This tracks whether a value is a context ref or generated from a context ref.
This lets us track mutations to context refs and treat it separately as we want
this to be more conservative than our existing inference.
ValueKind.Context is exactly like ValueKind.Mutable but is more conservative.
I realized we hadn't updated InferReferenceEffects to match our latest thinking
on hooks. Specifically, we will default to assuming that hooks can mutate their
arguments and return mutable values — this works with our model since we don't
treat hooks specially for reactive scope construction. Ie, first we figure out
what variables construct together, then we create scopes, then we prune scopes
that contain hooks. So changing the reference effects for hooks "just works".
Note that it is helpful for our unit tests to have an example hook that we know
_does_ freeze its input and return a frozen value, so i've temporarily added
`useFreeze()` to the list of defined hooks. That is meant as a stopgap: the
right solution is to allow some way to tell the compiler about specific custom
hooks and their semantics.
Effect.Capture is very similar to Effect.Read, but the only difference is that
this reference is stored somewhere via a Effect.Store.
Previously, any operand associated with a Effect.Store in the same instruction
would get aliased -- so there was no need to explicitly differentiate between a
"normal read" and "read that gets stored".
This difference is now explicit with FunctionExpression where every dependency
is "read" but only a few are "captured" for store (and mutation). In a follow up
PR, the mutating deps will have an Effect.Capture to differentiate from the
other non mutating deps (Effect.Read).
No need for InferAliasForStores to know about the semantics of each instruction
anymore. It's just a simple pass that iterates over every operand and lvalue.
The FunctionExpression is special cased because it's slightly different but I
have a follow up that removes this special casing.
This doesn't change codegen as the lvalue is unused but it lets us make this
pass be semantically the same across all instructions -- "alias lvalue and
operands of an instr".
Support TypeCastExpressions — `(x: TypeAnnotation)`. This is pretty
straightforward, it's semantically identical to a raw identifier.
One catch is that our prettier config is hard-coded to use the babel-ts parser,
i wasn't sure how to make that dynamic based on the file extension so for now i
just ignored .flow.js files in our pretter config.
As discussed, this repurposes OtherStatement as a catch all variant for
unsupported syntax or errors in the source. This also renames the previously
added ErrorTerminal to UnsupportedTerminal for consistency (plus makes it a
little bit less confusing that it's not an actual terminal representing an
Error).
Not loving the name but couldn't think of anything better, open to suggestions!
Adds `PropertyCall` and `ComputedCall`, which are a combination of
CallExpression and PropertyLoad/ComputedLoad, respectively. The goal is to
ensure that we correctly model the receiver of a call where the callee is a
member expression, and also accurately record scope dependencies in for both the
computed and non-computed (property) cases.
An alternative that I tried first was to add a `receiver: Place | null` to
CallExpression. That works well for HIR construction, but it's then very
difficult at codegen time to correctly reconstruct the original call: if the
receiver and callee share part of their structure then we can transform back to
a non-computed member expression, otherwise it has to be computed. Eg we have to
distinguish `a.b.c[d.foo]()` from `a.b.c[a.b.c.foo]()`. Given that our target is
high-level code, it seems reasonable to have a higher-level representation for
these cases.
I'm open to feedback but this feels pretty reasonable in terms of complexity /
precision of modeling.