RFC: Document unexpected behaviour of --fatal-warnings

[H.J. Lu]

On Wed, Jan 24, 2024 at 8:06 AM H.J. Lu <hjl.tools@gmail.com> wrote:
>
> On Wed, Jan 24, 2024 at 7:15 AM Sam James <sam@gentoo.org> wrote:
> >
> >
> > Nick Clifton <nickc@redhat.com> writes:
> >
> > > Hi Guys,
> >
> > Hi Nick,
> >
> > >
> > >   It was recently pointed out to me that the bfd linker's
> > >   --fatal-warnings option can behave in an unexpected manner.  For
> > >   example:
> > >
> > >     $ ld.bfd -z bad-option --fatal-warnings -e 0/dev/null
> > >     ld.bfd: warning: -z bad-option ignored
> > >     $ echo $?
> > >     0
> > >
> > >   ie the warning about the ignored option is not being treated as an
> > >   error.  This happens because the --fatal-warnings option only takes
> > >   affect after it has been processed, and we process the options in a
> > >   linear order.  So the following works:
> > >
> > >     $ ld.bfd --fatal-warnings -z bad-option -e 0 /dev/null
> > >     ld.bfd: warning: -z bad-option ignored
> > >     $ echo $?
> > >     1
> > >
> > >   This behaviour differs from gold and lld, both of which honour
> > >   --fatal-warnings no matter where it occurs on the command line.
> > >
> > >   So we could fix the linker, create a two pass argument scan and the
> > >   problem would be solved.  But a) I am lazy and b) we already have a
> > >   precedent for options on the command line only affecting options that
> > >   come after it.  (I am thinking of the -L option here, although there
> > >   are probably several others).  So instead I am considering documenting
> > >   the current behaviour as expected.  (See the patch below).
> > >
> > >   What do people think ?
> >
> > I'm OK with it either way. We're really used to this kind of linker
> > behaviour so it's not a problem to just document it. I've not hit
> > this as a problem at all.
> >
> > Then if someone wants to improve it later on, so be it.
> >
> > (We could maybe change the docs to say "may not affect" if we want
> > leeway to change it later.)
>
> I opened:
>
> https://sourceware.org/bugzilla/show_bug.cgi?id=31289
>
> and will submit a fix this week.
>

A patch is posted at:

https://sourceware.org/pipermail/binutils/2024-January/132106.html

-- 
H.J.