[ruby-dev:8035] ruby 1.4.2 sometimes crashes on m68k-linux

Ruby の Debian パッケージについてバグ報告がきたんですが, 
Ruby 自体のバグではなかろうか? ということなんで
こちらにメールしてみます. 

内容としては以下に添付した通りです. 
今のところ m68k アーキテクチャについてのみの報告ですね. 

-- 

 やまだ あきら <URL:http://arika.org/>
 ([e-mail] or [e-mail])


Received: via dmail-4.1(9) for +test-mtx; Sat, 16 Oct 1999 22:41:16 +0900 (JST)
Return-Path: <[e-mail]>
Delivered-To: [e-mail]
Received: from master.debian.org (master.debian.org [209.41.108.5])
	by qp.linux.or.jp (Postfix) with SMTP id 817944581F
	for <[e-mail]>; Sat, 16 Oct 1999 22:41:15 +0900 (JST)
Received: (qmail 30153 invoked by uid 1409); 16 Oct 1999 13:33:10 -0000
Delivered-To: [e-mail]
Received: (qmail 30123 invoked by uid 1227); 16 Oct 1999 13:33:09 -0000
Subject: Bug#47559: ruby_1.4.2-3(unstable): crashes on m68k
Reply-To: [e-mail],
	[e-mail]
Resent-From: Roman Hodek <[e-mail]>
Resent-To: [e-mail]
Resent-Cc: akira yamada <[e-mail]>
Resent-Date: Sat, 16 Oct 1999 13:33:08 GMT
Resent-Message-ID: <[e-mail]>
Resent-Sender: [e-mail]
X-Debian-PR-Message: report 47559
X-Debian-PR-Package: ruby
X-Debian-PR-Keywords: 
X-Loop: [e-mail]
Received: via spool by [e-mail] id=B.94008078129947
          (code B ref -1); Sat, 16 Oct 1999 13:33:08 GMT
Date: Sat, 16 Oct 1999 15:32:56 +0200 (MET DST)
Message-Id: <[e-mail]>
From: Roman Hodek <[e-mail]>
To: [e-mail]

Package: ruby
Version: 1.4.2-3

The current ruby version sometimes crashes on m68k. Concretely, it
crashes during the ruby build, when extmak.rb builds the ext/
directory. It debugged it a bit because I suspected a m68k-specific
gcc bug, but the facts more point to a generic ruby bug.

Ok, some facts:

 - The crash happens when "miniruby -Xext extmk.rb" gets to the
   ext/sockets dir. The process receives a SIGILL during load of
   ext/socket/extconf.rb.

 - This happens only if extmk.rb compiles the *whole* ext dir. If one
   just retries ext/socket (i.e., after the crash type "make" again)
   it works.

 - I traced the crash a bit with gdb: At the end of ruby.c:load_file()
   rb_io_close(f) is called. The OpenFile f at this point is damanged,
   the component 'f' point somewhere into the data segment of libc,
   not to a FILE structure. This bogus pointer is later passed to
   fclose(), which crashes.

     #4  0x800229ee in rb_io_close (io=3222189632) at io.c:935
     935         rb_io_fptr_close(fptr);
     (gdb) p fptr
     $6 = (OpenFile *) 0x800d1aa0
     (gdb) p *fptr
     $7 = {f = 0xc00d5fac, f2 = 0x800baa28, mode = 1, pid = 0, lineno =
     	169,
       path = 0x0, finalize = 0}

For me this looks as ruby at some point and under some circumstances
overwrites its memory... An m68k gcc bug isn't impossible, but
unprobable. I'm sorry that I can't debug more on this due to lack of
knowledge about ruby internals and spare time...

I've archived the necessary files for debugging at

  ftp://ftp.uni-erlangen.de/pub/Linux/LOCAL/tmp/ruby-core.tar.gz

Roman