Re: bfd/elf/mn10300 "dangerous error"

[Ian Lance Taylor <ian at airs dot com>]

DJ Delorie <dj@redhat.com> writes:

> > The notion of what dangerous means is target dependent.  Some
> > dangerous relocs should only be warnings.  Some should be fatal
> > errors.  I don't know which is the case for elf-m10300.c.
> 
> Yeah, I was hoping someone would know the history.  The relocs in
> question are non-pic relocs in a shared library, seemed kinda fatal to
> me.

Again, it depends.  You can put non-PIC code in a shared library, on
x86, for example.  The resulting code can't actually be shared, but
you might still want to use a shared library for name space control.
Or you might just be confused.

> Is there any historical precedent for having a linker (or emulation)
> option to control how fatal those cases should be?

There is --no-warn-mismatch and --warn-common, which are sort of
similar.  I can't think of anything else offhand.

I know!  We need a control variable for every error and warning
message, and then some way to manage the control variables from the
command line.  Isn't somebody doing that in gcc-land?

Ian