[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
[Emacs-diffs] /srv/bzr/emacs/trunk r108250: Try to fix building with gcc
From: |
Glenn Morris |
Subject: |
[Emacs-diffs] /srv/bzr/emacs/trunk r108250: Try to fix building with gcc >= 4.6 on Darwin. |
Date: |
Tue, 15 May 2012 23:15:52 -0700 |
User-agent: |
Bazaar (2.3.1) |
------------------------------------------------------------
revno: 108250
fixes bug(s): http://debbugs.gnu.org/9755
committer: Glenn Morris <address@hidden>
branch nick: trunk
timestamp: Tue 2012-05-15 23:15:52 -0700
message:
Try to fix building with gcc >= 4.6 on Darwin.
Eg, hydra builds have been failing for some time because (?) of this.
In gcc < 4.6, unrecognized -no-fo options just cause a warning.
In 4.6 and later, they cause an error.
* configure.in: (CPP): Do not unconditionally set it on Darwin.
Instead, try to test if -no-cpp-precomp is accepted.
(CPP_TEST_OPTIONS, SPECIFIED_CPP): New.
modified:
ChangeLog
configure.in
=== modified file 'ChangeLog'
--- a/ChangeLog 2012-05-15 19:13:17 +0000
+++ b/ChangeLog 2012-05-16 06:15:52 +0000
@@ -1,3 +1,10 @@
+2012-05-16 Glenn Morris <address@hidden>
+
+ * configure.in: Try to fix building with gcc >= 4.6 on Darwin.
+ (CPP): Do not unconditionally set it on Darwin.
+ Instead, try to test if -no-cpp-precomp is accepted.
+ (CPP_TEST_OPTIONS, SPECIFIED_CPP): New. (Bug#9755)
+
2012-05-15 Glenn Morris <address@hidden>
* Makefile.in (install-arch-dep): Replace use of MV_DIRS.
=== modified file 'configure.in'
--- a/configure.in 2012-05-12 18:10:37 +0000
+++ b/configure.in 2012-05-16 06:15:52 +0000
@@ -492,12 +492,11 @@
* ) unported=yes ;;
esac
opsys=darwin
- # Define CPP as follows to make autoconf work correctly.
- CPP="${CC-cc} -E -no-cpp-precomp"
# Use fink packages if available.
+ # FIXME? Is it really our business to decide this for the user?
if test -d /sw/include && test -d /sw/lib; then
GCC_TEST_OPTIONS="-I/sw/include -L/sw/lib"
- CPP="${CPP} ${GCC_TEST_OPTIONS}"
+ CPP_TEST_OPTIONS=${GCC_TEST_OPTIONS}
NON_GCC_TEST_OPTIONS=${GCC_TEST_OPTIONS}
fi
;;
@@ -581,9 +580,7 @@
machine=intel386
case "${canonical}" in
*-cygwin ) opsys=cygwin ;;
- *-darwin* ) opsys=darwin
- CPP="${CC-cc} -E -no-cpp-precomp"
- ;;
+ *-darwin* ) opsys=darwin ;;
*-sysv4.2uw* ) opsys=unixware; NON_GNU_CPP=/lib/cpp ;;
*-sysv5uw* ) opsys=unixware; NON_GNU_CPP=/lib/cpp ;;
*-sysv5OpenUNIX* ) opsys=unixware; NON_GNU_CPP=/lib/cpp ;;
@@ -633,13 +630,18 @@
# Save the value of CFLAGS that the user specified.
SPECIFIED_CFLAGS="$CFLAGS"
+# Save specified CPP (if any), before AC_PROG_CPP runs.
+SPECIFIED_CPP="$CPP"
+
dnl Sets GCC=yes if using gcc.
AC_PROG_CC
AM_PROG_CC_C_O
# Initialize gnulib right after verifying that the C compiler works.
+# Note: this ends up setting CPP.
gl_EARLY
+dnl Note: looks like gl_EARLY has already done this (via AC_PROG_CPP).
# On Suns, sometimes $CPP names a directory.
if test -n "$CPP" && test -d "$CPP"; then
CPP=
@@ -663,6 +665,46 @@
fi
fi
+## If using gcc, and on darwin, see if using Apple's gcc, where old
+## (pre gcc 3.3?) versions need -no-cpp-precomp to workaround some
+## (unrecorded) issue.
+## Frankly, I suspect this option is no longer needed on any system
+## still in use today. It is at best a no-op since Apple's version
+## of gcc 3.3 (?), which corresponds (?) to Xcode 1.5 from 2004.
+## This was for OS X 10.3, and the nextstep port (for example)
+## only supports 10.4 or later (although you could be building a
+## non-ns version on some older OS X I guess).
+if test x"$opsys" = xdarwin && test x"$GCC" = xyes; then
+ AC_MSG_CHECKING([whether we are using an Apple version of GCC])
+ AC_CACHE_VAL(emacs_cv_apple_gcc,
+ [AC_LINK_IFELSE([AC_LANG_PROGRAM([[]],
+[[#ifndef __APPLE_CC__
+fail;
+#endif
+]])], emacs_cv_apple_gcc=yes, emacs_cv_apple_gcc=no)])
+ AC_MSG_RESULT($emacs_cv_apple_gcc)
+
+ ## This only tests that adding the option does not cause an error.
+ ## FIXME It may still cause "unrecognized option"
+ ## (in gcc < 4.6 --no-foo does that; newer versions throw an error).
+ ## The message appears on stderr, so AC_EGREP_CPP is no use.
+ ## It would be better to test if it is actually needed.
+ ## For that, someone would have actually had to document what the
+ ## issues it worked around were, so we could test for them.
+ if test $emacs_cv_apple_gcc = yes; then
+ AC_MSG_CHECKING([whether preprocessor accepts -no-cpp-precomp])
+ save_CPP="$CPP"
+ test x$SPECIFIED_CPP = x && CPP="${CC-cc} -E"
+ CPP="$CPP -no-cpp-precomp"
+ dnl NB this will run AC_PROG_CPP first if it has not already happened.
+ AC_CACHE_VAL(emacs_cv_precomp,
+ [AC_PREPROC_IFELSE([AC_LANG_PROGRAM([[]],[[]])], emacs_cv_precomp=yes,
emacs_cv_precomp=no)])
+ AC_MSG_RESULT($emacs_cv_precomp)
+
+ test $emacs_cv_precomp = no && CPP="$save_CPP"
+ fi
+fi
+
#### Some systems specify a CPP to use unless we are using GCC.
#### Now that we know whether we are using GCC, we can decide whether
#### to use that one.
@@ -671,6 +713,8 @@
CPP="$NON_GNU_CPP"
fi
+test "x$CPP_TEST_OPTIONS" != x && CPP="$CPP $CPP_TEST_OPTIONS"
+
#### Some systems specify a CC to use unless we are using GCC.
#### Now that we know whether we are using GCC, we can decide whether
#### to use that one.
@@ -805,12 +849,16 @@
#### Some other nice autoconf tests.
-dnl checks for programs
+dnl Note: looks like gl_EARLY has already done this.
AC_PROG_CPP
AC_PROG_INSTALL
+
+dnl Note: looks like configure does this very early on anyway.
AC_PROG_MKDIR_P
+
+dnl Note: looks like gl_EARLY has already done this.
if test "x$RANLIB" = x; then
- AC_PROG_RANLIB
+ AC_PROG_RANLIB
fi
## Although we're running on an amd64 kernel, we're actually compiling for
[Prev in Thread] |
Current Thread |
[Next in Thread] |
- [Emacs-diffs] /srv/bzr/emacs/trunk r108250: Try to fix building with gcc >= 4.6 on Darwin.,
Glenn Morris <=