This is what really made our outlandishly large RWC project an outlier (now it takes less time than vscode!). Every time we went to "lookup" a file - be it for an exists or a read, we would iterate over every file read in the project. Talk about accidentally quadratic! This cuts ~30s of cpu time out of the RWC tests on my machine from our worst test (bringing wall-clock run time down by another ten seconds).