This is a big redesign of failure-stack and register handling, prompted
by bugs revealed when trying to add shy-groups. Overall, what happened is that loops are now structured a little differently, groups can be shy and the code is a little simpler. (enum re_opcode_t): Remove jump_past_alt, maybe_pop_jump, push_dummy_failure and dumy_failure_jump. Add on_failure_jump_(exclusive, loop and smart). Also fix the comment for (start|stop)_memory since they now only take one argument (the second has becomes unnecessary). (print_partial_compiled_pattern): Adjust for changes in re_opcode_t. (print_compiled_pattern): Use %ld to printf long ints and flush to make debugging a little easier. (union fail_stack_elt): Make the integer unsigned. (struct fail_stack_type): Add a `frame' element. (INIT_FAIL_STACK): Init `frame' as well. (POP_PATTERN_OP): New macro for re_compile_fastmap. (DEBUG_PUSH, DEBUG_POP): Remove. (NUM_REG_ITEMS): Remove. (NUM_NONREG_ITEMS): Adjust. (FAILURE_PAT, FAILURE_STR, NEXT_FAILURE_HANDLE, TOP_FAILURE_HANDLE): New macros for the cycle detection. (ENSURE_FAIL_STACK): New macro for PUSH_FAILURE_(REG|POINT). (PUSH_FAILURE_REG, POP_FAILURE_REG, CHECK_INFINITE_LOOP): New macros. (PUSH_FAILURE_POINT): Don't push registers any more. The pattern address pushed is not the destination of the jump but the source of it instead. (NUM_FAILURE_ITEMS): Remove. (POP_FAILURE_POINT): Adapt to the new stack structure (i.e. pop registers before the actual failure point). Don't hardcode any meaning for str==NULL anymore. (union register_info_type, REG_MATCH_NULL_STRING_P, IS_ACTIVE) (MATCHED_SOMETHING, EVER_MATCHED_SOMETHING, SET_REGS_MATCHED): Remove. (REG_UNSET_VALUE): Use NULL (why not?). (compile_range): Remove declaration since it doesn't exist. (struct compile_stack_elt_t): Remove inner_group_offset. (old_reg(start|end), reg_info, reg_dummy, reg_info_dummy): Remove. (regex_grow_registers): Remove dead code. (FIXUP_ALT_JUMP): New macro. (regex_compile): Add shy-groups Change loops to use on_failure_jump_smart&jump instead of on_failure_jump&maybe_pop_jump. Change + loops to eliminate the initial (dummy_failure_)jump. Remove c1_base (looks like unused variable to me). Use `jump' instead of `jump_past_alt' and don't bother with push_dummy_failure in alternatives since it is now unnecessary. Use FIXUP_ALT_JUMP. Eliminate a useless `#ifdef emacs' for (re)allocating the stack. (re_compile_fastmap): Remove dead variables i and num_regs. Exit from loop when bufp->can_be_null rather than jumping to `done'. Avoid jumping backwards so as to ensure termination. Use PATTERN_STACK_EMPTY and POP_PATTERN_OP. Improved handling of backreferences. Remove dead code in handling of `anychar'. (skip_noops, mutually_exclusive_p): New functions taken from the handling of `maybe_pop_jump' in re_match_2_internal. Slightly improve mutually_exclusive_p to handle ".+\n". ((lowest|highest)_active_reg, NO_(LOWEST|HIGHEST)_ACTIVE_REG) Remove. (re_match_2_internal): Use %p instead of 0x%x when printf'ing ptrs. Don't SET_REGS_MATCHED anymore. Remove many dead variables. Push register (in `start_memory') on the stack rather than storing it in old_reg(start|end). Remove the cycle detection from `stop_memory', replaced by the use of on_failure_jump_loop for greedy loops. Add code for the new on_failure_jump_<foo>. Remove ad-hoc code in `on_failure_jump' to push more registers in the case of a loop. Take out code from `maybe_pop_jump' into separate functions and adapt it to the semantics of `on_failure_jump_smart'. Remove jump_past_alt, dummy_failure_jump and push_dummy_failure. Remove dummy_failure handling and handling of `failures to jump to on_failure_jump' (this last one was already dead code, it seems). ((group|alt|common_op)_match_null_string_p): Remove.
Showing with 623 additions and 1219 deletions