From: Eric Wong <normalperson@...>
Date: 2013-11-27T10:29:53+00:00
Subject: [ruby-core:58638] Re: [ruby-trunk - Feature #8998] string keys for hash literals should use fstrings

"tmm1 (Aman Gupta)" <ruby@tmm1.net> wrote:
> Seeing some occasional segfaults in CI after this change. Not quite
> sure what's up yet.
>
> http://c5632.rubyci.org/~chkbuild/ruby-trunk/log/20131127T070302Z.log.html.gz
> 
> #3  0x0063fb97 in rb_bug (fmt=0x67363d "Segmentation fault") at error.c:341
> #4  0x00557a16 in sigsegv (sig=11, info=0x95f9c7c, ctx=0x95f9cfc)
>     at signal.c:699
> #5  <signal handler called>
> #6  search_nonascii (p=0x1c <Address 0x1c out of bounds>, 
>     len=<value optimized out>, enc=0x95df180) at string.c:202
> #7  coderange_scan (p=0x1c <Address 0x1c out of bounds>, 
>     len=<value optimized out>, enc=0x95df180) at string.c:232

So it seems GC is not preserving the hash used by rb_fstring, still.  I
wonder if something else changed since r43210 which caused this, perhaps
r43718?  I've never been a GC expert, especially not with the current GC.

> #8  0x0057ab6e in rb_enc_str_coderange (str1=184779220, str2=173130540)
>     at string.c:376