Hacker News Reader

I think serde should be part of the stdlib. A http server and client library is also really a must-have today. Maybe some abstraction for SQL database drivers.

Some other candidates include: logging, anync runtime (Tokyo), random generator, regex

View on HN

Let me give you some contrasting views on this: we already know that serde has (serious) limitations and sooner or later a completely new version will be needed. HTTP and Client library will not make sense unless async is also in the core. Moving async in the core is going to raise serious problems for people trying to port this to new ecosystems and there is not even agreement on exactly how async should continue into the future.

In fact, there are people (like me) who are very happily pushing back on moving stuff into the standard library unless it's clear that the code there can live for 30 years. And I do not see this for almost all the things you just mentioned.

What are Serde's serious limitations?

You're constrained by the internal serde data model. For instance serde has no bignum in the data model, so you can't express it in that data model which requires awful workarounds. I went into the types of abuses necessary here a bit: https://lucumr.pocoo.org/2021/11/14/abusing-serde/

What being a part of standard library makes the library better?

Reply on news.ycombinator.com