BUG: Binutils strip corrupts dll files when using cygwin 1.5.18 and latest snapshot (20051103)

[Christopher Faylor]

On Fri, Nov 04, 2005 at 06:44:44AM -0700, Eric Blake wrote:
>According to Corinna Vinschen on 11/4/2005 6:11 AM:
>>The problem is that there's one section in the file, .gnu_debuglink,
>>which is defined as a debugging section.  strip's job is to remove all
>>debug sections.  This is usually a simple job to do since the debug
>>sections are typically at the end of the file anyway.  Not so the
>>.gnu_debuglink section.  By stripping this out, strip generates an
>>invalid PE/COFF file since it leaves an undefined memory hole in the
>>address space of the file, which is a no-no with the somewhat overly
>>simple Windows runtime loader.
>
>Wow.  Learn something every day.  No wonder ELF is so much more popular
>in the open source community.

Yeah, we tried hard to convince the open source community to use PE/COFF
and to start adopting the use of extensions like '.exe', and '.dll' but
you know how they are.

>>It's an unlucky case.  You could argue that strip should know better
>>than to remove debug sections from the middle of a PE/COFF file but
>>OTOH the Cygwin DLL is a very special case.
>
>So the cygwin DLL is likely to be the only PE/COFF dll that strip ever
>encounters with a section marked as debug in the middle of a file?

Yes.  Let me quote from the link that I provided since it seems like you
haven't read it:

>cgf wrote:
>>The splitting of the debug information into a new file is something that
>>I've wanted to do for a long time but problems with the standard
>>binutils way of accomplishing this task had proven to be daunting for PE
>>format executables.  I took some time today to try to understand what
>>was going wrong and wrote a script to fix up the dll so that information
>>about cygwin1.dbg was included in cygwin1.dll in a way that does not
>>cause cygwin1.dll to be an invalid binary -- which was the problem
>>before.
>>
>>Eventually, I will fix binutils so that this workaround is no longer
>>necessary and, then, it will be possible to create separate debug files
>>for any binary, without the need for my fixup.

So, to clarify, I was wrong to say that it wasn't a bug before.  It is
at least arguably a bug that binutils creates a bad dll if you strip
cygwin1.dll.  However, if you read the above it should be pretty clear
that this is a problem that is understood and that notes of concern
about the behavior are not required.

cgf

--
Unsubscribe info:      http://cygwin.com/ml/#unsubscribe-simple
Problem reports:       http://cygwin.com/problems.html
Documentation:         http://cygwin.com/docs.html
FAQ:                   http://cygwin.com/faq/