* Doctest column name reference
* Document rusqlite assumption on column name reference
And move doctest as a test.
* Document when columns metadata should be extracted.
* Rustfmt doc (wrap_comments)
This change implements `TryFrom<&Row>` for tuples up to 16 fields. This
is a convenience function that can be used to map rows more easily.
The change includes tests for 1-tuple, 2-tuple and 16-tuple.
Replace `Row::get` by `Row::get_checked`,
And rename original `Row::get` to `Row::get_unwrap`.
`Stmt::query_map`, `Stmt::query_map_named`, `Stmt::query_row`,
`Conn::query_row` and `Conn::query_row_named` callback parameter must return a `Result`.
This is behind the `i128_blob` feature.
Blobs are stored as 16 byte big-endian values, with their most significant bit
flipped. This is so that sorting, comparison, etc all work properly, even with
negative numbers. This also allows the representation to be stable across
different computers.
It's possible that the `FromSql` implementation should handle the case that the
real value is stored in an integer. I didn't do this, but would be willing to
make the change. I don't think we should store them this way though, since I
don't think users would be able to sort/compare them sanely.
Support for `u128` is not implemented, as comparison with i128 values would work
strangely. This also is consistent with `u64` not being allowed, not that I
think that would be reason enough on it's own.
The `byteorder` crate is used if this feature is flipped, as it's quite small
and implements things more or less optimally. If/when `i128::{to,from}_be_bytes`
gets stabilized (https://github.com/rust-lang/rust/issues/52963), we should
probably use that instead.
