This reverts commit 5cd07648fd.
The separators construct is useful as an optimization. It turns out that
constructing a node for every chunk of whitespace in a document causes a
significant performance regression.
Conflicts:
src/compiler/build_tables/build_lex_table.cc
src/compiler/grammar.cc
src/runtime/parser.c
Now, grammars can handle whitespace by making it another ubiquitous
token, like comments.
For now, this has the side effect of whitespace being included in the
tree that precedes it. This was already an issue for other ubiquitous
tokens though, so it needs to be fixed anyway.
Now, the root node of a document is always a document node.
It will often have only one child node which corresponds to the grammar's
start symbol, but not always. Currently, it may have more than one child
if there are ubiquitous tokens such as comments at the beginning of the
document. In the future, it will also be possible be possible to have multiple
for the document to have multiple children if the document is partially parsed.
ParseItem no longer has a lookahead_sym field; it now represents
the 'core' of a parse item. The lookahead context is stored separately,
as a set per core item. This makes iterating, copying and merging item
sets more efficient, because before, the core items were repeated for each
different lookahead symbol.
Also, the memoization in sym_transitions(ParseItemSet) has been removed.
Maybe I'll add it back later.
Not needed at the moment because rule pointers are are always wrapped
in shared_ptrs. Still, don't want to forget this if I stopped using shared_ptrs
at some point.
The parser is no longer hard-coded to skip whitespace. Tokens
such as newlines, whose characters overlap with the separator
characters, can now be correctly recognized.
Now, you can't a particular occurrence of a symbol
in a grammar. You can only hide a symbol globally
(right now, by beginning its name with an underscore).
