libc VS go-sqlite3

I don't think I understand what the argument for that is, because I've only ever heard it articulated as "cgo isn't go" which doesn't really convey much information.

Is this go? https://gitlab.com/cznic/libc/-/blob/master/libc_openbsd.go?...

I mean technically I suppose it is code that conforms to the go language grammar, but I'm not sure why a language purist would accept this.

Maybe you can contribute a fix? The related discussion is here: https://gitlab.com/cznic/libc/-/issues/20

struggling figuring out how to make my cgo sqlite cross-compile to Windows

Plenty of people trying to fix that.

There's at least:

https://modernc.org/sqlite

Then there's https://github.com/zombiezen/go-sqlite that actually builds https://crawshaw.io/sqlite on top of modernc.

And there's mine that has both a low level and a database/sql driver builds and runs everywhere Go does: https://github.com/ncruces/go-sqlite3

If you're interested both SQLite's and my memdb VFSes implement safe locking.

Depending on your familiarity with Go, mine maybe easier to follow, or not.

https://github.com/ncruces/go-sqlite3/blob/f1b00a9944730eaa9...

One thing I tried to make sure, to avoid the pitfall modernc is having, is to make sure building "the WASM BLOB" is easily reproducible with widely available tools:

https://github.com/ncruces/go-sqlite3/blob/main/.github/work...

I do apply some light patches to SQLite, but so far they've always cleanly applied, and I can produce a new release within hours of being notified of SQLite releases.

Disclosure: I'm working on alternative Cgo-less bindings for SQLite, using wazero.

https://github.com/ncruces/go-sqlite3

One of the problems of the modernc approach (IMO) is that they're not just transpiring CPU/compute stuff, but entirely OS/platform stuff.

Each Go file of theirs is a xxx_os_arch.go that starts with 100s of OS-#defines-as-consts, and goes on to transpile fully #ifdefed code.

It also implements antithetical (in Go) stuff like goroutine local storage, because libc pthreads can't live without it.

And all IO is via direct syscalls that will never play nice with the Go scheduler, because, again this is OS level stuff.

WASM defines a cross platform CPU and an ABI, and using that for compute and the bottom OS layer in Go you get (IMO) a nicer end result.

Given the hard task of generating decent code from WASM at load time (wazero's compiler is pretty naive, a better one is being developed, but it will take seconds to generate good code for anything non trivial like SQLite) I wouldn't mind having a solution that translated to Go, or Go ASM, at build time.

The github.com/ncruces/go-sqlite3 is a link.

Using the stack pointer global is an interesting hack. I'd never thought of that. Need to compare with what I'm doing for SQLite (a kind of per connection arena).