fix: server.address before listen, chdir in test, basic cli test

[dominikg]

Description

today vite-2.6.0-beta.1 release broke cli on node < 16 and it went unnoticed because the current tests never use the cli

This PR adds a new cli playground that uses a custom serve.js implementation that runs vite, vite build and vite preview in a subprocess.

The tests themsselves are very basic, just checking that the page comes up with expected content. No tests for arguments or anything.

Additional context

killing of the subprocess is a pain, in vite-plugin-svelte i used tree-kill for that because there's another indirection. Here it relies on execa's kill.
There are relatively short timeouts on process handling to avoid running out of the 10s jest limit, hopefully this works on CI too, otherwise it might need a larger timeout.

---

What is the purpose of this pull request?

• Bug fix
• New Feature
• Documentation update
• Other

Before submitting the PR, please make sure you do the following

• Read the Contributing Guidelines.
• Read the Pull Request Guidelines and follow the Commit Convention.
• Check that there isn't already a PR that solves the problem the same way to avoid creating a duplicate.
• Provide a description in this PR that addresses what the PR is solving, or reference the issue that it solves (e.g. fixes #123).
• Ideally, include relevant tests that fail without this PR but pass with it.

[dominikg]

not cool windows, not cool... looks like e.killed isn't true when the process was actually killed :( might take me until tomorrow to get to the bottom of this.

If anyone has ideas how to actually kill a process on windows, please chime in.

[dominikg]

this is going to have to wait until #5060 has been merged as it's likely this needs to be adapted to be usable with pnpm.

[dominikg]

force-pushed to update to pnpm

[patak-dev]

I tried and failed to kill the process in windows. Maybe we should avoid this test in windows until we figure this out? At least we can check that the cli is working correctly in other environments (so missing imports at least will not pass).

[dominikg]

on windows it only works with taskkill see https://github.com/pkrumins/node-tree-kill/blob/cb478381547107f5c53362668533f634beff7e6e/index.js#L29

i'll add that later tonight. might also look into group killing for unix to be extra safe.

[dominikg]

i'm at a loss here, the tests run fine locally, but fail on github ci with a strange error that the server isn't starting
https://github.com/vitejs/vite/runs/3710496041?check_suite_focus=true#step:9:87

stdout:

        vite v2.6.0-beta.2 dev server running at:

stderr:

      error when starting dev server:
      TypeError: Cannot read property 'address' of null
          at isAddressInfo (/home/runner/work/vite/vite/packages/vite/src/node/logger.ts:149:57)
          at Object.printHttpServerUrls (/home/runner/work/vite/vite/packages/vite/src/node/logger.ts:150:7)
          at CAC.<anonymous> (/home/runner/work/vite/vite/packages/vite/src/node/cli.ts:109:7)

[patak-dev]

🚀🥳

Thanks for not giving up 👏

[dominikg]

this is now ready for review.

originally i only wanted to add the testsuite itself, but it uncovered 2 problems along the way, both of which are fixed here too.

1. the testsuite for postcss-caching uses process.chdir which caused all tests after it to run from a different directory. It's a miracle that nothing else broke due to this.
2. refactor: change location of server start log messages #5016 changed the logging of server information but tried to access server.address() before server.listen() was called and it changed the log output of startup time.

[patak-dev]

Merging this one to be able to do a beta.3 release