Skip to content

Conversation

robhogan
Copy link
Owner

Summary

Changelog

[CATEGORY] [TYPE] - Message

Test Plan

@github-actions
Copy link

Fails
🚫

❗ Base Branch - The base branch for this PR is something other than main or a -stable branch. Are you sure you want to target something other than the main branch?

🚫

📋 Verify Changelog Format - A changelog entry has the following format: [CATEGORY] [TYPE] - Message.

CATEGORY may be:
  • General
  • iOS
  • Android
  • JavaScript
  • Internal (for changes that do not need to be called out in the release notes)

TYPE may be:

  • Added, for new features.
  • Changed, for changes in existing functionality.
  • Deprecated, for soon-to-be removed features.
  • Removed, for now removed features.
  • Fixed, for any bug fixes.
  • Security, in case of vulnerabilities.

MESSAGE may answer "what and why" on a feature level. Use this to briefly tell React Native users about notable changes.

Generated by 🚫 dangerJS against c548dd0

robhogan pushed a commit that referenced this pull request Jan 26, 2024
Summary:
Pull Request resolved: facebook#41466

## Context
In open source, all apps use the same turbomodulemanager delegate (i.e: the default delegate).

This diff introduces the buck infra that makes the oss default delegate work for meta apps.

Concretely, we are going to make React Native use the same  delegate for **all** Meta apps.

Each Meta app will:
1. At build time, generate a unique TMProvider map
2. At app init time, initialize the default delegate with the TMProvider map.

## Implementation
**Step #1:** At build time, generate a unique TMProvider map

**Insight:** Buck genrules can accept, as input, the output of a buck query.

So, here's how we get this done:
1. Buck query (i.e: input to Genrule): Given the app's deps, query all the schemas in the app.
2. Genrule: Read the schemas to generate the TMProvider map. The TMProvider map will also contain **all** the app's C++ module codegen.

Concretely:
1. This diff introduces a macro: rn_codegen_appmodules(deps).
2. rn_codegen_appmodules(deps) generates appmodules.so, which contains the TMProvider map.

**Step #2:** At app init time, initialize the default delegate with the TMProvider map.

This is how we'll initialize the DefaultTurboModuleManagerDelegate:
1. DefaultTurboModuleManagerDelegate will load appmodules.so during init.
2. When loaded, appmodules.so will assign the code-generated TMProvider map to DefaultTurboModuleManagerDelegate.

## Impact
This should allow us to:
1. Get one step closer to getting rid of the `js1 build turbomodule-manager-delegates --target <app>` script
3. Remove the TurboModuleManagerDelegate from React Native's public API. (Because we use one delegate for all React Native apps in Meta and OSS)

Changelog: [Internal]

Reviewed By: mdvacca

Differential Revision: D50988397

fbshipit-source-id: 0ca5dec14e2dae89ec97f5d39a182c7937c5c7bf
robhogan pushed a commit that referenced this pull request Feb 19, 2024
…gets

Summary:
Changelog: [Internal]

# This diff

1. Provides all Targets with an `executorFromThis()` method, which can be used from within a Target to access a *`this`-scoped main thread executor* = a `std::function` that will execute a callback asynchronously iff the current Target isn't destroyed first.
2. Refactors how (all) Target objects are constructed and retained, from a plain constructor to `static shared_ptr create()`. This is because `executorFromThis()` relies internally on `enable_shared_from_this` plus two-phase construction to populate the executor.
3. Creates utilities for deriving scoped executors from other executors and `shared_ptr`s.

The concept is very much like `RuntimeExecutor` in reverse: the #1 use case is moving from the JS thread back to the main thread - where "main thread" is defined loosely as "anywhere it's legal to call methods on Target/Agent objects, access session state, etc". The actual dispatching mechanism is left up to the owner of each `PageTarget` object; for now we only have an iOS integration, where we use `RCTExecuteOnMainQueue`.

Coupling the ownership/lifetime semantics with task scheduling is helpful, because it avoids the footgun of accidentally/nondeterministically moving `shared_ptr`s (and destructors!) to a different thread/queue .

# This stack
I'm refactoring the way the Runtime concept works in the modern CDP backend to bring it in line with the Page/Instance concepts.

Overall, this will let us:

* Integrate with engines that require us to instantiate a shared Target-like object (e.g. Hermes AsyncDebuggingAPI) in addition to an per-session Agent-like object.
* Access JSI in a CDP context (both at target setup/teardown time and during a CDP session) to implement our own engine-agnostic functionality (`console` interception, `Runtime.addBinding`, etc).
* Manage CDP execution contexts natively in RN, and (down the line) enable first-class debugging support for multiple Runtimes in an Instance.

The core diffs in this stack:

* ~~Introduce a `RuntimeTarget` class similar to `{Page,Instance}Target`. ~~
* ~~Make runtime registration explicit (`InstanceTarget::registerRuntime` similar to `PageTarget::registerInstance`). ~~
* ~~Rename the existing `RuntimeAgent` interface to `RuntimeAgentDelegate`.~~
* ~~Create a new concrete `RuntimeAgent` class similar to `{Page,Instance}Agent`.~~
* ~~Provide `RuntimeTarget` and `RuntimeAgent` with primitives for safe JSI access, namely a `RuntimeExecutor` for scheduling work on the JS thread.~~
  * Provide RuntimeTarget with mechanism for scheduling work on the "main" thread from the JS thread, for when we need to do more than just send a CDP message (which we can already do with the thread-safe `FrontendChannel`) in response to a JS event. *← This diff*

## Architecture diagrams

Before this stack:
https://pxl.cl/4h7m0

After this stack:
https://pxl.cl/4h7m7

Reviewed By: hoxyq

Differential Revision: D53356953

fbshipit-source-id: 152c784eb64e9b217fc2966743b33f61bd8fd97e
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant