Memoize getName/getNameParts on namespaced AST nodes

[chrisdp]

Summary

Memoize the result of getName / getNameParts on NamespacedVariableNameExpression, NamespaceStatement, FunctionStatement, ClassStatement, and InterfaceStatement. Validators call these methods per potential namespace reference per scope, and each call walks the AST (findAncestor) and rebuilds the qualified name string from scratch. The result is a pure function of the parsed AST, so caching it on the node is safe and eliminates a large amount of repeated work.

Measured impact

Profiled on a brighterscript build of ~1300 source files (using node --cpu-prof --heap-prof):

Metric	Before	After	Δ
Total CPU sampled time	13.58 s	12.03 s	-1.55 s (-11.4%)
Validate phase wall-clock	8.74 s	7.60 s	-1.14 s (-13%)
GC self-time	1.37 s	1.08 s	-290 ms (-21%)
Total heap allocations	876 MB	869 MB	flat (within variance)

CPU profile attribution before this change: NamespacedVariableNameExpression.getName 455 ms self, getNameParts 247 ms self, Statement.getName and the (get name) getter chain together another ~800 ms self. After this change those frames drop out of the top-25 entirely.

Heap stays essentially flat: each relevant AST node grows by two string | undefined slots (~16 bytes), but that's offset by the eliminated intermediate string and array allocations from repeated calls.

Implementation notes

getNameParts returns a defensive slice. Two existing callers mutate the returned array — BrsFile.ts:999 uses .shift(), symbolUtils.ts:141 uses .pop(). The slice on each call is trivial relative to the AST walk it replaces, and preserves the original mutable-array contract for any external callers.

Statement-level getName only writes to the cache when this.parent is set. Without that guard, NamespaceStatement's constructor (which calls this.name to build a SymbolTable label before the AST walk that establishes parent links) would lock in the parentless local name forever. The guard defers caching until after parent linking, so a nested namespace correctly caches its full qualified name on first post-walk read.

Cache fields use the _cached* prefix. AstNode.spec's clone-equality test was updated to skip those slots: the cache populates lazily and need not match between a node and its clone (each populates independently on first read).

🤖 Generated with Claude Code

[rokucommunity-bot]

Hey there! I just built a new temporary npm package based on 52394bb. You can download it here or install it by running the following command:

npm install https://github.com/rokucommunity/brighterscript/releases/download/v0.0.0-packages/brighterscript-0.71.1-perf-name-memoization.20260429001621.tgz

[TwitchBronBron]

I don't really want to do this at this time. Too much difficulty knowing when to clear these cached values anytime AST edits are made. We need to think about this more and come up with a better more broad caching system.