Remove limit on the number of backrefs possible.

Changes named sub-expressions to use different hashing scheme: high order bit is now always set to clashes between hashes and indexes don't happen until 2^30 or 2^62 sub-expressions in 32 and 64 bit code respectively.
Changes bitmask of seen sub-expressions to use dynamic storage for sub-expression indexes above 64.
Adds tests for the above.
Fixes https://github.com/boostorg/regex/issues/75.
This commit is contained in:
jzmaddock
2020-01-19 11:28:36 +00:00
parent b5d60694cc
commit 4bb4d392e4
7 changed files with 81 additions and 20 deletions

View File

@ -609,7 +609,7 @@ bool perl_matcher<BidiIterator, Allocator, traits>::match_backref()
// or PCRE.
//
int index = static_cast<const re_brace*>(pstate)->index;
if(index >= 10000)
if(index >= hash_value_mask)
{
named_subexpressions::range_type r = re.get_data().equal_range(index);
BOOST_ASSERT(r.first != r.second);
@ -758,7 +758,7 @@ inline bool perl_matcher<BidiIterator, Allocator, traits>::match_assert_backref(
{
// Have we matched subexpression "index"?
// Check if index is a hash value:
if(index >= 10000)
if(index >= hash_value_mask)
{
named_subexpressions::range_type r = re.get_data().equal_range(index);
while(r.first != r.second)
@ -782,7 +782,7 @@ inline bool perl_matcher<BidiIterator, Allocator, traits>::match_assert_backref(
// Have we recursed into subexpression "index"?
// If index == 0 then check for any recursion at all, otherwise for recursion to -index-1.
int idx = -(index+1);
if(idx >= 10000)
if(idx >= hash_value_mask)
{
named_subexpressions::range_type r = re.get_data().equal_range(idx);
int stack_index = recursion_stack.empty() ? -1 : recursion_stack.back().idx;