Bug #17732
Updated by nobu (Nobuyoshi Nakada) about 3 years ago
Reported by @yahonda in https://github.com/ruby/ruby/pull/4119#issuecomment-800189841 ### The bug `rb_enc_interned_str` doesn't properly handle autoloaded encodings that are not yet loaded: ``` [BUG] Segmentation fault at 0x0000000000000000 -- C level backtrace information ------------------------------------------- ruby(rb_print_backtrace+0xf) [0x101b06c92] vm_dump.c:758 ruby(rb_vm_bugreport) vm_dump.c:1042 ruby(rb_vm_bugreport) (null):0 ruby(bug_report_end+0x0) [0x101929f02] error.c:801 ruby(rb_bug_for_fatal_signal) error.c:801 ruby(sigsegv+0x5b) [0x101a6289b] signal.c:960 /usr/lib/system/libsystem_platform.dylib(_sigtramp+0x1a) [0x7fff71b64f5a] (null)((null)) (null):0 ruby(rb_enc_precise_mbclen+0x15) [0x101914dd5] encoding.c:1239 ruby(coderange_scan+0x63) [0x101a79773] string.c:602 ruby(rb_enc_str_coderange+0xd1) [0x101a79581] string.c:713 ruby(rb_str_hash+0x32) [0x101a78592] string.c:3290 ruby(do_hash+0x6) [0x101a6cd75] st.c:320 ruby(rb_st_update) st.c:1390 ruby(register_fstring+0x4c) [0x101a87cd5] string.c:398 ruby(rb_enc_interned_str) string.c:11502 ruby(ibf_load_object+0xa6) [0x1018fb176] compile.c:11816 ruby(ibf_load_object_regexp+0x129) [0x1018fba09] compile.c:11428 ruby(ibf_load_object+0xa6) [0x1018fb176] compile.c:11816 ruby(ibf_load_code+0x1000361bd) [0x1018dae7c] compile.c:10482 ruby(ibf_load_iseq_each) compile.c:11122 ruby(ISEQ_COMPILE_DATA_CLEAR+0x0) [0x1018dba8b] compile.c:11997 ruby(rb_ibf_load_iseq_complete) compile.c:11998 ruby(ibf_load_iseq) compile.c:12052 ruby(rb_iseq_ibf_load+0x4f) [0x1018db87f] compile.c:12158 ruby(iseqw_s_load_from_binary+0x12) [0x1019892d2] iseq.c:3430 ruby(vm_call_cfunc_with_frame+0x160) [0x101afc580] ./vm_insnhelper.c:2924 ruby(vm_sendish+0x572) [0x101af4e82] ruby(vm_exec_core+0x3606) [0x101ada706] insns.def:789 ruby(rb_vm_exec+0xafb) [0x101aef13b] vm.c:2162 ruby(rb_ec_exec_node+0x132) [0x1019354b2] eval.c:317 ruby(ruby_run_node+0x57) [0x101935327] eval.c:375 ruby(main+0x71) [0x10189a061] ./main.c:47 ``` Other `rb_enc_*` functions go through `enc_check_encoding()`, but because `rb_enc_interned_str` rely on `rb_setup_fake_str`, it bypass this check. ### Occurence While unlikely, this yhis crash can be caused by C extensions starting in ruby 3.0.0-p0. However https://github.com/ruby/ruby/pull/4119 made `RubyVM::InstructionSequence.load_from_binary` rely on `rb_enc_interned_str` and make this error very likely, mostly because `net/http` has a `Windows-31J` regexp (which is likely a bug too, see https://github.com/ruby/net-http/pull/18). So I believe this fix should be backported to the 3.0 branch. ### Patch I created a Pull Request with a patch: https://github.com/ruby/ruby/pull/4290