bca5af1c2f360
Summary:
Pull Request resolved: https://github.com/facebook/react-native/pull/35226
Changelog: [Internal]
This adds scaffolding for the C++ side of NativePerformanceObserver module.
Thanks to christophpurrer for helping set this up, as this is the first one of this kind inside core/OSS.
Reviewed By: rubennorte
Differential Revision: D41028555
fbshipit-source-id: 4acf0e71a254a42044cbbe5f94f40938342c6aa2
Summary:
[Changelog][Internal]
This adds module specs for the native part of PerformanceObserver, as well as the interaction logic vs the NativePerformanceObserver API.
See https://fb.quip.com/MdqgAk1Eb2dV for more detail.
Reviewed By: rubennorte
Differential Revision: D40897006
fbshipit-source-id: 77475f21dad9ee9dbe15df5a989eb08d314e6db2
Summary:
[Changelog][Internal]
This is to prevent an accidental use of PerformanceObserver while it's still a work in progress.
PerformanceObserver is also moved into a dedicated folder, `Libraries/WebPerformance`.
Reviewed By: rubennorte
Differential Revision: D40978749
fbshipit-source-id: 09645a95bed72902870ebc00c1b4a3d81ab4c829
Ruslan Shestopalyuk
Xin Chen
Rubén Norte
Christoph Purrer