Re: [PATCH 2/2] MIPS: Fix n64 `ldconfig' for 32-bit objects

["Maciej W. Rozycki" <macro at codesourcery dot com>]

On Wed, 18 Mar 2009, Joseph S. Myers wrote:

> Thanks, applied, although using a union between the two header types might 
> be cleaner than casting a pointer to one to a pointer to the other.

 Thanks (for both).  I agree that a union might be cleaner iff both 
Elf32_Ehdr and Elf64_Ehdr were already used explicitly (rather than 
through ElfW()).  As it is, the use of lone Elf32_Ehdr would suffice, so a 
union would be all but obvious and I thought it might be worth expressing 
code like this so that such a mistake is not repeated if someone in the 
future refers to the header from the n64 path and misses the fact in a 
multilib Linux system any of the {o32,n32,n64} `ldconfig' binaries is used 
to maintain common "ld.so.cache".  With my version there is a chance they 
will ask themselves what this cast in the o32/n32 path is all about.

 And the correct behaviour of "ld.so.cache" is difficult to verify as 
AFAICS the path is hardcoded and you have to actually have a system (or 
chroot(2) jail) available to spare for a fresh libc installation and 
verification rather than just running the testsuite.  So it better be as 
prone to human errors as it can.  Hmm, perhaps I better expressed it with 
some comment in the code actually...

  Maciej