Hi Marek,

> On 19 May 2023, at 00:32, Marek Polacek via Gcc-patches <gcc-patches@gcc.gnu.org> wrote:
> 
> On Tue, May 16, 2023 at 09:11:27PM +0100, Iain Sandoe wrote:
>> Hi Marek,
>> 
>>> On 16 May 2023, at 16:29, Marek Polacek via Gcc-patches <gcc-patches@gcc.gnu.org> wrote:
>>> 
>>> Ping.
>> 
>> I’m trying this on Darwin (since I have a local patch to do this for modern [darwin20+]
>> versions, which do not allow non-PIE)
> 
> Thanks, this patch is unfortunately more complicated than I expected,
> so any testing is very much appreciated.

Yeah, one reason my local patches have not been posted, it’s time-consuming work.

Unfortunately, we are not quite there yet...

On Darwin [Mach-o] PIC and PIE have different semantics from ELF.
  - PIE is applicable to linking executables
  - To link PIE, the input objects must be built PIC.
  - “-pie” is not applicable to linking other DSOs (dylibs, bundles, modules[notc++20 ones!])

so … I’m attaching a patch addition that deals with making host_pie imply host_shared
for Darwin, which means a top-level configure change to recognise —enable-host-pie there.

However - that does not fix a second issue that I am now getting diagnostics about “-pie
being passed to link lines for non-exec outputs (but maybe I can “fix” that in Darwin’s
specs).  It remains to be seen if other non-binutils linkers have similar issues.

Might I suggest that you also try to do a build of GCC with GMP and friends as sources
symlinked in-tree (I would expect that a similar situation would exist where they need to
be built PIC or PIE (if that is sufficient)).

Trying Solaris and AIX on the cfarm might also be a worthwhile diversion.

>> I think you are missing a hunk to deal with Ada.
> 
> You're right (and I missed Modula2 too).  I've fixed that problem in
> the ada/ and m2/ hunks.  CCing Eric and Gaius thus.  I've tested this
> patch with all languages, with/without --enable-host-pie, non/bootstrap,
> and all combinations on both Debian and Fedora.  Does this seem OK for
> trunk?
> 
> Thanks again for spotting the Ada problem; it would have been unpleasant
> to deal with it after pushing the patch.

I will re-combine and re-test.

Iain