@0no-co/graphqlsp@1.16.0

Minor Changes

• Reduce type-checker work in findAllCallExpressions. The gql.tada function detection and schema-name lookups are now memoized per callee, so ordinary calls with a string-literal first argument (e.g. t('key'), it('name', fn)) no longer trigger a type probe per call site. Fragment unrolling is deduplicated for repeated references to the same fragment, and the third argument of findAllCallExpressions now also accepts an options object ({ searchExternal?: boolean; collectFragments?: boolean }), where collectFragments: false skips fragment collection entirely for callers that only consume nodes
  Submitted by @JoviDeCroock (See #392)
• Rewrite unused field detection as a single-pass, symbol-driven analysis. Instead of issuing a find-all-references request per binding and enumerating the lexical scope per reference, the file's AST is walked once to index identifier occurrences, and query-result bindings are resolved through getSymbolAtLocation against a selection-set trie. This removes the quadratic language-service calls from getSemanticDiagnostics, scales to files with many documents and consumers, and makes the analysis work without resolved document types.
  The analysis also tracks more usage patterns, reducing false positives: values passed as function arguments (including property chains like format(data.pokemon.weight)), spreads and object-literal assignments now mark the affected sub-selections as used, for...of loop bindings are followed like array-method callbacks, and assignments (existing = result.data.pokemon) and conditional expressions now alias like variable declarations
  Submitted by @JoviDeCroock (See #393)

Patch Changes

• ⚠️ Fix multi-level property accessor when resolving fragments in gql.tada generate-persisted
  Submitted by @felamaslen (See #389)