[Greg Noel <GregNoel@san.rr.com>] Re: [GregNoel@san.rr.com] libc/1326: Bug (or at least undocumented restriction) in DL routines

[Andreas Jaeger <aj at suse dot de>]

Hi,

Greg Noel asked me to forward this because his provider is in the
blacklist of open relays.

Geoff, I hope you're successful with mozilla.

Andreas

---

• To: Andreas Jaeger <aj at suse dot de>
• Subject: Re: [GregNoel@san.rr.com] libc/1326: Bug (or at leastundocumented restriction) in DL routines
• From: Greg Noel <GregNoel at san dot rr dot com>
• Date: Fri, 1 Oct 1999 03:24:40 -0700
• Cc: "GNU C. library Alpha Tester List" <libc-alpha at sourceware dot cygnus dot com>
• Delivery-date: Fri, 01 Oct 1999 12:15:20 +0200
• Envelope-to: aj@localhost

At 5:12 PM +0200 9/27/99, Andreas Jaeger wrote:
>we've received the appended bug report.  Is there really such a
>limitation in the PowerPC dl-routines of glibc?  Could anybody check
>(and fix;-) this, please?

I've got some additional information.

I've been in touch with Geoff Keating and run some tests for him.  There's
some good news and some bad news.

The good news is that there doesn't seem to be a limitation on either the
number of libraries or the number of symbols.

The bad news is that the failure occurs at different places if different
debugging output is requested.  Other tests where the only difference is
the background load also fail at different locations.  Running it under a
debugger also affects where it fails (since the debugger has to get the
symbols for the shared objects being loaded and therefore adds load).

So it looks like a timing bug.  It's likely that having debugging symbols
failed more quickly simply because processing the additional symbols
changed the load on the system.

I know that glibc-2.0 had some races; it could be that they haven't all
been eradicated for the PowerPC architecture yet.

For a timing bug, it's surprising how reproducible it is: with predictable
load, if fails in the same place virtually all the time.  This gives some
hope that it will be easy to find.

Geoff Keating is looking at it; I believe he's building Mozilla even as I type.

Thanks for your help,
-- Greg Noel, retired UNIX guru

P.S. A question for Herr Jaeger: I know it's an anagram, but was it
deliberate that SuSE chose a name whose anagram means "sweet"?

- jgn

---

-- 
 Andreas Jaeger   
  SuSE Labs aj@suse.de	
   private aj@arthur.rhein-neckar.de