Improve frontend startup performance

[sdirix]

The frontend awaits 'initialize', 'configure' and 'onStart' for all frontend contributions. It's therefore very important that no expensive code is run there.

By not awaiting expensive operations in

• DebugFrontendApplicationContribution.onStart (~300ms)
• EditorNavigationContribution.onStart (~36-400ms)
• TerminalFrontendContribution.onStart (~100-300ms)

the reported startup time without using plugins is reduced by ~400-1000ms which is an improvement of ~20-40%.

Contributed on behalf of STMicroelectronics

What it does

It removes some awaited instructions in the critical execution paths which are not really necessary to be awaited. The remaining await instruction in the common-frontend-contribution which takes about ~200ms of time to be executed was not removed as this might have some side effects in theory, although in practice I was not able to observe any.

---

Note: I also tried to run all initialize, configure and onStart phases each in "parallel", using Promise.all. However when combined with the remaining changes this actually lead to a slowdown, probably because the Javascript thread is already fully utilized.

How to test

Verify that Theia behaves as before. Please have a look at the code paths to see whether something was missed.

You can start Theia with logLevel=debug for more extensive logging and verify the reported startup times. For proper logging in production mode, you need to disable class name substitution. This can be achieved by adding the following to the webpack.config.js:

const TerserPlugin = require('terser-webpack-plugin');
configs[0].optimization = {
    minimizer: [
        new TerserPlugin({
            terserOptions: {
                keep_classnames: true
            }
        })
    ]
};

Review checklist

• As an author, I have thoroughly tested my changes and carefully followed the review guidelines

Reminder for reviewers

• As a reviewer, I agree to behave in accordance with the review guidelines

[JonasHelming]

@kittaakos : Maybe you are interested in reviewing this one?

[msujew]

The changes look good to me. I cannot identify any regressions with the approach and the startup time of frontend contributions is indeed reduced.

Though not as much as advertised, but the initial startup time wasn't as long anyway. Reduced the FrontendApplicationContribution#onStart time from ~500ms to ~300ms for me.

[sdirix]

I measured the times only on my own machine ;)

[sdirix]

I rebased the branch on latest master to solve conflicts in the Changelog. Nothing else changed.

[msujew]

@sdirix I wasn't fast enough to press the merge button. Please rebase again :)