To support internal compilations, dart2wasm must be able to run in
phases. There are a few reasons for this:
1) Kernel transforms are run on the program after the CFE has run. We
must emit a dill that frameworks can transform and then pass the
transformed dill back to dart2wasm.
2) This allows us to avoid forge limits by running each phase of the
compiler in separate blaze actions. TFA has the chance of running long
on large programs and so it might be beneficial to run it as its own
action.
This implementation currently supports 3 phases: "cfe", "tfa", "codegen"
They can be run collectively or in any consecutive combination. Phases
are specified via a '--phases' multi-option. Any data that needs to be
passed between the phases is encoded directly into the serialized dill.
This also opens up the opportunity to make "opt" its own phase that
runs wasm-opt on the wasm emitted from the codegen phase.
Change-Id: Ide830763f7063c7ab880e8e54dc47bd32fd4e7cd
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/455280
Commit-Queue: Nate Biggs <natebiggs@google.com>
Reviewed-by: Martin Kustermann <kustermann@google.com>
Nate Biggs
79fa199a2b