Hi folks,
This sunday I went down the rabbit hole of Meteor’s server runtime, bundle format, and boot sequence, an area I still felt weaker on and wanted to understand better because why not ? ![]()
After reading through the code, sketching a few diagrams, asking Claude way too many questions, and sanity-checking some ideas with ChatGPT, I ended up spotting a few places where the current server runtime still reflects older Node-era constraints, even though modern Node now gives us better native tools for some of these cases ![]()
So I’m starting a small PR series focused on a few very small, pragmatic cleanups in the current Node server runtime
changes that seem useful on their own, with a good signal/risk ratio, and that should be easy to review independently ![]()
The goal is simple: make a few parts of the current runtime a bit more modern, a bit less fragile, and a bit easier to reason about.
First PRs
- Add
sourceURLpragma to vm-evaluated server packages
PR: Add sourceURL pragma to vm-evaluated server packages by dupontbertrand · Pull Request #14282 · meteor/meteor · GitHub - Lazy-load
clusterinwebappwhenUNIX_SOCKET_PATHis used
PR: Lazy-load cluster in webapp when UNIX_SOCKET_PATH is used by dupontbertrand · Pull Request #14283 · meteor/meteor · GitHub
Validation so far
I validated both changes locally. Current results:
sourceURLpragma present in modified bundle- baseline and modified bundles boot and return HTTP 200
- stack traces captured successfully in both bundles
- stack trace attribution remained equivalent in my local check
UNIX_SOCKET_PATHstill worksmeteor runstill boots correctly
I’ll use this thread for any broader discussion/questions about the overall direction, so the PRs can stay focused on the diffs themselves ![]()
Thanks!