== Performance improvement ideas ==

=== 1. Low-level parsing ===

==== 1.0.1 Check name validity only once? ====

Perhaps we can combine SymbolTable, and name validation access
such that we only store valid names in SymbolTable. As such,
validity only needs to be checked if there's no hit.


==== 1.1 Integrate char decoding and scanning ====

(BIG EFFORT)

It would be possible to combine character decoding (byte -> UTF-8 char)
and low-level XML scanning, so that effectively we would have
somewhat XML-token-aware Reader. This could speed up parsing a bit,
maybe by 10% or so. Downside is that Reader would be clunkier, and
each decoder would be bigger than they are now.
An additional benefit would be 100% accurate byte offsets for the reader.



=== 2. DTD handling ===

==== 2.1 Efficient next-state data struct for DFA ====

Currently standard HashMap is used for figuring out next-state for
DFAs. It would be possible to create type-safe alternative, that
would be more compact (lower mem usage), and potentially even slightly
faster.