fix(sqlite): allow queries without arguments to be persistent

[link2xt]

No description provided.

[link2xt]

The problem is that sqlx forces statements without arguments to be non-persistent. Because of this there is currently no way to make all the queries persistent except by adding arguments to all of them artificially.

This is actually a workaround for deltachat/deltachat-ios#1129 It's likely caused by sqlx finalizing the query too early. It's hard to prove that query is not finalized by the time sqlite3_step is executed because StatementHandle accesses raw sqlite statements via C pointer but does not actually own it and VirtualStatement can be dropped independently, causing the statement to be finalized. Making all statements persistent ensures that they are only finalized when they are evicted from the cache, and this happens after executing 100 other different queries by default, so it's unlikely that they are freed too early. Fixing the original problem would be nice, but I don't see the reason not to allow statements without arguments to be cached anyway and it was unexpected to see them evicted immediately despite not setting .persistent(false) explicitly.

[mehcode]

So this is actually largely intentional. Any query built from sqlx::query( ... ) ( or that family of functions ) is by-default persistent.

The "low level" execution APIs intentionally do not touch the statement cache.

In other words,

// persistent
sqlx::query("SELECT 10").execute(&mut conn).await?;

// not persistent
conn.execute("SELECT 10").await?;

[link2xt]

So this is actually largely intentional. Any query built from sqlx::query( ... ) ( or that family of functions ) is by-default persistent.

Is it documented somewhere? It then makes sense to add one non-persistent statement like this to the test and close this PR without merging.

@r10s tests this workaround on iOS now, if it works we can just wrap every query in deltachat-core-rust in sqlx::query.

The real bug is that sqlite driver does not guarantee that sqlite3_step-ped statements are not finalized prior to that, because StatementWorker only gets a StatementHandle which does not actually own the statement and VirtualStatement can be dropped asynchronously. Making all queries persistent is a workaround, but not a fix anyway.

[link2xt]

I opened #1165 as a replacement of this PR.