The previous approach to error recovery relied on special error-recovery
states in the parse table. For each token T, there was an error recovery
state in which the parser looked for *any* token that could follow T.
Unfortunately, sometimes the set of tokens that could follow T contained
conflicts. For example, in JS, the token '}' can be followed by the
open-ended 'template_chars' token, but also by ordinary tokens like
'identifier'. So with the old algorithm, when recovering from an
unexpected '}' token, the lexer had no way to distinguish identifiers
from template_chars.
This commit drops the error recovery states. Instead, when we encounter
an unexpected token T, we recover from the error by finding a previous
state S in the stack in which T would be valid, popping all of the nodes
after S, and wrapping them in an error.
This way, the lexer is always invoked in a normal parse state, in which
it is looking for a non-conflicting set of tokens. Eliminating the error
recovery states also shrinks the lex state machine significantly.
Signed-off-by: Rick Winfrey <rewinfrey@github.com>
We previously maintained a set of individual productions that were
involved in conflicts, but that was subtly incorrect because
we don't compare productions themselves when comparing parse items;
we only compare the parse items properties that could affect the
final reduce actions.
For some reason, there was previously some extra logic that prevented
the external scanner from being invoked if the only valid external
token also had an internal definition.
It's surprising to not call the external scanner if an external
token is valid.
This speeds up parser generation by increasing the likelihood that we'll recognize
parse item sets as equivalent in advance, rather than having to merge their states
after the fact.
While generating the parse table, keep track of which tokens can follow one another.
Then use this information to evaluate token conflicts more precisely. This will
result in a smaller parse table than the previous, overly-conservative approach.
This fixes some randomized test failures in the C grammar, relating to Object-like macros.
The object-like macro rule relies on a whitespace token in order to distinguish object-like
macros whose values begin with a '(' from function-like macros. The presence of that
whitespace token means that other nodes should not be reusable in that state.
This fixes a bug in the C++ grammar where the `>>` token was merged into
a state where it was previously not valid, but the `>` token *was*
valid. This caused nested templates like -
std::vector<std::pair<int, int>>
to not parse correctly.
