parser: more efficient treatment of string literals.

In this patch we switch the string literal parser from right
recursion to left recursion, so that it doesn't require a Yacc
stack depth proportional to the number of characters in the
literal.  Secondly, we build the string directly in a
syntax-directed way, rather than building a list of characters
and then walking the list to build a string.

This was discovered as a byacc regression, though the
fix is not for the sake of byacc but basic efficiency.

The Sep 29, 2015 commit
111650e235ab2e529fa1529b1c9a23688a11cd1f
"Implementation of static slots for structures."
extended the string literal in the struct.tl test case in such
a way that if the parser is generated by byacc rather than GNU
Bison, the test case fails with a "yacc stack overflow". I
haven't done any regression testing with byacc in over two
years so I didn't notice this.

Quasiliterals could use this treatment also.
Word list literals benefit from this change, but they still
use a Yacc stack depth proportional to the number of words,
since the accumulation of words is right recursive.

* parser.y (lit_char_helper): Static function removed.
(restlitchar): New grammar nonterminal symbol.
(strlit, quasi_item, wordslit): No need to call
lit_char_helper.
(litchars): A litchars is now either a single LITCHAR,
or else a LITCHARS followed by a sequence of more.
This sequence is a separate production rule called
restlitchar, which is purely left recursive. (If
litchars is made directly left recursive without
this helper rule, intractable reduce/reduce and
shift/reduce conflicts arise.)

0 files changed, 0 insertions, 0 deletions