Bun claims their SQLite driver is 3-6x faster than Node.js with better-sqlite3

[Prinzhorn]

Bun (some sort of Node.js alternative based on JavaScriptCore) is claiming their SQLite driver (inspired by better-sqlite3) is 3-6x faster than better-sqlite3 running in Node.js (see claim in the docs here https://github.com/oven-sh/bun/blob/682406c42ceee590c2b90900175f0eb1eaed5637/docs/api/sqlite.md). The benchmark is located here https://github.com/oven-sh/bun/blob/682406c42ceee590c2b90900175f0eb1eaed5637/bench/sqlite/node.mjs

I've already posted a lengthy post in their tracker here oven-sh/bun#4776

I'm mostly posting this here because I'd be very curious to hear what @JoshuaWise thinks why the SELECT * case would be so much faster with Bun. I can't believe that this would be purely a JavaScriptCore vs V8 thing, there must be something else going on. Maybe their bridge does something better when converting the SQLite data into the JavaScript world that better-sqlite3 could adopt?

[Prinzhorn]

It might be because of this: https://bun.sh/docs/api/ffi

So this wouldn't be a SQLite thing per say, just that their native API is better? (nodejs/node#46233)

[CovertCode]

I started creating my CMS using bun and when I uploaded the to my server I kept on getting memory dump SIGNIT something like this error. Some errors were solved by bun update and some by kernel. But then again I started getting those errors. Most of the time I spent on solving those errors. So I moved my project to nodejs. Recently I saw another issue raised in bun git core dump.

These kinds of errors are mostly related to bun core and I don't think developer needs to spend 30% of time in solving major memory errors.
I won't be using bun for next whole year. There are some major issues which are not known to their team because it's based on OS and kernel. I never got any error in WSL while I got errors in digitalocean VPS.

[Prinzhorn]

I'm not sure what this adds to the discussion and how this relates to better-sqlite3 in any way? This is not an anti-Bun post, I simply want to understand why the SQLite-JS interop is faster so we can potentially improve better-sqlite3 as well.

[sidesteps]

This is buns' "bridge":
https://github.com/oven-sh/bun/blob/main/src/bun.js/bindings/sqlite/JSSQLStatement.cpp

The driving logic is in JSC_DEFINE_HOST_FUNCTION(jsSQLStatementExecuteStatementFunction...
methods

At the moment it has plenty of leftover/misplaced code from pull requests.
Also, functionality is missing in bun. Like returning affected row count with statement.run().
Bun also doesn't return rowid for run() statement like better-sqlite3 does. Although that's probably a good thing.