Performance of getSemanticDiagnostics in compiler API

The setup is somewhat similar to the one in the compiler API example (https://github.com/Microsoft/TypeScript/wiki/Using-the-Compiler-API#incremental-build-support-using-the-language-services)

The problem, as in the example, is that errors are reported only for files that were directly changed. It doesn't cover the case in which the changed file was a dependency of another file, and even though the file depending on it was not changed, it will still have errors, which won't be reported.

To be more specific, the setup I am talking about is ts-loader + webpack.

What happens is that getSemanticDiagnostics(file) has to be called for every file in the project, which ends up increasing the incremental build times considerably. Even program.getSemanticDiagnostics() iterates through all of the source files as it is seen in the typescript source code.

I can only see this problem fixed from the compiler API. It could expose a method which finds all files dependent on a given file. Because in an incremental build only a single file changed, this should be faster than calling getSemanticDiagnostics() for each and every project file.

The alternative would be some sort of caching on getSemanticDiagnostics() itself, but the compiler would probably have to do internally the same work as above to invalidate the cache on file change. This seems to be equivalent to a getSemanticDiagnosticsForFileAndDependents(file)-like function.

Start by reading the TypeScript compiler API implementation of getSemanticDiagnostics(file) and program.getSemanticDiagnostics(), including the source code referenced in the issue. Determine how incremental builds identify files dependent on a changed file, then assess an API or caching approach that avoids checking every project file while still reporting all affected diagnostics.