Re: [bug-diffutils] Bug: diff crashes on Windows when run from Clearcase

[Eric Blake]

[please keep the list in the loop]

On 11/10/2010 03:21 AM, address@hidden wrote:
>> In general, it is considered poor netiquette to send a 
>> screenshot when a
>> textual transcription will do.
> 
> Sorry for this. I know this rule from usegroups, but here we have a
> mailing list dedicated to but reports (i.e. one with low traffic), so
> I thought it is perfectly fine. Next time I will upload the screenshot
> to the web and just include the link (I think providing the screenshot
> might be safer than just reproducing the text, because you might then
> perhaps be able to see whether the message is generated by diff itself,
> or by Windows).
> 
>> In this case, your version of diff appears to have been built for
>> Windows.  Which distribution did you get it from?  Cygwin, mingw,
>> Interix, something else?  
> 
> It identifies itself by
> 
> diff (GNU diffutils) 2.8.7
> Written by Paul Eggert, Mike Haertel, David Hayes,
> Richard Stallman, and Len Tower.
> 
> Copyright (C) 2004 Free Software Foundation, Inc.
> 
> and it is part of the Gnu utilities for Windows.
> 
> See for example: http://gnuwin32.sourceforge.net/packages/diffutils.htm

This port does not have a very good track record of staying up-to-date.
 I can personally recommend the cygwin port as something that is more
likely to work in this scenario.

> 
>> It may be that your particular pre-built
>> binary is missing some prerequisite libraries, which would be an issue
>> for your distribution to resolve, rather than upstream diffutils.
> 
> Wouldn't this mean that it should always fail, and not only when stored in 
> a Clearcase dynamic view?

Not necessarily - it may just mean that the gnuwin32 port of GNU
utilities to Windows does not have all the workarounds built in like
other ports such as cygwin.  Therefore, your bug report is better
directed to the gnuwin32 folks.

> 
>> Also, the ClearCase MVFS file system is notoriously buggy on Windows.
> 
> Indeed, it is! Still, I wonder what this diff is doing different. For example,
> we have an older version of Gnu diff on Windows, which *does* work. The only
> reason why we don't use that old version, is that it contains a bug which is
> relevant to our work. That's why we switched to the new one.
> 
> BTW, the old and the new version differ also in that the new one uses a DLL
> and the old one is statically linked. I just mention this for completeness,
> because we *do* have applications using DLLs which work within Clearcase, so
> it is not that DLLs in general cause trouble.
> 
>> Otherwise, I would suggest raising a bug report with the
>> folks at IBM about their ClearCase product not obeying Windows file
>> system conventions, although given that ClearCase is not open 
>> source, I
>> can't promise any chance of a decent resolution with that approach.
> 
> Certainly something worth doing, though not very promising - I have yet to
> see a "quick fix" from them.
> 
> Wouldn't it make sense from your side too to find out, *which* system call
> fails? Maybe there is an easy workaround; and if it is not, at least the 
> incompatibility to Clearcase could be documented.

Not from my side - I intentionally stay as far away from ClearCase as
possible.

> 
> BTW, I forgot to mention: The problem appears on Windows 2000 and Windows XP.
> Don't have a chance to test other Win32 platforms.
> 
> Ronald

-- 
Eric Blake   address@hidden    +1-801-349-2682
Libvirt virtualization library http://libvirt.org

signature.asc
Description: OpenPGP digital signature