GitLab

This extracts the mojom parsing logic from mojom_bindings_generator.py
into a new mojom_parser.py script within the mojom tool directory.

Apart from simply moving the parsing logic to a new script, this also
changes how the parse result is persisted by an intermediate build step
for subsequent build steps to consume.

Prior to this change, the parsing step emitted a ".p" file for each
processed .mojom file. This was nothing more than a serialized version
of the parsed AST with [EnableIf] feature filtering applied. Every
subsequent step for the mojom in question would load this AST, then
recursively process its import dependencies while loading their ASTs in
turn, before finally processing all of the aggregated data (resolving
references, computing field packing data, etc) to produce a Module
object suitable for bindings generation.

This CL rolls most of that processing into the parsing step (with more
to follow in future CLs), and the resulting computed Module object is
persisted to a ".mojom-module" file.

To generate bindings for a given .mojom file, the generator can simply
call Module.Load() on the corresponding .mojom-module produced by the
parser. The loaded Module is sufficient to inform code generation
with no need to resolve dependencies, process other modules, or
perform other complex computations on the model.

Each mojom-module is thus entirely self-contained, replicating the full
contents of all transitive import dependencies. This means the build
consumes more disk space with intermediate build artifacts, but it
also means generation steps are faster and simpler.

This CL should result in no net changes to the final build outputs.

Bug: 1060471
Change-Id: Ic21b737409ac1329f0f8a8751bf021f2de4768a4
Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/2128966
Commit-Queue: Ken Rockot <rockot@google.com>
Reviewed-by: Oksana Zhuravlova <oksamyt@chromium.org>
Cr-Commit-Position: refs/heads/master@{#755952}