Memory leak from child compilation in watch mode (Sources strings)

[helloitsjoe]

Bug report

NOTE: This issue is very similar to #17851 (child compilations), but these are two separate leaks. This one is related to duplicate strings added to the in-memory cache.

What is the current behavior?

When running in watch mode, Webpack caches source code from child compilations (e.g. HtmlWebpackPlugin) without removing them:

Duplicated strings in memory

Retained Sources

The leak is worse when we have build-time dependencies, as in the reproduction repo:

Duplicated build-time dependencies source

Investigation done

One place these are being cached is in a Map in MemoryCachePlugin. When a child compiler caches them it uses the child compiler ID, for example (note the HtmlWebpackPlugin_0-7, the 7th time I ran a recompilation):

'HtmlWebpackCompiler|0|Compilation/codeGeneration|javascript/esm|data:text/javascript,__webpack_public_path__ = __webpack_base_uri__ = htmlWebpackPluginPublicPath;|HtmlWebpackPlugin_0-7'

This seems problematic because entry will never be found, since the key uses an ID unique to each iteration.

If the current behavior is a bug, please provide the steps to reproduce.

Minimal reproduction repo including steps to reproduce here: https://github.com/helloitsjoe/webpack-cache-memory-leak

What is the expected behavior?

Dependency source code should be cached only once (or maybe twice including the child compilation). A few potential options:

1. Remove the child compiler sources from the cache during _cleanupLastCompilation, but I don't see a way to clear entries from the MemoryCache map currently.
2. Only cache compiler sources for the parent compilation, or replace the previous child compilation's cached sources.

I'm also not sure MemoryCache is the only place the sources are being referenced, so fixing this leak might require more changes. I'm happy to help however I can.

Other relevant information:
webpack version: 5.89.0
Node.js version: 20.5.1
Operating System: MacOS (darwin 22.5.0)
Additional tools: HtmlWebpackPlugin (child compiler), custom loader in https://github.com/helloitsjoe/webpack-cache-memory-leak to cache build-time dependency sources

[alexander-akait]

Thank you for the issue, do you want send a PR with fix?

[helloitsjoe]

I will see if I can determine the root cause here, it's not clear to me yet. This might actually be an issue in HtmlWebpackPlugin, if that turns out to be the source I'll close this issue!

[helloitsjoe]

I think I've narrowed this down to a commit in HtmlWebpackPlugin. I've opened an issue over there, I'll follow up here if it's confirmed to be a problem there: jantimon/html-webpack-plugin#1835

[helloitsjoe]

@alexander-akait I noticed you're a maintainer of HtmlWebpackPlugin as well, does the change I'm suggesting in this PR look like a reasonable fix for this? jantimon/html-webpack-plugin#1836

I'm happy to look more into this if not.

[helloitsjoe]

Thank you!