Re: cwrappers vs (old)wrappers: status and path forward?

[Jörg Sonnenberger <joerg%bec.de@localhost>]

On 8/24/24 7:44 PM, Tobias Nygren wrote:
> On Sat, 24 Aug 2024 13:04:24 -0400
> Greg Troxel <gdt%lexort.com@localhost> wrote:
>
> > I don't have the energy to work on this at all, but recent discussion
> > about optional rust reminded me that we are still perhaps partway
> > through a wrappers transition.
> >
> > We have wrappers from before, and cwrappers.  cwrappers has been default
> > for a really long time.  I wonder then:
> >
> >    - Do people believe that they are totally equivalent functionally?
> >      Or do we have "(old)wrappers is buggy in case X; fixed in
> >      cwrappers"?
> >
> >    - Do people believe that it makes sense to still have the old
> >      wrappers?  Are they for bootstrapping?  In case of bugs?   Not
> >      needed?  Something else?
> >
> >    - What do people think would happen if someone set USE_CWRAPPERS=no,
> >      ran into a problem, and reported it?  Anything other than being told
> >      they should use cwrappers?
>
> cwrappers lack support for coalescing arguments which
> is needed in some scenarios on legacy platforms.

Coalescing arguments by itself doesn't make sense and that's why it 
doesn't do anything like it.

> For example the HP-UX linker requires
> -Wl,R/usr/pkg -Wl,-R/usr/pkg/foo
> to be transformed into -Wl,+b,/usr/pkg:/usr/pkg/foo.
> At most one +b flag must appear on the command line.

...but that's because there is no support for the !ELF !OSX backend 
transforms at the moment. The only platform in that set that I ever 
cared about was AIX and even that was quite a while ago. Given that AIX 
has moved to faking a lot of ELF in later versions, I'm not even sure 
how much of the current wrapper transformations are still needed.

Joerg