feat: sqlx sqlite expose de/serialize

[mattrighetti]

This PR exposes two new functions for sqlite (#192):

• serialize
• deserialize

I've tried to follow what @abonander suggested #192 (comment)

Related docs

• serialize
• deserialize
• sqlite malloc

Serialize

serialize_nocopy uses sqlite3_serialize with the flag SQLITE_SERIALIZE_NOCOPY enabled so no new buffer is allocated. It is implemented on LockedHandle so that we're sure the buffer does not outlive the handle lifetime as noted in the sqlite docs:

After the call, if the SQLITE_SERIALIZE_NOCOPY bit had been set, the returned buffer content will remain accessible and unchanged until either the next write operation on the connection or when the connection is closed, and applications must not modify the buffer. If the bit had been clear, the returned buffer will not be accessed by SQLite after the call.

I've then implemented the serialize function on SqliteConnection that basically locks the handle and copies the buffer returned by serialize_nocopy into an owned type Vec<u8> that will be returned to the caller.

Deserialize

deserialize requires a bit more work. First of all I've created a wrapper type SqliteOwnedBuf for a buffer allocated with sqlite3_malloc. That buffer is useful in sqlite3_deserialize since it enables us to use the SQLITE_DESERIALISE_FREEONCLOSE flag that will notify sqlite3_deserialize to manage the buffer and free the allocation when it's done with it.

Notes

This still needs testing, but I wanted to first understand if there are other changes needed for this implementation.

Also, I really doubt that the Err I've implemented are correct, I'd need some guidance on how we should handle those failures

Looking forward to your feedback, thanks!

[mattrighetti]

I've implemented the serialize method on SqliteConnection so that it returns an SqliteOwnedBuf. I'm still unwrapping some type conversions, I'll fix those in a follow-up change.

It would be cool to have a single serialize function on SqliteConnection which could return both the default and the nocopy buffers but I still have to figure out how to achieve the latter.

[mattrighetti]

Update: I've moved serialize() and deserialize() on the worker thread but I have little knowledge of the codebase so feel free to point me in the right direction.

I had to introduce two new Command enumerations Serialize + Deserialize so that they can be used and passed in the worker tx channel.

serialize() probably requires more work, as @abonander pointed out

it may have to wait for concurrent write transactions from other connections to complete

plus, I've implemented Send + Sync for SqliteOwnedBuf just to silence the compiler complaints because NonNull<u8> is not thread-safe ofc.

I've also removed the NOCOPY implementation until we figure out how to handle that.

[abonander]

@mattrighetti on further review, there were a lot of things I wanted to change and it would have been unwieldy to do it all in review comments. I took the liberty of pushing a commit with the changes:

• expanded docs better matching the style of the rest of the project
• better error handling
• integration tests

Take a look and let me know what you think.

[mattrighetti]

lgtm!

Not 100% convinced about the serialize(Option<&str>) but I guess it makes sense for this first iteration.

In the future we could introduce a new function serialize_schema(&str) where we explicitly ask for a schema while keeping serialize() as "main" schema serialization.