[Boost-bugs] [Boost C++ Libraries] #12624: bug error calling boost::make_u32regex()

Subject: [Boost-bugs] [Boost C++ Libraries] #12624: bug error calling boost::make_u32regex()
From: Boost C++ Libraries (noreply_at_[hidden])
Date: 2016-11-25 05:22:18

#12624: bug error calling boost::make_u32regex()
---------------------------------------------+-------------------------
 Reporter:  shane.quinlivan@…                |      Owner:  johnmaddock
     Type:  Bugs                             |     Status:  new
Milestone:  To Be Determined                 |  Component:  regex
  Version:  Boost 1.63.0                     |   Severity:  Problem
 Keywords:  regex,solaris,bus error, SIGBUS  |
---------------------------------------------+-------------------------
 os: SunOS 5.11 11.2 sun4v sparc sun4v
 gcc:gcc (GCC) 4.8.2
 boost regex:tried 1.59->1_63+

 When calling boost::make_u32regex() with a regex
 like: '302-Found \\([0-9]+A[0-9]+\\)'

 The core occurs:

 #0  0x00142854 in boost::re_detail_106300::basic_regex_creator<int,
 boost::icu_regex_traits>::append_set (this=0xffbfd5b0, char_set=...) at
 ./boost/regex/v4/basic_regex_creator.hpp:380
 380        result->cclasses = char_set.classes();

 I added some debug for char_set, and it was fine. The issue is with
 result ( boost::re_detail_106300::re_set_long<unsigned long long> *);
 this structure has 2 long long variables (8byte): cclasses, cnclasses
 after 3 * 4byte variables (csingles,cranges,cequivalents).

 from boost/regex/v4/states.hpp, the structure is:

 203 /*** struct re_set_long
 ***********************************************
 204 A wide character set of characters, following this structure will be
 205 an array of type charT:
 206 First csingles null-terminated strings
 207 Then 2 * cranges NULL terminated strings
 208 Then cequivalents NULL terminated strings
 209
 ***********************************************************************/
 210 template <class mask_type>
 211 struct re_set_long : public re_syntax_base
 212 {
 213    unsigned int            csingles, cranges, cequivalents;
 214    mask_type               cclasses;
 215    mask_type               cnclasses;
 216    bool                    isnot;
 217    bool                    singleton;
 218 };

 When this struct definition is changed to force alignment:

 203 /*** struct re_set_long
 ***********************************************
 204 A wide character set of characters, following this structure will be
 205 an array of type charT:
 206 First csingles null-terminated strings
 207 Then 2 * cranges NULL terminated strings
 208 Then cequivalents NULL terminated strings
 209
 ***********************************************************************/
 210 template <class mask_type>
 211 struct __attribute__((__packed__)) re_set_long : public re_syntax_base
 212 {
 213    unsigned int            csingles, cranges, cequivalents;
 214    mask_type               cclasses;
 215    mask_type               cnclasses;
 216    bool                    isnot;
 217    bool                    singleton;
 218 };

 All our unit tests pass, and no core occurs.

 This change also resolves the solaris issue mentioned here:
 http://lists.boost.org/boost-users/2010/03/57717.php

-- 
Ticket URL: <https://svn.boost.org/trac/boost/ticket/12624>
Boost C++ Libraries <http://www.boost.org/>
Boost provides free peer-reviewed portable C++ source libraries.