[Bug c/69291] New: [6 Regression] wrong code at -O1 for ruby-2.3.0/regcomp.c:985:compile_length_quantifier_node()

["t_nissie at yahoo dot co.jp" <gcc-bugzilla at gcc dot gnu dot org>]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=69291

            Bug ID: 69291
           Summary: [6 Regression] wrong code at -O1 for
                    ruby-2.3.0/regcomp.c:985:compile_length_quantifier_nod
                    e()
           Product: gcc
           Version: 6.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c
          Assignee: unassigned at gcc dot gnu.org
          Reporter: t_nissie at yahoo dot co.jp
  Target Milestone: ---

gcc-6 gives wrong code at -O1 for
ruby-2.3.0/regcomp.c:985:compile_length_quantifier_node()
and cannot build the ruby.  Even the -O1 optimization option fails.
Adding __attribute__((optimize("O0"))) before the function can evade this
issue.
I confirmed this issue on CentOS 6.7 (x86_64-pc-linux-gnu) and
Mac OS X 10.11.2 (x86_64-apple-darwin15.0.0).

However, I found strange behavior. If I built the ruby in /tmp on CentOS 6.7,
this issue does not stop to `make encdb.h`. But anyway, it stops to `make rdoc`
with Segmentation fault. Adding __attribute__((optimize("O0"))) can eliminate
the Segmentation fault, too.

Please note that there are two compile_length_quantifier_node() in
ruby-2.3.0/regcomp.c. gcc-6 fails to optimize second one at the line 985.

This may be a regression, because gcc-4.9.2 could build ruby-2.3.0.
clang of Mac OS X (Apple LLVM version 7.0.2 (clang-700.1.81)) could
build ruby-2.3.0, too.

================================
$ wget https://cache.ruby-lang.org/pub/ruby/2.3/ruby-2.3.0.tar.xz
$ shasum ruby-2.3.0.tar.xz | grep 96e620e38af351c8da63e40cfe217ec79f912ba1
$ tar xf ruby-2.3.0.tar.xz
$ mkdir  ruby-2.3.0/build_on_MacOSX
$ cd     ruby-2.3.0/build_on_MacOSX
$ /usr/local/bin/gcc --version
gcc (GCC) 6.0.0 20151115 (experimental)
$ ../configure CC=/usr/local/bin/gcc
$ make -j4
    :
generating encdb.h
../template/encdb.h.tmpl:63:in `block (4 levels) in <main>': x_emoji.h:8:
ENC_REPLICATE: UTF-8 is not defined yet. (replica UTF8-DoCoMo) (ArgumentError)
    :
$ sed -e '985 i\
> __attribute__((optimize("O0")))' ../regcomp.c > regcomp.c
$ diff -u ../regcomp.c .
--- ../regcomp.c        2015-12-09 16:30:44.000000000 +0900
+++ ./regcomp.c 2016-01-14 08:21:52.798816125 +0900
@@ -982,6 +982,7 @@
 #else /* USE_COMBINATION_EXPLOSION_CHECK */
 static int
+__attribute__((optimize("O0")))
 compile_length_quantifier_node(QtfrNode* qn, regex_t* reg)
 {
   int len, mod_tlen;
$ make V=1 encdb.h
    :
./miniruby -I../lib -I. -I.ext/common  ../tool/generic_erb.rb -c -o encdb.h
../template/encdb.h.tmpl ../enc enc
encdb.h updated
$ make -j4
$ make check
    :
# Running tests:

Finished tests in 453.556568s, 34.7388 tests/s, 4931.0101 assertions/s.
15756 tests, 2236492 assertions, 0 failures, 0 errors, 53 skips

ruby -v: ruby 2.3.0p0 (2015-12-25 revision 53290) [x86_64-darwin15]
check succeeded
$
================================

Tank you for your constant efforts on GCC,
Takeshi