[ruby-core:63723] [ruby-trunk - Bug #10037] Since r46798 on Solaris, "[BUG] rb_vm_get_cref: unreachable" during make

[normalperson@yhbt.net]

Issue #10037 has been updated by Eric Wong.


 ngotogenome@gmail.com wrote:
 > I think it is possbile that the first 8 byte (in 64-bit architecture)
 > of "struct rb_iseq_struct" is sometimes used for some other purposes
 > (maybe VALUE ?), and on big-endian 64-bit architecture (including
 > SPARC64), it is frequently misunderstood as the other type.
 
 That sounds strange.  I'll wait for ko1 to comment.
 
 I tried valgrind, but that did not report errors on x86-64 when I tried
 "make newline.c".  tool/transcode-tblgen.rb ran without errors.
 
 > By the way, I think "int stack_max" is better than "uint32_t
 > stack_max" because it seems that all calculations of stack_max in
 > compile.c(iseq_set_sequence) and vm_insnhelper.c(vm_push_frame) are
 > conducted using int.
 
 I agree, that may be changed to int.
 
 Since enum is also signed int type, maybe having 32-bit signed and
 unsigned types next to each other is confusing the compiler?
 
 Can you try "int stack_max"?  Thanks.

----------------------------------------
Bug #10037: Since r46798 on Solaris, "[BUG] rb_vm_get_cref: unreachable" during make
https://bugs.ruby-lang.org/issues/10037#change-47769

* Author: Naohisa Goto
* Status: Open
* Priority: Normal
* Assignee: Koichi Sasada
* Category: 
* Target version: 
* ruby -v: ruby 2.2.0dev (2014-07-13) [sparc64-solaris2.10]
* Backport: 2.0.0: UNKNOWN, 2.1: UNKNOWN
----------------------------------------
Since r46798, failed to build on Solaris 10 running on SPARC64.

~~~
(snip)
converter from WINDOWS-874 to UTF-8
converter from UTF-8 to WINDOWS-874
enc/trans/single_byte.trans:22: [BUG] rb_vm_get_cref: unreachable
ruby 2.2.0dev (2014-07-13) [sparc64-solaris2.10]

-- Control frame information -----------------------------------------------
c:0008 p:---- s:0056 e:000053 TOP    [FINISH]
c:0007 p:---- s:0052 e:000051 CFUNC  :require
c:0006 p:0015 s:0048 e:000047 METHOD enc/trans/single_byte.trans:22
c:0005 p:0244 s:0041 E:0011d8 EVAL   enc/trans/single_byte.trans:48 [FINISH]
c:0004 p:---- s:0037 e:000036 CFUNC  :eval
c:0003 p:0045 s:0030 e:000029 METHOD /XXXXXXXXXX/lib/erb.rb:850
c:0002 p:1228 s:0026 E:000cf8 EVAL   ./tool/transcode-tblgen.rb:1043 [FINISH]
c:0001 p:0000 s:0002 E:001d68 TOP    [FINISH]

-- Ruby level backtrace information ----------------------------------------
./tool/transcode-tblgen.rb:1043:in `<main>'
/XXXXXXXXXX/lib/erb.rb:850:in `result'
/XXXXXXXXXX/lib/erb.rb:850:in `eval'
enc/trans/single_byte.trans:48:in `<main>'
enc/trans/single_byte.trans:22:in `transcode_tblgen_singlebyte'
enc/trans/single_byte.trans:22:in `require'
/XXXXXXXXXX/enc/trans/windows-1250-tbl.rb:1:in `<top (required)>'

-- Other runtime information -----------------------------------------------

* Loaded script: ./tool/transcode-tblgen.rb

* Loaded features:

    0 enumerator.so
    1 /XXXXXXXXXX/lib/optparse.rb
    2 /XXXXXXXXXX/lib/cgi/util.rb
    3 /XXXXXXXXXX/lib/erb.rb
    4 /XXXXXXXXXX/lib/fileutils.rb
    5 /XXXXXXXXXX/lib/prettyprint.rb
    6 /XXXXXXXXXX/lib/pp.rb
    7 /XXXXXXXXXX/enc/trans/iso-8859-1-tbl.rb
    8 /XXXXXXXXXX/enc/trans/iso-8859-2-tbl.rb
    9 /XXXXXXXXXX/enc/trans/iso-8859-3-tbl.rb
   10 /XXXXXXXXXX/enc/trans/iso-8859-4-tbl.rb
   11 /XXXXXXXXXX/enc/trans/iso-8859-5-tbl.rb
   12 /XXXXXXXXXX/enc/trans/iso-8859-6-tbl.rb
   13 /XXXXXXXXXX/enc/trans/iso-8859-7-tbl.rb
   14 /XXXXXXXXXX/enc/trans/iso-8859-8-tbl.rb
   15 /XXXXXXXXXX/enc/trans/iso-8859-9-tbl.rb
   16 /XXXXXXXXXX/enc/trans/iso-8859-10-tbl.rb
   17 /XXXXXXXXXX/enc/trans/iso-8859-11-tbl.rb
   18 /XXXXXXXXXX/enc/trans/iso-8859-13-tbl.rb
   19 /XXXXXXXXXX/enc/trans/iso-8859-14-tbl.rb
   20 /XXXXXXXXXX/enc/trans/iso-8859-15-tbl.rb
   21 /XXXXXXXXXX/enc/trans/iso-8859-16-tbl.rb
   22 /XXXXXXXXXX/enc/trans/windows-874-tbl.rb

[NOTE]
You may have encountered a bug in the Ruby interpreter or extension libraries.
Bug reports are welcome.
For details: http://www.ruby-lang.org/bugreport.html

Abort
make[1]: *** [enc/trans/single_byte.c] Error 134
make[1]: Leaving directory `/XXXXXXXXXX'
make: *** [srcs-enc] Error 2

~~~

The following patch fixed the problem (tested with r46821), but this completely breaks the benefit of r46798.

~~~
Index: vm_core.h
===================================================================
--- vm_core.h   (revision 46821)
+++ vm_core.h   (working copy)
@@ -205,6 +205,7 @@
        ISEQ_TYPE_MAIN,
        ISEQ_TYPE_DEFINED_GUARD
     } type;              /* instruction sequence type */
+    uint32_t dummy;
     uint32_t stack_max; /* for stack overflow check */
 
     rb_iseq_location_t location;
~~~

I think it is possbile that the first 8 byte (in 64-bit architecture) of "struct rb_iseq_struct" is sometimes used for some other purposes (maybe VALUE ?), and on big-endian 64-bit architecture (including SPARC64), it is frequently misunderstood as the other type.

By the way, I think "int stack_max" is better than "uint32_t stack_max" because it seems that all calculations of stack_max in compile.c(iseq_set_sequence) and vm_insnhelper.c(vm_push_frame) are conducted using int.



-- 
https://bugs.ruby-lang.org/