mruby-regexp: relocate the lookaround offsets with the code they name - #7213

enum re_opcode gives four opcodes an absolute code index in offset: the lookarounds hold the end of their sub-pattern there, the same way RE_JMP, RE_SPLIT and RE_SPLITNG hold a jump target. Two places in re_compile.c move compiled code and have to carry those indices along, and each carried its own list of which opcodes to fix. Both lists named the three jumps and stopped.

The defect

A lookaround that goes through either relocator keeps an index into where its sub-pattern used to be. The stale index lands on the sub-pattern's own RE_MATCH, which ends the outer match at that point, so the pattern answers about a match it never made.

Three shapes reach a relocator:

• * or ? around a group, where insert_inst() puts a SPLIT in front of it and shifts the rest down
• {n,m}, where emit_atom_copy() copies the group and re-points the copy at itself
• alternation, where compile_alt() inserts a SPLIT at the first branch once every branch is compiled

The answers flip in both directions, and a MatchData survives the failure looking successful: the match reports a position while m[0] is nil. Each line is begin(0) and m[0], against CRuby 4.0.6:

/(?:(?=a)b)*x/.match("a")    # CRuby: nil      mruby: 0, nil
/(?:(?=a)a){2}/.match("aa")  # CRuby: 0, "aa"  mruby: nil
/(?=a)a|z/.match("ax")       # CRuby: 0, "a"   mruby: 0, nil
/(?:(?!b)b)*x/.match("ax")   # CRuby: 1, "x"   mruby: 0, nil
/(?:(?=a)ab)+/.match("ab")   # CRuby: 0, "ab"  mruby: 0, "ab"

The last line is the same group with nothing to relocate, and it has always agreed.

The fix

Ask one function, op_holds_code_index(), which opcodes hold a code index, rather than keeping the question answered separately in each relocator. The two lists disagreeing with the opcode set is the defect, and a list per relocator invites it back the next time an opcode gains an index.

RE_LB_WIDTH stays out, since it carries a character count in a, as do RE_SAVE and RE_BACKREF, whose offset is a slot number and a case-fold flag.

The ambiguity insert_inst() already resolves for a target equal to the insertion point needs no new arm. A lookaround's sub-pattern ends ahead of it, so it is always the forward-reference case, and leaving the end where it is puts the inserted instruction after the sub-pattern rather than inside it, which is what the quantifier wrapping the whole group is asking for.

Tests

mrbgems/mruby-regexp/test/regexp_syntax.rb gains one assertion block covering the three relocation shapes, one case each, plus the group with nothing to relocate as the control.

The asan build reports no AddressSanitizer or UBSan diagnostic.

Environment

# host (build_config/default.rb)
gcc -std=gnu99 -g -O3 -Wall -Wundef -Werror-implicit-function-declaration -Wwrite-strings -DMRBGEM_MRUBY_REGEXP_VERSION=0.0.0 -DHAVE_MRUBY_REGEXP_GEM -DMRB_USE_SET -DHAVE_MRUBY_IO_GEM -DMRB_USE_RATIONAL -DMRB_USE_COMPLEX -DMRB_USE_BIGINT -DMRB_USE_DEBUG_HOOK

# full-debug (build_config/ci/gcc-clang.rb)
gcc -std=gnu99 -g -O3 -Wall -Wundef -Werror-implicit-function-declaration -Wwrite-strings -g3 -O0 -DMRB_GC_STRESS -DMRB_USE_DEBUG_HOOK -DMRBGEM_MRUBY_REGEXP_VERSION=0.0.0 -DMRB_DEBUG -DMRB_USE_BIGINT -DMRB_USE_COMPLEX -DHAVE_MRUBY_ENCODING_GEM -DMRB_UTF8_STRING -DHAVE_MRUBY_IO_GEM -DMRB_USE_RATIONAL -DHAVE_MRUBY_REGEXP_GEM -DMRB_USE_SET -DMRB_USE_TASK_SCHEDULER

# bintest (build_config/ci/gcc-clang.rb)
gcc -std=gnu99 -g -O3 -Wall -Wundef -Werror-implicit-function-declaration -Wwrite-strings -DMRB_GC_FIXED_ARENA -DMRBGEM_MRUBY_REGEXP_VERSION=0.0.0 -DMRB_USE_BIGINT -DMRB_USE_COMPLEX -DHAVE_MRUBY_ENCODING_GEM -DMRB_UTF8_STRING -DHAVE_MRUBY_IO_GEM -DMRB_USE_RATIONAL -DHAVE_MRUBY_REGEXP_GEM -DMRB_USE_SET -DMRB_USE_TASK_SCHEDULER -DMRB_USE_DEBUG_HOOK

# cxx_abi (build_config/ci/gcc-clang.rb), compiled by gcc -x c++, g++ only links
gcc -g -O3 -Wall -Wundef -Wwrite-strings -x c++ -std=gnu++03 -DMRB_GC_FIXED_ARENA -DMRBGEM_MRUBY_REGEXP_VERSION=0.0.0 -DMRB_USE_CXX_EXCEPTION -DMRB_USE_CXX_ABI -DMRB_USE_BIGINT -DMRB_USE_COMPLEX -DHAVE_MRUBY_ENCODING_GEM -DMRB_UTF8_STRING -DHAVE_MRUBY_IO_GEM -DMRB_USE_RATIONAL -DHAVE_MRUBY_REGEXP_GEM -DMRB_USE_SET -DMRB_USE_TASK_SCHEDULER

# byte-string (build_config/ci/gcc-clang.rb)
gcc -std=gnu99 -g -O3 -Wall -Wundef -Werror-implicit-function-declaration -Wwrite-strings -DMRBGEM_MRUBY_REGEXP_VERSION=0.0.0 -DMRB_USE_BIGINT -DMRB_USE_COMPLEX -DHAVE_MRUBY_IO_GEM -DMRB_USE_RATIONAL -DHAVE_MRUBY_REGEXP_GEM -DMRB_USE_SET -DMRB_USE_TASK_SCHEDULER

# ascii-case (build_config/ci/gcc-clang.rb)
gcc -std=gnu99 -g -O3 -Wall -Wundef -Werror-implicit-function-declaration -Wwrite-strings -DMRB_USE_ASCII_CASE -DMRBGEM_MRUBY_REGEXP_VERSION=0.0.0 -DMRB_USE_BIGINT -DMRB_USE_COMPLEX -DHAVE_MRUBY_ENCODING_GEM -DMRB_UTF8_STRING -DHAVE_MRUBY_IO_GEM -DMRB_USE_RATIONAL -DHAVE_MRUBY_REGEXP_GEM -DMRB_USE_SET -DMRB_USE_TASK_SCHEDULER

# asan (build_config/asan.rb)
clang -std=gnu99 -g -O3 -Wall -Wundef -Werror-implicit-function-declaration -Wwrite-strings -Wzero-length-array -fsanitize=address,undefined -g3 -O0 -DMRBGEM_MRUBY_REGEXP_VERSION=0.0.0 -DMRB_DEBUG -DMRB_USE_BIGINT -DMRB_USE_COMPLEX -DHAVE_MRUBY_ENCODING_GEM -DMRB_UTF8_STRING -DHAVE_MRUBY_IO_GEM -DMRB_USE_RATIONAL -DHAVE_MRUBY_REGEXP_GEM -DMRB_USE_SET -DMRB_USE_TASK_SCHEDULER

Summary by CodeRabbit

• Bug Fixes

  • Improved regular expression handling for lookaround patterns, alternations, and repeated groups.
  • Fixed cases where inserting or copying pattern instructions could produce incorrect matches or match boundaries.

• Tests

  • Added regression coverage for lookaround endpoint preservation and related matching scenarios.