Not all assert calls are being removed from v0.js by babel transforms

[rsimha]

#23377 updated the babel transforms to remove assert calls. This should have paved the way for the removal of AmpPass.java from the custom closure compiler runner. However, removing AmpPass.java increases the number of instances of assert in v0.js from 20 to 80.

Before removing AmpPass.java: 20 instances
After removing AmpPass.java: 80 instances

See discussion at #23377 (comment)

[rsimha]

This is still an issue. Reopening based on the discussion at #24047 (comment)

[kevinkimball]

@erwinmombay is this issue still relevant?

[rsimha]

@erwinmombay is this issue still relevant?

Yes, this is still relevant, and is the only thing that's blocking #24047. Start reading from #24047 (comment) for the latest status.

/cc @kristoferbaxter who is waiting on #24047 to be merged in order to fix a concurrency bug with our custom closure compiler runner.

[rsimha]

@jridgewell @erwinmombay Can we prioritize this fix and unblock #24047? Our inability to remove AmpPass.java is resulting in cascading build problems, because (as discovered by @kristoferbaxter), runner.jar does not reliably run in a concurrent fashion due to the code in AmpPass.java.

[kristoferbaxter]

I’m out on paternity leave, but started work on a refactor of the Babel transforms necessary to complete this transition.

Note, once complete you can use the async pool implementation (used internally in filesize and open sourced separately) and remove nailgun. Those combined improve compilation times for everyone significantly.

All together on a sample machine it was a 12x reduction in build time.

An additional optimization to finish later would be to request the compilation of all resources together as a complete array instead of extensions separately from runtime.

[rsimha]

Reopening due to non-trivial diffs in extension code, leading to test failures: #24047 (comment)