Skipping puppeteer download seems to save about 15 seconds on Linux, let's see if caching yarn cache works.

• Baseline (2m58s): https://dev.azure.com/xtermjs/xterm.js/_build/results?buildId=2643
• Cache yarn cache, no puppeteer
  • Miss (3m22s, cache took 28s): https://dev.azure.com/xtermjs/xterm.js/_build/results?buildId=2647
  • Hit: (2m29s): https://dev.azure.com/xtermjs/xterm.js/_build/results?buildId=2648
• Cache node_modules (~300mb), no puppeteer
  • Miss (3m12s)
  • Hit (2m15s)

@Tyriar So the difference is basically a minute for a single pipeline?

@jerch yeah, integration tests are mostly unchanged at but I'll see if I can reduce that a bit too. So right now we'll get faster compile/unit test feedback.

Caching node_modules for integration tests didn't seem to work, also it ended up as 1gb so it was going to be slower anyway if it did.

Final results gives faster compile and unit test builds to let us iterate faster on PRs:

* The actual run took 2m 23s but that's because cache restore took an abnormally long time just this run, number was adjusted for typical runtime
** This is the approx value, Windows wasn't running tests before so it's the actual number 2m 55s plus 28s for unit tests

Integration tests are basically the same.

This passed fully on a rebuild