Arbitrary indexing into Witness fixes the API of last and second_to_last to be more flexible.
This patch started off as an addition of third_to_last, but ended up evolving
into arbitrary indexing to allow for future use cases.
A list of the indices of the start byte for each witness element is stored as an ordered
contiguous group of u32s represented as 4 bytes each in the Vec<u8> contents.
The bytes are stored using to_ne_bytes for performance reasons. A helper function is added
to the tests to allow for easier contruction of the contents Vec in test vectors. u32 was
chosen because 22 bits are needed to store 4,000,000 which is the maximum weight limit for
a block. This might need to be reworked in the event of consensus limits increasing, but
u32 can hold 1000x the current limit, so it should be fine for the forseeable future.
The push and consensus_deserialize functions utilize rotate_left and rotate_right to move
the indices to the end of the new allocation. Depending on the size of the data, this
might be more of a performance hit than just allocating a new temporary Vec to store the
indices and append them after parsing is completed. However, for a majority of cases
rotating the indices should be faster. Suggestions to use VecDeque instead of Vec for
contents necessitate other considerations, since it is not a public facing change,
those optimizations can be dealt with in future patches.
The Index<usize> trait is implemented using the new nth method with expect.
The Iter struct is reworked to make use of the new data representation. This new data
structure makes it trivial to implement DoubleEndedIterator and other such traits, but
I have decided to leave this as out of scope for this patch.
4226d60205