This is mostly to kick off conversation, i think we should go with a
modified version of the implemented approach that i'll describe here.
The playground currently serves two roles. The primary one we think
about is for verifying compiler output. We use it for this sometimes,
and developers frequently use it for this, including to send us repros
if they have a potential bug. The second mode is to help developers
learn about React. Part of that includes learning how to use React
correctly — where it's helpful to see feedback about problematic code —
and also to understand what kind of tools we provide compared to other
frameworks, to make an informed choice about what tools they want to
use.
Currently we primarily think about the first role, but I think we should
emphasize the second more. In this PR i'm doing the worst of both:
enabling all the validations used by both the compiler and the linter by
default. This means that code that would actually compile can fail with
validations, which isn't great.
What I think we should actually do is compile twice, one in
"compilation" mode and once in "linter" mode, and combine the results as
follows:
* If "compilation" mode succeeds, show the compiled output _and_ any
linter errors.
* If "compilation" mode fails, show only the compilation mode failures.
We should also distinguish which case it is when we show errors:
"Compilation succeeded", "Compilation succeeded with linter errors",
"Compilation failed".
This lets developers continue to verify compiler output, while also
turning the playground into a much more useful tool for learning React.
Thoughts?
---
[//]: # (BEGIN SAPLING FOOTER)
Stack created with [Sapling](https://sapling-scm.com). Best reviewed
with [ReviewStack](https://reviewstack.dev/facebook/react/pull/33777).
* #33981
* __->__ #33777
Squashed, review-friendly version of the stack from
https://github.com/facebook/react/pull/33488.
This is new version of our mutability and inference model, designed to
replace the core algorithm for determining the sets of instructions
involved in constructing a given value or set of values. The new model
replaces InferReferenceEffects, InferMutableRanges (and all of its
subcomponents), and parts of AnalyzeFunctions. The new model does not
use per-Place effect values, but in order to make this drop-in the end
_result_ of the inference adds these per-Place effects.
I'll write up a larger document on the model, first i'm doing some
housekeeping to rebase the PR.
---
[//]: # (BEGIN SAPLING FOOTER)
Stack created with [Sapling](https://sapling-scm.com). Best reviewed
with [ReviewStack](https://reviewstack.dev/facebook/react/pull/33494).
* #33571
* #33558
* #33547
* #33543
* #33533
* #33532
* #33530
* #33526
* #33522
* #33518
* #33514
* #33513
* #33512
* #33504
* #33500
* #33497
* #33496
* #33495
* __->__ #33494
* #33572
(Almost) all pragmas are now one of the following:
- `@...TestOnly`: custom pragma for test fixtures
- `@<configName>` | `@<configName>:true`: enables with either true or a
default enabled value
- `@<configName>:<json value>`
Adds HIR version of `PropagateScopeDeps` to handle optional chaining.
Internally, this improves memoization on ~4% of compiled files (internal links: [1](https://www.internalfb.com/intern/paste/P1610406497/))
Summarizing the changes in this PR.
1. `CollectOptionalChainDependencies` recursively traverses optional blocks down to the base. From the base, we build up a set of `baseIdentifier.propertyA?.propertyB` mappings.
The tricky bit here is that optional blocks sometimes reference other optional blocks that are *not* part of the same chain e.g. a(c?.d)?.d. See code + comments in `traverseOptionalBlock` for how we avoid concatenating unrelated blocks.
2. Adding optional chains into non-null object calculation.
(Note that marking `a?.b` as 'non-null' means that `a?.b.c` is safe to evaluate, *not* `(a?.b).c`. Happy to rename this / reword comments accordingly if there's a better term)
This pass is split into two stages. (1) collecting non-null objects by block and (2) propagating non-null objects across blocks. The only significant change here was to (2). We add an extra reduce step `X=Reduce(Union(X, Intersect(X_neighbors)))` to merge optional and non-optional nodes (e.g. nonNulls=`{a, a?.b}` reduces to `{a, a.b}`)
3. Adding optional chains into dependency calculation.
This was the trickiest. We need to take the "maximal" property chain as a dependency. Prior to this PR, we avoided taking subpaths e.g. `a.b` of `a.b.c` as dependencies by only visiting non-PropertyLoad/LoadLocal instructions. This effectively only recorded the property-path at site-of-use.
Unfortunately, this *quite* doesn't work for optional chains for a few reasons:
- We would need to skip relevant `StoreLocal`/`Branch terminal` instructions (but only those within optional blocks that have been successfully read).
- Given an optional chain, either (1) only a subpath or (2) the entire path can be represented as a PropertyLoad. We cannot directly add the last hoistable optional-block as a dependency as MethodCalls are an edge case e.g. given a?.b.c(), we should depend on `a?.b`, not `a?.b.c`
This means that we add its dependency at either the innermost unhoistable optional-block or when encountering it within its phi-join.
4. Handle optional chains in DeriveMinimalDependenciesHIR.
This was also a bit tricky to formulate. Ideally, we would avoid a 2^3 case join (cond | uncond cfg, optional | not optional load, access | dependency). This PR attempts to simplify by building two trees
1. First add each hoistable path into a tree containing `Optional | NonOptional` nodes.
2. Then add each dependency into another tree containing `Optional | NonOptional`, `Access | Dependency` nodes, truncating the dependency at the earliest non-hoistable node (i.e. non-matching pair when walking the hoistable tree)
ghstack-source-id: a2170f2628
Pull Request resolved: https://github.com/facebook/react/pull/31037
Resubmission of #30079 -- core logic unchanged, but needed to rebase past #30573
### Quick background
#### Temporaries
The compiler currently treats temporaries and named variables (e.g. `x`) differently in this pass.
- named variables may be reassigned (in fact, since we're running after LeaveSSA, a single named identifier's IdentifierId may map to multiple `Identifier` instances -- each with its own scope and mutable range)
- temporaries are replaced with their represented expressions during codegen. This is correct (mostly correct, see #29878) as we're careful to always lower the correct evaluation semantics. However, since we rewrite reactive scopes entirely (to if/else blocks), we need to track temporaries that a scope produces in `ReactiveScope.declarations` and later promote them to named variables.
In the same example, $4, $5, and $6 need to be promoted: $2 ->`t0`, $5 ->`t1`, and $6 ->`t2`.
```js
[1] $2 = LoadGlobal(global) foo
[2] $3 = LoadLocal bar$1
scope 0:
[3] $4 = Call $2(<unknown> $3)
scope 1:
[4] $5 = Object { }
scope 2:
[5] $6 = Object { a: $4, b: $5 }
[6] $8 = StoreLocal Const x$7 = $6
```
#### Dependencies
`ReactiveScope.dependencies` records the set of (read-only) values that a reactive scope is dependent on. This is currently limited to just variables (named variables from source and promoted temporaries) and property-loads.
All dependencies we record need to be hoistable -- i.e. reordered to just before the ReactiveScope begins. Not all PropertyLoads are hoistable.
In this example, we should not evaluate `obj.a.b` without before creating x and checking `objIsNull`.
```js
// reduce-reactive-deps/no-uncond.js
function useFoo({ obj, objIsNull }) {
const x = [];
if (isFalse(objIsNull)) {
x.push(obj.a.b);
}
return x;
}
```
While other memoization strategies with different constraints exist, the current compiler requires that `ReactiveScope.dependencies` be re-orderable to the beginning of the reactive scope. But.. `PropertyLoad`s from null values will throw `TypeError`. This means that evaluating hoisted dependencies should throw if and only if the source program throws. (It is also a bug if source throws and compiler output does not throw. See https://github.com/facebook/react-forget/pull/2709)
---
### Rough high level overview
1. Pass 1
Walk over instructions to gather every temporary used outside of its defining scope (same as ReactiveFunction version). These determine the sidemaps we produce, as temporaries used outside of their declaring scopes get promoted to named variables later (and are not considered hoistable rvals).
2. Pass 2 (collectTemporariesSidemap)
Walk over instructions to generate a sidemap of temporary identifier -> named variable and property path (e.g. `$3 -> {obj: props, path: ["a", "b"]}`)
2. Pass 2 (collectHoistablePropertyLoads)
a. Build a sidemap of block -> accessed variables and properties (e.g. `bb0 -> [ {obj: props, path: ["a", "b"]} ]`)
b. Propagate "non-nullness" i.e. variables and properties for which we can safely evaluate `PropertyLoad`.
A basic block can unconditionally read from identifier X if any of the following applies:
- the block itself reads from identifier X
- all predecessors of the block read from identifier X
- all successors of the block read from identifier X
4. Pass 3: (collectDependencies)
Walks over instructions again to record dependencies and declarations, using the previously produced sidemaps. We do not record any control-flow here
5. Merge every scope's recorded dependencies with the set of hoistable PropertyLoads
Tested by syncing internally and (1) checking compilation output differences ([internal link](https://www.internalfb.com/intern/everpaste/?handle=GPCfUBt_HCoy_S4EAJDVFJyJJMR0bsIXAAAB)), running internally e2e tests ([internal link](https://fburl.com/sandcastle/cs5mlkxq))
---
### Followups:
1. Rewrite function expression deps
This change produces much more optimal output as the compiler now uses the function CFG to understand which variables / paths are assumed to be non-null. However, it may exacerbate [this function-expr hoisting bug](https://github.com/facebook/react/blob/main/compiler/packages/babel-plugin-react-compiler/src/__tests__/fixtures/compiler/bug-invalid-hoisting-functionexpr.tsx). A short term fix here is to simply call some form of `collectNonNullObjects` on every function expression to find hoistable variable / paths. In the longer term, we should refactor out `FunctionExpression.deps`.
2. Enable optional paths
(a) don't count optional load temporaries as dependencies (e.g. `collectOptionalLoadRValues(...)`).
(b) record optional paths in both collectHoistablePropertyLoads and dependency collection
ghstack-source-id: 2507f6ea75
Pull Request resolved: https://github.com/facebook/react/pull/30894
This PR updates to use SSA form through the entire compilation pipeline. This means that in both HIR form and ReactiveFunction form, `Identifier` instances map 1:1 to `IdentifierId` values. If two identifiers have the same IdentifierId, they are the same instance. What this means is that all our passes can use this more precise information to determine if two particular identifiers are not just the same variable, but the same SSA "version" of that variable.
However, some parts of our analysis really care about program variables as opposed to SSA versions, and were relying on LeaveSSA to reset identifiers such that all Identifier instances for a particular program variable would have the same IdentifierId (though not necessarily the same Identifier instance). With LeaveSSA removed, those analysis passes can now use DeclarationId instead to uniquely identify a program variable.
Note that this PR surfaces some opportunties to improve edge-cases around reassigned values being declared/reassigned/depended-upon across multiple scopes. Several passes could/should use IdentifierId to more precisely identify exactly which values are accessed - for example, a scope that reassigns `x` but doesn't use `x` prior to reassignment doesn't have to take a dependency on `x`. But today we take a dependnecy.
My approach for these cases was to add a "TODO LeaveSSA" comment with notes and the name of the fixture demonstrating the difference, but to intentionally preserve the existing behavior (generally, switching to use DeclarationId when IdentifierId would have been more precise).
Beyond updating passes to use DeclarationId instead of Identifier/IdentifierId, the other change here is to extract out the remaining necessary bits of LeaveSSA into a new pass that rewrites InstructionKind (const/let/reassign/etc) based on whether a value is actually const or has reassignments and should be let.
ghstack-source-id: 69afdaee5f
Pull Request resolved: https://github.com/facebook/react/pull/30573
Updates the prettier config to format all `.ts` and `.tsx` files in the
repo using the existing defaults and removing overrides.
The first commit in this PR contains the config changes, the second is
just the result of running `yarn prettier-all`.
Our previous logic for aligning scopes to block scopes constructs a tree of block and scope nodes. We ensured that blocks always mapped to the same node as their fallthroughs. e.g.
```js
// source
a();
if (...) {
b();
}
c();
// HIR
bb0:
a()
if test=... consequent=bb1 fallthrough=bb2
bb1:
b()
goto bb2
bb2:
c()
// AlignReactiveScopesToBlockScopesHIR nodes
Root node (maps to both bb0 and bb2)
|- bb1
|- ...
```
There are two issues with the existing implementation:
1. Only scopes that overlap with the beginning of a block are aligned correctly. This is because the traversal does not store information about the block-fallthrough pair for scopes that begin *within* the block-fallthrough range.
```
\# This case gets handled correctly
┌──────────────┐
│ │
block start block end
scope start scope end
│ │
└───────────────┘
\# But not this one!
┌──────────────┐
│ │
block start block end
scope start scope end
│ │
└───────────────┘
```
2. Only scopes that are directly used by a block is considered. See the `align-scopes-nested-block-structure` fixture for details.
ghstack-source-id: 327dec5019
Pull Request resolved: https://github.com/facebook/react/pull/29891
Joseph SavonaandGitHub
Jordan BrownandGitHub
mofeiZandGitHub
Mofei Zhang
Joe Savona
Jan KassensandGitHub
Niklas MollenhauerandGitHub