Re: bug#14911: automake-1.14 t/ccnoco-lt failure on OS X

[Stefano Lattarini]

tags 14911 + patch
severity 14911 minor
thanks

On 07/19/2013 10:24 PM, Nick Bowler wrote:
> On 2013-07-19 11:48 -0400, Hanspeter Niederstrasser wrote:
>> On OS X 10.7, t/ccnoco-lt fails with the following error (from 
>> test-suite.log).  For completeness sake, t/depcomp2 also failed, but I 
>> applied the patch from #14706 and that allowed depcomp2 to pass.
>>
>> ------- 8< ------- 8< ------- 8< -------
>>
>> FAIL: t/ccnoco-lt
>> =================
> [...]
>> libtool: link: 
>> /sw/build.build/automake1.14-1.14-1/automake-1.14/t/ccnoco-lt.dir/compile 
>> /sw/build.build/automake1.14-1.14-1/automake-1.14/t/ax/cc-no-c-o 
>> -dynamiclib -Wl,-undefined -Wl,dynamic_lookup -o .libs/libwish.0.dylib 
>> .libs/libwish.o   -L/sw/lib  -O2   -install_name 
>> /usr/local/lib/libwish.0.dylib -compatibility_version 1 -current_version 
>> 1.0 -Wl,-single_module
>> /sw/build.build/automake1.14-1.14-1/automake-1.14/t/ax/cc-no-c-o: both 
>> '-o' and '-c' seen on the command line
>> make: *** [libwish.la] Error 2
> 
> Looks like a false negative to me.
> 
> Automake uses a deliberately-crippled compiler wrapper which is designed
> to reject command-lines with -c and -o for testing purposes, found in
> automake/t/ax/cc-no-c-o.  The problem appears to be that the wrapper's
> test of the command line arguments is too simplistic, and incorrectly
> considers the above command line as a using both -c and -o, when it
> clearly does not.
> 
> So the test will need to be made more robust.
>
Agreed.  Could anybody test the patch below?

Thanks,
  Stefano

---- 8< --- 8< ---- 8< --- 8< ---- 8< --- 8< ---- 8< --- 8< ----

>From 4d7dcafc0f419378cd80e46f9390950c6fbaffa0 Mon Sep 17 00:00:00 2001
Message-Id: <address@hidden>
From: Stefano Lattarini <address@hidden>
Date: Sun, 21 Jul 2013 13:46:48 +0100
Subject: [PATCH] test: avoid false positives in 'cc-no-c-o' script

Fixes automake bug#14991.

* t/ax/cc-no-c-o.in: Be more careful in determining whether both the
'-c' and '-o' options have been passed on the command line to the
compiler.  In particular, do not spuriously complain in the face of
options like '-compatibility_version' or '-current_version' (seen on
Mac OS X 10.7).
* THANKS: Update.

Signed-off-by: Stefano Lattarini <address@hidden>
---
 THANKS            |  1 +
 t/ax/cc-no-c-o.in | 20 ++++++++++++++++----
 2 files changed, 17 insertions(+), 4 deletions(-)

diff --git a/THANKS b/THANKS
index b708943..1482da2 100644
--- a/THANKS
+++ b/THANKS
@@ -145,6 +145,7 @@ Gwenole Beauchesne              address@hidden
 H.J. Lu                         address@hidden
 H.Merijn Brand                  address@hidden
 Hans Ulrich Niedermann          address@hidden
+Hanspeter Niederstrasser        address@hidden
 Harald Dunkel                   address@hidden
 Harlan Stenn                    address@hidden
 He Li                           address@hidden
diff --git a/t/ax/cc-no-c-o.in b/t/ax/cc-no-c-o.in
index c18f9b9..bbc9ec9 100644
--- a/t/ax/cc-no-c-o.in
+++ b/t/ax/cc-no-c-o.in
@@ -19,11 +19,23 @@

 am_CC=${AM_TESTSUITE_GNU_CC-'@GNU_CC@'}

-case " $* " in
- *\ -c*\ -o* | *\ -o*\ -c*)
+seen_c=false
+seen_o=false
+
+for arg
+do
+  case $arg in
+    -c)
+      seen_c=true;;
+    # It is acceptable not to leave a space between the '-o' option
+    # and its argument, so we have to cater for that.
+    -o|-o*)
+      seen_o=true;;
+  esac
+  if $seen_c && $seen_o; then
     echo "$0: both '-o' and '-c' seen on the command line" >&2
     exit 2
-    ;;
-esac
+  fi
+done

 exec $am_CC "$@"
-- 
1.8.3.1.605.g85318f5