Create chunks for virtual modules when preserving modules #2511
Conversation
lukastaegert
force-pushed
the
virtual-preserved-modules
branch
2 times, most recently
from
ca90a85
to
fcf68aa
Compare
lukastaegert
force-pushed
the
virtual-preserved-modules
branch
2 times, most recently
from
8b20bcd
to
9d26dd2
Compare
lukastaegert
force-pushed
the
external-execution-order
branch
from
2601256
to
df0caff
Compare
…g module order test
lukastaegert
force-pushed
the
external-execution-order
branch
from
df0caff
to
dbbd52d
Compare
lukastaegert
force-pushed
the
virtual-preserved-modules
branch
from
9d26dd2
to
f505bfd
Compare
|
Damn it, seems like Windows really does not like my tests. Will keep on investigating tomorrow. |
| description: 'Generates actual files for virtual modules when preserving modules', | ||
| options: { | ||
| input: ['main.js'], | ||
| experimentalCodeSplitting: true, |
There was a problem hiding this comment.
arent those like conflicting settings? preserving modules & code splitting?
There was a problem hiding this comment.
Not in rollup < 1.0. experimentalCodeSplitting internally uses the "new" code-splitting pipeline which is a prerequisite for preserving modules, which is like the ultimate form of code-splitting. In rollup 1.0 of course, this is the only pipeline that exists.
| .generate({ format: 'esm' }) | ||
| .then(generated => | ||
| assert.deepEqual(Object.keys(generated.output), [ | ||
| 'ROLLUP__virtualModule', |
There was a problem hiding this comment.
wouldnt it be better to generate hashed filename?
There was a problem hiding this comment.
I tend to disagree here. Two reasons:
- It is unclear how the file name should be generated.
- The point of preserving modules is to preserve as much as possible. Nothing else is hashed and thus there is no benefit in hashing just this file. On the contrary, this will cause friction as it is not a priori clear which file names will be generated.
src/Chunk.ts
Outdated
| this.id = normalize( | ||
| isAbsolute(this.entryModule.id) | ||
| ? relative(preserveModulesRelativeDir, sanitizedId) | ||
| : 'ROLLUP_' + basename(sanitizedId) |
There was a problem hiding this comment.
Perhaps make these a folder - _virtual/..., or something like that, in case there are a lot of them?
| const inputBase = commondir( | ||
| chunks.filter(chunk => chunk.entryModule).map(chunk => chunk.entryModule.id) | ||
| chunks | ||
| .filter(chunk => chunk.entryModule && isAbsolute(chunk.entryModule.id)) |
There was a problem hiding this comment.
Doesn't the ID generation for preserve modules now guarantee this is always true?
There was a problem hiding this comment.
Not if the id is generated by a plugin. And this extends to other workflows as well.
| @@ -2,6 +2,7 @@ import * as path from './path'; | |||
|
|
|||
| // ported from https://github.com/substack/node-commondir | |||
| export default function commondir(files: string[]) { | |||
| if (files.length === 0) return '/'; | |||
lukastaegert
force-pushed
the
virtual-preserved-modules
branch
3 times, most recently
from
ba6cf4e
to
4abb2e0
Compare
lukastaegert
force-pushed
the
virtual-preserved-modules
branch
from
4abb2e0
to
b68ce59
Compare
|
I went for the |
This PR contains:
Are tests included?
Breaking Changes?
List any relevant issue numbers:
Resolves #2503
Description
When preserving modules, we also need to create chunks for virtual modules injected by plugins (e.g. babel helpers etc.). This PR solves this by doing the following:
\0characters are replaced with_ROLLUP_to reduce the danger of name clashes and turned into top-level modules (there is no check if such an id is already taken, though, therefore the prefix)