rgarciasuarez hates software: Some fresh rpm hate

• From: Rafael Garcia-Suarez
  Date: 08:33 on 15 Mar 2006
  Subject: Some fresh rpm hate
  
  And now, some hate for developers. Sorry in advance for our
  non-C-fluent readers.
  
  Somewhere in the rpm library headers, up to version 4.4.4, you have
  this nice typedef:
  
  typedef void * (*rpmCallbackFunction)
                  (/*@null@*/ const void * h,
                  const rpmCallbackType what,
                  const unsigned long amount,
                  const unsigned long total,
                  /*@null@*/ fnpyKey key,
                  /*@null@*/ rpmCallbackData data)
  
  So far so good. Except that in 4.4.5 the "amount" and "total"
  arguments were changed to unsigned long long (without any mention in
  the release changelog, of course, that would be telling).
  
  So, if you have some C program that defines an rpmCallbackFunction,
  when it gets called, the call stack is blown, your program segfaults
  and gdb is confused. Hard to trace. Depending on the C application,
  this might also have unnice consequences on your system's rpm
  database, I suppose.
  
  But it doesn't stop there. I haven't been able to figure out a way to
  check for the version of the rpmlib you're linking against via plain
  #ifdefs. Which means that I have to fallback to some Makefile magic to
  add a CPP symbol that says whether I need an unsigned long or an
  unsigned long long here. Duh.
  
  

• From: Luke Kanies
  Date: 15:54 on 15 Mar 2006
  Subject: Re: Some fresh rpm hate
  
  On Wed, 15 Mar 2006, Rafael Garcia-Suarez wrote:
  
  > And now, some hate for developers. Sorry in advance for our
  > non-C-fluent readers.
  > [...]
  
  I'm pretty sure that's a C hate, not an RPM hate.  Yes, it's true, RPM does
  use C, so it's to blame, too, of course.
  
  
  

• From: Bruce Richardson
  Date: 22:20 on 15 Mar 2006
  Subject: Re: Some fresh rpm hate
  
  On Wed, Mar 15, 2006 at 09:54:23AM -0600, Luke Kanies wrote:
  > On Wed, 15 Mar 2006, Rafael Garcia-Suarez wrote:
  > 
  > > And now, some hate for developers. Sorry in advance for our
  > > non-C-fluent readers.
  > > [...]
  > 
  > I'm pretty sure that's a C hate, not an RPM hate.  Yes, it's true, RPM does
  > use C, so it's to blame, too, of course.
  
  No, I'm pretty sure it counts as an RPM hate.  C didn't force the
  RPM developers to change the definitions without telling anybody.  So we
  hate the software, because it represents the combined and accumulated
  stupidity/evil/carelessness of its developers.
  
  
  
  

• From: Peter da Silva
  Date: 23:55 on 15 Mar 2006
  Subject: Re: Some fresh rpm hate
  
  Definitely RPM hate. You don't change an interface incompatibly without 
  either changing a testable version number (so that old RPMs are 
  rejected), or providing an emulation of the old interface (so that old 
  RPMs work but don't get the benefit of the new stuff) without a very 
  good reason.
  
  This has nothing to do with C. It doesn't matter whether the interface 
  is a C subroutine call, a wire protocol, a system call, or the price of 
  a postage stamp.
  
  
  

Generated at 10:28 on 16 Apr 2008 by mariachi