Re: VAX Ultrix bootstrap failure with gcc-2.96

["Kaveh R. Ghazi" <ghazi at caip dot rutgers dot edu>]

 > From: Jeffrey A Law <law@cygnus.com>
 > 
 >   In message <200006051926.PAA00793@hiauly1.hia.nrc.ca>you write:
 >   > > There is an autoconf test for this misfeature.
 >   > > With that test added to configure, you can simply:
 >   > > 
 >   > > > +extern int                    getgroups(int, GETGROUPS_T *);
 >   > > 
 >   > 
 >   > I believe that the following patch tests for the misfeature as per
 >   > your suggestion.  I have checked that auto-hosts.h contains the
 >   > GETGROUPS_T define.  However, I haven't tested yet whether this
 >   > define gets propagated to the fixed sys/unistd.h under Ultrix.
 >   > Hopefully, it will.
 >   > 
 >   > Dave
 >   > -- 
 >   > J. David Anglin                                  dave.anglin@nrc.ca
 >   > National Research Council of Canada              (613) 990-0752 (FAX: 952-6
 >   > 605)
 >   > 
 >   > 2000-06-05  J. David Anglin  <dave@hiauly1.hia.nrc.ca>
 >   > 
 >   > 	* configure.in: Add AC_TYPE_GETGROUPS test.
 >   > 	* sys-protos.h: Use GETGROUPS_T * for second arg of getgroups.
 >   > 	* configure: Rebuilt.
 > 
 > Thanks.  Installed.
 > jeff


Jeff, two problems:

1.  You didn't regen config.in.

2.  Even with config.in regenerated, this patch doesn't work.

The definition of GETGROUPS_T does not get propagated to unistd.h on
sunos4.  I'd be surprised if it did on Ultrix.  Dave broke our testing
rule and flat out said so, thus IMHO it should not have been checked in.

Okay, spilled milk.  Nuff said. :-)


The problem lies in how xsys-protos.h is created and then used.  The
creation step is:

 > cat deduced.h ../../egcs-CVS20000613/gcc/sys-protos.h > tmp-fixtmp.c
 > mv tmp-fixtmp.c fixtmp.c
 > ./xgcc -B/usr/local/sparc-sun-sunos4.1.4/bin/ -B./ -isystem
 >        /usr/local/sparc-sun-sunos4.1.4/include fixtmp.c -w
 >        -U__SIZE_TYPE__ -U__PTRDIFF_TYPE__ -U__WCHAR_TYPE__ -E | sed -e
 >        's/ / /g' -e 's/ *(/ (/g' -e 's/ [ ]*/ /g' -e 's/( )/()/' |
 >        ./gen-protos >xsys-protos.hT
 > gen-protos: 1347 entries 340 collisions

The temporary fixtmp.c file created from deduced.h & sys-protos.h
never sees auto-host.h here so there's no opportunity to do the
GETGROUPS_T substitution.  IMO we rightfully don't include auto-host.h
in this step because of all the other gobbledygook auto-host.h does.
(E.g. think about the empty autoconf definition of "const" on K&R
hosts and how it will screw up all the prototypes.)  Thus I don't
recommend we simply cat auto-host.h in front of deduced.h.

Once xsys-protos.h is created, we're too late because its entry for
getgroups looks like this:

 >   {"getgroups", "int", "int, GETGROUPS_T *", 0},

Notice that GETGROUPS_T is inside a double quoted string, so cpp macro
substitution won't occur even though the use of xsys-protos.h in
fix-header.c does include auto-host.h.


Solution options:

1.  Handle GETGROUPS_T like e.g. __SIZE_TYPE__ and teach gcc about it
with a default of gid_t and an autoconf substitution in gcc source via
auto-host.h when necessary.  Then we can leave GETGROUPS_T in unistd.h
and we don't care cause it'll be substituted by gcc when compiling
user code just like __SIZE_TYPE__ is.

Probably overkill IMO.

2.  Somehow do the substituion in the xsys-protos.h creation pipeline.

E.g. a quick hack might be to change the first line of the creation
step to:

(grep GETGROUPS_T auto-host.h ; cat deduced.h ../../egcs-CVS20000613/gcc/sys-protos.h) > tmp-fixtmp.c

This would stick "#define GETGROUPS_T int" at the top of fixtmp.c and
it will DTRT when passed through gen-protos.  Maybe a cleaner more
elegant variation of this approach can be created.

3.  Something else?

		--Kaveh

PS: Dave would be please confirm whether the currently installed
GETGROUPS hackery works/fails for you on ultrix?  Remember to regen
config.in before testing.  Thanks.

--
Kaveh R. Ghazi			Engagement Manager / Project Services
ghazi@caip.rutgers.edu		Qwest Internet Solutions