chore: make terser an optional dependency

[sapphi-red]

BREAKING CHANGE: terser must be installed for minification with terser

Description

This PR makes terser an optional dependency since it is not a default option now.
This will reduce package size much.

Result of npm publish --dry-run

Before

npm notice === Tarball Details ===
npm notice name:          vite
npm notice version:       2.9.8
npm notice filename:      vite-2.9.8.tgz
npm notice package size:  1.0 MB
npm notice unpacked size: 4.6 MB
npm notice shasum:        a23ece1600b5dc3d3f1d5cc360b019e5c70d8215
npm notice integrity:     sha512-G/0LVsqqnhzBe[...]ch0ft85Fb/87A==
npm notice total files:   39

After

npm notice === Tarball Details ===
npm notice name:          vite
npm notice version:       2.9.8
npm notice filename:      vite-2.9.8.tgz
npm notice package size:  838.7 kB
npm notice unpacked size: 3.6 MB
npm notice shasum:        8872ab7e4da1c508cf458adedbda0967cdaa5763
npm notice integrity:     sha512-7KiUjzUvu05XM[...]21fevZYCZBn+g==
npm notice total files:   38

	before	after	diff
package size	1.0 MB	838.7 kB	-161.3 kB (-19.2%)
unpacked size	4.6 MB	3.6 MB	-1.0 MB (-21.8%)

Additional context

---

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.

[TrySound]

Probably not "chore" but breaking?

[sapphi-red]

https://github.com/vitejs/vite/blob/main/.github/commit-convention.md#tldr
There is no ! or breaking: in the convention. Instead, I added BREAKING CHANGE: footer.

[sapphi-red]

I selected 5.4.0 because #8045 (comment).

[sapphi-red]

This commit (925fd58) worked. CI is now happy. 🙌
So the reason why node.js crashes might be because of a race condition when dynamic import is called parallelly.
It might happen with static imports too.
The incomprehensible crash when running Vitest might be sharing the same reason.

[haoqunjiang]

I think we can use createRequire here instead https://nodejs.org/docs/latest-v14.x/api/module.html#module_module_createrequire_filename

[haoqunjiang]

Actually the one used in the CSS plugin should be changed to createRequire too

[sapphi-red]

Is it safe to remove resolving from root? I now found these related things: ddfcbce, #3988, #2612.

[haoqunjiang]

Instead of

const resolved = require.resolve(name, [root, ...fallbackPaths]);
require(resolved)

I think it's safer to use:

const requireFromRoot = createRequire(path.resolve(root, 'index.cjs'))
requireFromRoot(name)

[haoqunjiang]

See import-js/eslint-plugin-import#1591

[haoqunjiang]

Wait 🤔 This problem is more complicated than I thought…

To clarify, I'm concerned about the require method used here is not "clean" (would be interfered with by require.cache, etc.).
Both require and require.resolve should be created from the root path so that we can get a fresh instance of terser, sass, etc. The point is that require should be returned from createRequire(some-file-in-the-root-dir) too.

Here comes the complexity:

• The above-mentioned refactor will fix a theoretical problem. Though I think it's better, no known issues have been found with the old way.
• But the refactor would break a real-world use case (resolve fallbacks): fix: allow preprocessor to be installed outside of the root directory #3988
• On the other hand, if I understand correctly, the use case seems to already be broken by the ESM refactor: the _require used in css.ts now is a newly created function, which doesn't inherit require.resolve.paths from the framework that wraps Vite https://github.com/sapphi-red/vite/blob/b6ec807e863e0e27cb16846f995c8e3a5c48b9fd/packages/vite/src/node/plugins/css.ts#L1282

[haoqunjiang]

Anyway, my original suggestion is to make requireResolveFromRootWithFallback requireFromRootWithFallback (that is, put the require part into the function too).

It's a nice-to-have.

[sapphi-red]

I'm leaving this for now because I tried now but I didn't make it work.

[patak-dev]

@sodatea do you think we could merge this then? And then work on your request as an enhancement? Or is this still blocking the PR?

[haoqunjiang]

Yeah the PR looks good to me. It's not a blocking change.

[sapphi-red]

CI become unhappy now. 😢
It's failing 12 times in a row, several times with "Test serve", most times with "Test build".

Mostly it is failing with this error.

FATAL ERROR: v8::FromJust Maybe value is Nothing.
 1: 00007FF70DEE7A1F v8::internal::CodeObjectRegistry::~CodeObjectRegistry+114207
 2: 00007FF70DE76096 DSA_meth_get_flags+65542
 3: 00007FF70DE76F4D node::OnFatalError+301
 4: 00007FF70E795CD5 v8::V8::FromJustIsNothing+53
 5: 00007FF70DD92B7A v8::internal::MicrotaskQueue::microtasks_policy+22170
 6: 00007FF70DD929ED v8::internal::MicrotaskQueue::microtasks_policy+21773
 7: 00007FF70DD8CD10 v8::internal::OSROptimizedCodeCache::OSROptimizedCodeCache+42848
 8: 00007FF70DF31235 uv_tcp_close_reset+405
 9: 00007FF70DF46A73 uv_run+739
10: 00007FF70DF156C5 node::SpinEventLoop+309
11: 00007FF70DDAE740 v8::internal::wasm::SignatureMap::Freeze+35776
12: 00007FF70DDA9D98 v8::internal::wasm::SignatureMap::Freeze+16920
13: 00007FF70DF36F2D uv_poll_stop+557
14: 00007FF70ED50200 v8::internal::compiler::RepresentationChanger::Uint32OverflowOperatorFor+146416
15: 00007FFCEBEC4ED0 BaseThreadInitThunk+16
16: 00007FFCEC8AE39B RtlUserThreadStart+43
 ELIFECYCLE  Command failed with exit code 134.

This happened once.

Administrator: Windows PowerShell[5500]: c:\ws\src\node_file-inl.h:160: Assertion `finished_' failed.
 1: 00007FF6B2407A1F v8::internal::CodeObjectRegistry::~CodeObjectRegistry+114207
 2: 00007FF6B2396096 DSA_meth_get_flags+65542
 3: 00007FF6B2396451 DSA_meth_get_flags+66497
 4: 00007FF6B237F83E v8::base::CPU::has_sse+33486
 5: 00007FF6B2380D66 v8::base::CPU::has_sse+38902
 6: 00007FF6B246A997 uv_timer_stop+1207
 7: 00007FF6B2466F2B uv_async_send+331
 8: 00007FF6B24666BC uv_loop_init+1292
 9: 00007FF6B246685A uv_run+202
10: 00007FF6B24356C5 node::SpinEventLoop+309
11: 00007FF6B22CE740 v8::internal::wasm::SignatureMap::Freeze+35776
12: 00007FF6B22C9D98 v8::internal::wasm::SignatureMap::Freeze+16920
13: 00007FF6B2456F2D uv_poll_stop+557
14: 00007FF6B3270200 v8::internal::compiler::RepresentationChanger::Uint32OverflowOperatorFor+146416
15: 00007FFBD91B4ED0 BaseThreadInitThunk+16
16: 00007FFBD9E8E39B RtlUserThreadStart+43
 ELIFECYCLE  Command failed with exit code 134.

[sapphi-red]

Rerunning all CI worked. I'm not sure why rerunning only the failed test didn't work 🤔.

[bluwy]

Minor docs nitpick, but this looks great

[patak-dev]

@sapphi-red would you check the merge conflicts?