Use id of last module in chunk as name base for auto-generated chunks #3025
Conversation
|
@jakearchibald I assume this would also solve the original issue of #2793 even though it is not configurable? |
Codecov Report
@@ Coverage Diff @@
## unified-file-emission-api #3025 +/- ##
=============================================================
+ Coverage 88.63% 88.65% +0.01%
=============================================================
Files 165 165
Lines 5693 5694 +1
Branches 1738 1737 -1
=============================================================
+ Hits 5046 5048 +2
Misses 388 388
+ Partials 259 258 -1
Continue to review full report at Codecov.
|
* Use new FileEmitter for basic cases around assets. TODO: Replace asset hooks completely, create tests for assets using the new hooks. * Migrate assets to new file emitter * Remove assetsById from Graph * Implement emitFile for assets * Internally use EmittedFile in the file emitter * Deprecate emitAsset and ROLLUP_ASSET_URL * Deprecate getAssetFileName * Merge chunk emission into unified API and deprecated previous API * Allow emitting files with fixed names * Support unnamed assets * Improve chunk name assignment * Initial support for chunk file names * Allow specifying explicit file names for emitted chunks * Fix some TODOs * Test ids remain stable when the transform hook is cached and make test more stable * Refine error handling * Test some more errors * Refine file emission * Refactor file emitter to have a single code path for asset finalization * Deduplicated emitted assets without a specific file name * Only use the alias as name for a manual chunk if the chunk is not facade for an entry point * Generate separate facades for duplicate named user-defined entry points * Always create facades for explicit file names * Test edge cases * Test and refactor handling of dynamic relative imports * Use async-await in generate function, remove error condition * Improve and test pattern validation * Test file emitter edge cases * Improve plugin error handling * Add documentation
|
This will be huge. I was just pondering if this was achievable via a plugin using some real goofy code but I like this version way better 😄 |
|
@lukastaegert happy to give it a spin. Is it in a testable state now? |
|
Definitely testable. |
|
Yeah this is worlds better. GoogleChromeLabs/proxx#493 I still get a few chunk names named |
|
Maybe, but I'm not really sure this is always a good idea. Also somewhat difficult from where the assignment happens as it does not know the other names. |
|
Maybe |
* Switch to a code-splitting build and update dependencies * Fix case where default exports were not properly deconflicted against chunk names
lukastaegert
force-pushed
the
better-generated-chunk-names
branch
from
6e6dec9
to
dd0ed65
Compare
|
I think I will start with the current approach as it is very simple implementation-wise and can be rolled out immediately and we can think about further improvements in another task. Other approaches will both require more implementation work and have more edge-cases that need to be considererd and I'm not sure about some of them. Maybe it is actually possible to do it based on conflicts and then maybe create names such as |
…#3025) * Use new FileEmitter for basic cases around assets. TODO: Replace asset hooks completely, create tests for assets using the new hooks. * Migrate assets to new file emitter * Remove assetsById from Graph * Implement emitFile for assets * Internally use EmittedFile in the file emitter * Deprecate emitAsset and ROLLUP_ASSET_URL * Deprecate getAssetFileName * Merge chunk emission into unified API and deprecated previous API * Allow emitting files with fixed names * Support unnamed assets * Improve chunk name assignment * Initial support for chunk file names * Allow specifying explicit file names for emitted chunks * Fix some TODOs * Test ids remain stable when the transform hook is cached and make test more stable * Refine error handling * Test some more errors * Refine file emission * Refactor file emitter to have a single code path for asset finalization * Deduplicated emitted assets without a specific file name * Only use the alias as name for a manual chunk if the chunk is not facade for an entry point * Generate separate facades for duplicate named user-defined entry points * Always create facades for explicit file names * Test edge cases * Test and refactor handling of dynamic relative imports * Use async-await in generate function, remove error condition * Improve and test pattern validation * Test file emitter edge cases * Improve plugin error handling * Add documentation * Use id of last module in chunk as name base for auto-generated chunks
|
FWIW, looks like this change broke a few things for a few people. I outlined my issue in this comment: #2839. |
|
FWIW I did a POC with some small plugins to tweak dynamic import names using the file emission API and relying on chunk deduping by ids. Haven't tested them in a real environment and need some externals and error handling but I think they may be good enough to solve most chunk naming issues. https://gist.github.com/isidrok/d42cb4e71e9c54fe0260ffe300125ef1 |
|
@isidrok that is really interesting |
This PR contains:
Are tests included?
Breaking Changes?
List any relevant issue numbers:
Description
This will avoid the generic
chunkname for auto-generated chunks by using the last module in the execution order of a chunk as base for the[name]part of the file name. The rationale is that since dependencies are always sorted first in the execution order, the last module in a chunk is what was originally imported by at least part of the chunks that depend on the auto-generated chunk.To get something close to the old naming scheme, one could use
output.chunkFileNames: 'chunk-[hash].js'