This commit completely rewrites this crate from the ground up,
supporting serde at the lowest levels as I believe serde support was
intended to do. This is a major change from the previous versions of
this crate, with a summary of changes being:
* Serialization directly to TOML is now supported without going through
a `Value` first.
* Deserialization directly from TOML is now supported without going
through a `Value`. Note that due to the TOML format some values still
are buffered in intermediate memory, but overall this should be at a
minimum now.
* The API of `Value` was overhauled to match the API of
`serde_json::Value`. The changes here were to:
* Add `is_*` accessors
* Add `get` and `get_mut` for one-field lookups.
* Implement panicking lookups through `Index`
The old `index` methods are now gone in favor of `get` and `Index`
implementations.
* A `Datetime` type has been added to represent a TOML datetime in a
first-class fashion. Currently this type provides no accessors other
than a `Display` implementation, but the idea is that this will grow
support over time for decomposing the date.
* Support for the `rustc-serialize` crate has been dropped, that'll stay
on the 0.2 and 0.1 release trains.
* This crate no longer supports the detection of unused fields, for that though
you can use the `serde_ignored` crate on crates.io
Rationale:
- The path has nothing to do with the result.
- The path has no need to live as long as the Value/self.
- In some cases it can be hard to actually build a path that meets
the same lifetime requirements as the Value or String slice result.
The new algorithm allows the explicit usage of "" and '' to denote key
names. This is useful for accessing tables or keys that are named in a
non-conventional manner.
Mutable access may sometimes be desired in order to change values
in the toml table. This can be used for dynamic configurations which
will be easy to modify and store.
lookup_mut requires a recursive method due to the borrow checker
not allowing to have more than one mutable reference in the same
scope.
This also changes the calculated line and column numbers. Without this
patch, if an error occurs at the end of a line, the returned line and
column numbers will point at the start of the next line.
After this patch, the line and column numbers will correctly point at
the end of the line where the actual parse error happened.
