>Just for argument's sake, the problem with the "separate library"
>approach is that it gives an easy out for lazy programmers; "add
>-lgoddamned_stupid_broken_gets to the LDFLAGS line" will become
>part of the lore of how to port programs to NetBSD, and will almost
>certainly become built-in to GNU's configure script.  Having played
>Language Lawyer on comp.*.c for a long time, I'm all too familiar
>with the the class of programmers who literally don't care if something
>is correct as long as it "works" _once_ on their machine.  I actually
>think the ideal scheme would be to take the linker-warning approach,
>but have ld prompt for permission to continue rather than simply printing
>a diagnostic (legal by ANSI C, it's part of the implementation-defined
>diagnostic process).  Upon doing so, ld completes the link while shooting
>gouts of flame out the CRT screen, enveloping the developer (you have to
>finish the link, it's a conforming program, but again anything's fair
>in a diagnostic "message" if it's documented).  Of course, the additional
>hardware to support this feature might be a bit expensive.
>

warning everywhere is also not a good idea:
  1. people tend to ignor warnings
  2. it is silly if "that is what I want" 
  3. if the linker stops for permission to continue, it breaks the
     compilation stream.
  4. if the linker does not stop, nobody will see the message (it
     will fly off the screen.)
  
--Shao


------------------------------------------------------------------------------