I suspect this to be an issue with the lua script call. I cannot work around this in database.lua in any way.

I can't reproduce this.

There's also no obviously related changes between 3.2.2 and 3.2.4.

Do the contents of the database look sane after first login (to create the account)?
$ sqlite3 user_auth.sqlite
SQLite version 3.46.1 2024-08-13 09:16:08
Enter ".help" for usage hints.
sqlite> select * from fcdb_auth;
1|cazfi|e7e941b1f09f266540c6780db51d5f58||1782983871|1782983871|::1|::1|1

Do the contents of the database look sane after first login (to create the account)?

Yes, they look very similar for me:

1|vego|e10adc3949ba59abbe56e057f20f883e||1782984826|1782984826|::1|::1|1

Marko Lindqvist wrote in #note-6:

Attached is a patch to add some logging to database.lua. Can you test with it, so we would learn whether the error happens already within database.lua user_exists(), or is it later in the server binary?

I have applied the patch to my installation for the file at /etc/freeciv/database.lua.

Then, doing the registration and re-login procedure gave the following output on stdout:

3: Connection request from vego from X.
3: c1 has client version 3.2.4
3: User does not exist
>
3: vego has connected from X.
>
3: Lost connection: vego from X (client disconnected) (player Vego).
3: Removing player Vego.
3: AI*1 has been added as Easy level AI-controlled player (classic).
>
3: Connection request from vego from X
3: c2 has client version 3.2.4
3: User exists
>
1: lua error:
    /etc/freeciv/database.lua:287: LuaSQL: UNIQUE constraint failed: fcdb_auth.name
stack traceback:
    : in function 'assert'
    /etc/freeciv/database.lua:287: in function 'user_save'
1: Error writing to database for: vego

I did a little searching in the codebase yesterday to figure out whether it was something I could quickly hack around and came to the conclusion that the lua return value is not properly written back to the C code.

Now I have also never worked with this, but I feel like the `bool exists` (which is passed by reference to lua via va_args) is not the size the lua code expects it to be.

Please take my considerations with a grain of salt, it's a new codebase for me and I didn't get around to testing any of them.

Could also be related to this:

man va_arg

...
If ap is passed to a function that uses va_arg(ap,type), then the value of ap is undefined after the return of
that function.
...

We first pass ap to a function to extract lua call arguments. Later we pass it to another function extracting return values, assuming it's still in the state that we left it in the first function.

Marko Lindqvist wrote in #note-8:

Another logging patch (with one cast added), this time for the C-code side.

Thank you for the patch! I'll report back as soon as I get the cross build working or directly on my server.