[csw-maintainers] libiconv and --fast --xnolibmopt, ./configure hangs at "checking for stdbool.h that conforms to C99"

[Sebastian Kayser]

Hi,

i am seeing some weird issues while trying to build libiconv on
build8s. The regular build with GAR works just fine, but when i try
to incorporate the optimization flags that Phil has used so far
(--fast --xnolibmopt), the ./configure on build8s just hangs at

  checking for stdbool.h that conforms to C99...

GAR build description is at [1]. build8x is fine, i can only see
this issue on build8s. The ptree looks something like

        25961 gmake build-isa-sparcv8
          26068 gmake MODULATION=isa-sparcv8 ISA=sparcv8 build-modulated
            26886 /bin/sh -c cd work/build-isa-sparcv8/libiconv-1.13 && prefix="/opt/csw" exec_pr
              26887 /bin/bash ./configure --prefix=/opt/csw --exec_prefix=/opt/csw --bindir=/opt/cs
                29070 /bin/bash ./configure --prefix=/opt/csw --exec_prefix=/opt/csw --bindir=/opt/cs
                  29071 /opt/studio/SOS11/SUNWspro/bin/cc -c -fast -xnolibmopt -xarch=v8 -I/opt/csw/inc
                    29072 /opt/studio/SOS11/SUNWspro/prod/bin/acomp -xintrinsic -xldscope=global -depend

Trussing the acomp shows a myriad of brk()s and the process hogs
memory like there is no tomorrow.

   PID USERNAME  SIZE   RSS STATE  PRI NICE      TIME  CPU PROCESS/NLWP
 29072 skayser   684M  670M cpu13   30    0   0:01.23 3.1% acomp/1

When i take the conftest.c that ./configure tries to compile [2] and
try to compile it manually it simply bails out like it does on
build8x [3] (cannot find stdbool.h), but it does not hang. Anyone
seen something like this before? Or any other idea what could be the
culprit here?

Sebastian

[1] https://gar.svn.sf.net/svnroot/gar/csw/mgar/pkg/libiconv/trunk/Makefile
[2] http://pastebin.com/m6010e35c
[3] http://pastebin.com/m18121654

P.S.: Building with these optimization flags on build8x makes the
libiconv test suite complete in about 1:35 - 1:40 minutes versus 1:55
minutes for the non-optimized version, so it seems worth making the
opimization flags work.