The result here probably does not change for a specific key once set? In which case it may be worthwhile to pass the already derived results back into the analysis as a starting point?
BUT: This should be done as a follow up (so you can open an issue), not in this PR, even if my guess is right. It is a performance optimisation, and we do not know whether this has big performance implications yet.
Originally posted by @maximilianruesch in #2274 (comment)
analyze_modifier_calls expands the call graph by computing calls to custom-modified methods. In ENGINE.check, we recompute analyze_modifier_calls for scratch every time (till fix point), but since we do not recompute the resolved_modified_calls, there is room for performance improvement.
Originally posted by @maximilianruesch in #2274 (comment)
analyze_modifier_callsexpands the call graph by computing calls to custom-modified methods. InENGINE.check, we recomputeanalyze_modifier_callsfor scratch every time (till fix point), but since we do not recompute theresolved_modified_calls, there is room for performance improvement.