IZ2635: Fails to build libs/uti/sge_getloadavg.c with gcc 4.3.1

[paulmillar]

[Imported from gridengine issuezilla ​http://gridengine.sunsource.net/issues/show_bug.cgi?id=2635]

        Issue #:      2635             Platform:     All       Reporter: paulmillar (paulmillar)
       Component:     gridengine          OS:        All
     Subcomponent:    build            Version:      current      CC:    None defined
        Status:       NEW              Priority:     P3
      Resolution:                     Issue type:    DEFECT
                                   Target milestone: ---
      Assigned to:    andreas (andreas)
      QA Contact:     andreas
          URL:
       * Summary:     Fails to build libs/uti/sge_getloadavg.c with gcc 4.3.1
   Status whiteboard:
      Attachments:
                      Date/filename:                           Description:                                           Submitted by:
                      Wed Jul 2 01:53:00 -0700 2008: 2635.diff A source diff that should fix the problem (text/plain) andreas
                      Wed Jul 2 01:59:00 -0700 2008: 2635.diff New try (former diff was buggy) (text/plain)           andreas

     Issue 2635 blocks:
   Votes for issue 2635:


   Opened: Tue Jul 1 16:36:00 -0700 2008 
------------------------


I tried to build gridengine from cvs on Debian sid.  The current HEAD failed
whilst building the libraries, in particular with libs/uti/sge_getloadavg.c.
The problem was with line 1306 and I've copied the output below:


_________C_O_R_E__S_Y_S_T_E_M_____________
gcc -O3 -Wall -Werror -Wstrict-prototypes -D__GRIDENGINE_FD_SETSIZE=8192 -DLINUX
-DLINUX86 -DLINUX86_26 -D_GNU_SOURCE -DGETHOSTBYNAME_R6 -DGETHOSTBYADDR_R8
-DLOAD_OPENSSL -I/vol2/SW/db-4.4.20/lx26-x86/include/ -DSGE_ARCH_STRING=lx26-x86
-DTARGET_32BIT  -DSPOOLING_dynamic -DSECURE
-I/vol2/tools/SW/openssl-0.9.8g-origin/lx26-x86/include -Wno-strict-aliasing
-D_FILE_OFFSET_BITS=64 -DCOMPILE_DC -D__SGE_COMPILE_WITH_GETTEXT__
-D__SGE_NO_USERMAPPING__ -I../common -I../libs -I../libs/uti -I../libs/juti
-I../libs/gdi -I../libs/japi -I../libs/sgeobj -I../libs/cull -I../libs/rmon
-I../libs/comm -I../libs/comm/lists -I../libs/sched -I../libs/evc -I../libs/evm
-I../libs/mir -I../libs/lck -I../daemons/common -I../daemons/qmaster
-I../daemons/execd -I../daemons/schedd -I../clients/common -I.
-I/usr/lib/jvm/java-6-sun/include -I/usr/lib/jvm/java-6-sun/include/linux  -fPIC
-c ../libs/uti/sge_getloadavg.c
cc1: warnings being treated as errors
../libs/uti/sge_getloadavg.c: In function 'get_cpu_load':
../libs/uti/sge_getloadavg.c:1306: error: array subscript is above array bounds
../libs/uti/sge_getloadavg.c:1306: error: array subscript is above array bounds
make: *** [sge_getloadavg.o] Error 1

I've not traced the logic of the function, but the code doesn't pass the "sniff
test".  I've copied a patch that fixes this issue, allowing the compilation to
progress, although it failed later on.


Index: libs/uti/sge_getloadavg.c
===================================================================
RCS file: /cvs/gridengine/source/libs/uti/sge_getloadavg.c,v
retrieving revision 1.38
diff -u -r1.38 sge_getloadavg.c
--- libs/uti/sge_getloadavg.c   15 Apr 2008 12:40:54 -0000      1.38
+++ libs/uti/sge_getloadavg.c   1 Jul 2008 23:19:22 -0000
@@ -1302,10 +1302,11 @@
    /* calculate percentages based on overall change, rounding up */
    half_total = total_change / 2l;
    for (i = 0; i < cnt; i++) {
-      *out = ((double)((*diffs++ * 1000 + half_total) / total_change))/10;
+      *out = ((double)((*diffs * 1000 + half_total) / total_change))/10;
       DPRINTF(("diffs: %lu half_total: %lu total_change: %lu -> %f",
             *diffs, half_total, total_change, *out));
       out++;
+      diffs++;
    }

    DEXIT;


Naturally, someone who understands the precise semantics of this function should
review the patch.

Cheers,

Paul.

PS. Can one attach patches to to bug with this issue-tracker?  I'd guess that
posting patches in-line is fragile.

   ------- Additional comments from andreas Wed Jul 2 01:53:01 -0700 2008 -------
Created an attachment (id=176)
A source diff that should fix the problem

   ------- Additional comments from andreas Wed Jul 2 01:59:21 -0700 2008 -------
Created an attachment (id=177)
New try (former diff was buggy)

   ------- Additional comments from andreas Wed Jul 2 02:02:08 -0700 2008 -------
Paul,

could you try the second diff that I attached to this issue and let me know the
result?

Note, the first one was buggy, since DPRINTF expressions are evaluated in
monitoring mode only. For that reason increments must be done outside the
DPRINTF statements.

Regards,
Andreas

   ------- Additional comments from paulmillar Wed Jul 2 14:48:01 -0700 2008 -------
Hi Andreas,

Thanks for looking into this.

Both patches look broken to me.  The first patch *only* increments the two ptrs
inside the DPRINTF, which (as you say) is broken if monitoring is switched off;
the second patch increments both inside and outside the DPRINT, which is broken
if monitoring is switched on!

Could you have another look at my patch?  I still believe this is the correct fix.

Cheers,

Paul.

PS.  Is it possible to use unified output for diffs ("cvs diff -u")?  I find
these easier to read.

   ------- Additional comments from paulmillar Mon Jul 14 17:06:04 -0700 2008 -------
Hi Andreas,

A couple of updates on this issue:

The first point is I've tried the second version of the patch, as you recommended.

At first it seemed to work; however, I was concerned that the compiler was
somehow factoring out the DPRINTF macro (hence the diffs++ and out++ within the
DPRINTF macro are never evaluated).  This would hide the problem until someone
attempts to compile with an enabled DPRINTF.  To test this, I replaced the
DPRINTF macro with a simple printf and the compilation broke again:

gcc -O3 -Wall -Werror -Wstrict-prototypes -D__GRIDENGINE_FD_SETSIZE=8192
[... many more arguments ...]
-I/usr/lib/jvm/java-6-sun-1.6.0.07/jre/include/linux  -fPIC -c
../libs/uti/sge_getloadavg.c
cc1: warnings being treated as errors
../libs/uti/sge_getloadavg.c: In function 'get_cpu_load':
../libs/uti/sge_getloadavg.c:1305: error: array subscript is above array bounds
../libs/uti/sge_getloadavg.c:1305: error: array subscript is above array bounds
../libs/uti/sge_getloadavg.c:1305: error: array subscript is above array bounds
../libs/uti/sge_getloadavg.c:1305: error: array subscript is above array bounds
make: *** [sge_getloadavg.o] Error 1

I believe this demonstrates that the second version (of the patch) is still
broken --- although I know why DPRINTF is not having any affect: is DPRINTF not
available for the uti library?

The second point is that I've just noticed that there's a function called
percentages_new() in the same file that is similar to percentages().

The patches so far only fix percentages() and not percentages_new().  The latter
looks to have the same problem as the former, but was not picked up by gcc as
the code is wrapped by some preprocessor tests for (I believe) compilation
architecture.

HTH,

Paul.

[dlove]

Seems to be fixed by [2819].