OProfile

Email Archive: oprofile-list (read-only)

Re: oprofile 0.9 and g++ 2.95?

From: Jason Lunz <lunz@fa...> - 2005-06-02 20:47

levon@... said:
> I've applied a modified version of your patch. It'd be great if you
> could verify it's OK when it reaches anon cvs.

Any idea how long that'll take?

In the meantime, I noticed that debian woody has g++ 3.0.4. After
rerunning ./configure with CXX=g++-3.0, the vanilla oprofile 0.9 source
compiles perfectly. (again, i have no QT). So for me, at least, no
patching is needed at all.

So maybe you want to just recommend that alternative? I don't know who
else might need to support g++ that old. If you switch to requiring
g++ 3.0, you could also use string::clear() everywhere.

Jaso

Thread View

Thread	Author	Date
oprofile 0.9 and g++ 2.95?	Jason Lunz <lunz@fa...>

Re: oprofile 0.9 and g++ 2.95?

From: John Levon <levon@mo...> - 2005-06-02 17:16

On Thu, Jun 02, 2005 at 04:16:38PM +0000, Jason Lunz wrote:

> 0.9 isn't compiling though:

Sigh :)

> op_bfd.cpp:96: aggregate `struct stat st' has incomplete type and cannot be initialized

Try #include <sys/stat.h>

> op_bfd.cpp:265: implicit declaration of function `int remove_if(...)'

Try #include <algorithm>

> Is g++ 2.95.4 still supported by oprofile?

In theory yes, but I've not personally got it for testing. I probably
should have...

regards
john

Re: oprofile 0.9 and g++ 2.95?

From: Jason Lunz <lunz@fa...> - 2005-06-02 19:38

On Thu, Jun 02, 2005 at  6:16PM +0100, John Levon wrote:
> > Is g++ 2.95.4 still supported by oprofile?
> In theory yes, but I've not personally got it for testing. I probably
> should have...

The patch below gets 0.9 compiling on debian woody. I don't claim that
#define _GNU_SOURCE is necessarily the right thing to do; but it works.
It causes INFINITY to be defined with older glibc.

The patched source still compiles on debian sarge with g++ 3.3.5, though
I didn't attempt to compile any of the QT parts in either case.

Jason


diff -urp oprofile-0.9/libpp/diff_container.cpp oprofile-0.9.oldgcc/libpp/diff_container.cpp
--- oprofile-0.9/libpp/diff_container.cpp	Fri May 27 14:06:30 2005
+++ oprofile-0.9.oldgcc/libpp/diff_container.cpp	Thu Jun  2 15:17:51 2005
@@ -9,6 +9,7 @@
  * @author John Levon
  */
 
+#define _GNU_SOURCE
 #include "diff_container.h"
 
 #include <cmath>
diff -urp oprofile-0.9/libpp/format_output.cpp oprofile-0.9.oldgcc/libpp/format_output.cpp
--- oprofile-0.9/libpp/format_output.cpp	Mon Apr 11 23:09:35 2005
+++ oprofile-0.9.oldgcc/libpp/format_output.cpp	Thu Jun  2 15:18:00 2005
@@ -9,6 +9,7 @@
  * @author John Levon
  */
 
+#define _GNU_SOURCE
 #include <sstream>
 #include <iomanip>
 #include <iostream>
diff -urp oprofile-0.9/libutil++/bfd_support.cpp oprofile-0.9.oldgcc/libutil++/bfd_support.cpp
--- oprofile-0.9/libutil++/bfd_support.cpp	Thu May  5 10:43:46 2005
+++ oprofile-0.9.oldgcc/libutil++/bfd_support.cpp	Thu Jun  2 14:35:11 2005
@@ -18,6 +18,7 @@
 #include <iostream>
 #include <fstream>
 #include <sstream>
+#include <string>
 
 using namespace std;
 
@@ -548,7 +549,8 @@ find_nearest_line(bfd_info const & b, op
 
 fail:
 	info.found = false;
-	info.filename.clear();
+	// some stl lacks string::clear()
+	info.filename.erase(info.filename.begin(), info.filename.end());
 	info.line = 0;
 	return info;
 }
diff -urp oprofile-0.9/libutil++/op_bfd.cpp oprofile-0.9.oldgcc/libutil++/op_bfd.cpp
--- oprofile-0.9/libutil++/op_bfd.cpp	Thu May  5 10:43:46 2005
+++ oprofile-0.9.oldgcc/libutil++/op_bfd.cpp	Thu Jun  2 14:26:31 2005
@@ -14,9 +14,11 @@
 #include "config.h"
 
 #include <fcntl.h>
+#include <sys/stat.h>
 
 #include <cstdlib>
 
+#include <algorithm>
 #include <iostream>
 #include <iomanip>
 #include <sstream>

Re: oprofile 0.9 and g++ 2.95?

From: John Levon <levon@mo...> - 2005-06-02 19:46

On Thu, Jun 02, 2005 at 03:38:12PM -0400, Jason Lunz wrote:

> The patch below gets 0.9 compiling on debian woody. I don't claim that
> #define _GNU_SOURCE is necessarily the right thing to do; but it works.
> It causes INFINITY to be defined with older glibc.

Is there no way to get INFINITY defined without using _GNU_SOURCE on
these older glibc's ?

regards,
john

Re: oprofile 0.9 and g++ 2.95?

From: John Levon <levon@mo...> - 2005-06-02 20:25

On Thu, Jun 02, 2005 at 03:38:12PM -0400, Jason Lunz wrote:

> The patch below gets 0.9 compiling on debian woody. I don't claim that
> #define _GNU_SOURCE is necessarily the right thing to do; but it works.
> It causes INFINITY to be defined with older glibc.

I've applied a modified version of your patch. It'd be great if you
could verify it's OK when it reaches anon cvs.

thanks,
john

Re: oprofile 0.9 and g++ 2.95?

From: Jason Lunz <lunz@fa...> - 2005-06-02 20:35

levon@... said:
> Is there no way to get INFINITY defined without using _GNU_SOURCE on
> these older glibc's ?

I'm sure there is.  It's actually guarded by #define __USE_ISOC99 in math.h,
and features.h indicates that you can turn that on by #defining _ISOC99_SOURCE.
But #defining _only_ _ISOC99_SOURCE causes this:

	make[2]: Entering directory `/home/lunz/oprofile-0.9.oldgcc/libpp'
	source='diff_container.cpp' object='diff_container.o' libtool=no \
	depfile='.deps/diff_container.Po' tmpdepfile='.deps/diff_container.TPo' \
	depmode=gcc /bin/sh ../depcomp \
	g++ -DHAVE_CONFIG_H -I. -I. -I..  -I ../libop -I ../libutil -I ../libdb -I ../libopt++ -I ../libutil++ -I ../libop++ -I ../libregex  -W -Wall -fno-common -ftemplate-depth-50 -g -O2 -c -o diff_container.o `test -f 'diff_container.cpp' || echo './'`diff_container.cpp
	In file included from profile_container.h:20,
			from diff_container.h:15,
			from diff_container.cpp:13:
	../libutil++/op_bfd.h:224: syntax error before `('
	make[2]: *** [diff_container.o] Error 1
	make[2]: Leaving directory `/home/lunz/oprofile-0.9.oldgcc/libpp'

lines 219-224 of that file are:

        op_bfd_symbol const create_artificial_symbol();

        /* Generate symbols using bfd functions for
         * the image file associated with the ibfd arg.
         */
        uint process_symtab(bfd_info * bfd, uint start);

I guess the uint causes trouble? I didn't feel like tracking that down.

_GNU_SOURCE turns on a whole pile of options. The minimal subset that still
gets INFINITY defined and also compiles is to #define _ISOC99_SOURCE along with
either _BSD_SOURCE or _SVID_SOURCE.

There is also the compiler's -std option, but my attempts to use that haven't
worked.

-- 
Jason Lunz			Reflex Security
lunz@...		http://www.reflexsecurity.com/

Re: oprofile 0.9 and g++ 2.95?

From: Jason Lunz <lunz@fa...> - 2005-06-02 20:47

levon@... said:
> I've applied a modified version of your patch. It'd be great if you
> could verify it's OK when it reaches anon cvs.

Any idea how long that'll take?

In the meantime, I noticed that debian woody has g++ 3.0.4. After
rerunning ./configure with CXX=g++-3.0, the vanilla oprofile 0.9 source
compiles perfectly. (again, i have no QT). So for me, at least, no
patching is needed at all.

So maybe you want to just recommend that alternative? I don't know who
else might need to support g++ that old. If you switch to requiring
g++ 3.0, you could also use string::clear() everywhere.

Jaso

Re: oprofile 0.9 and g++ 2.95?

From: John Levon <levon@mo...> - 2005-06-02 21:13

On Thu, Jun 02, 2005 at 08:37:47PM +0000, Jason Lunz wrote:

> > could verify it's OK when it reaches anon cvs.
> 
> Any idea how long that'll take?

Nope. Just check the ChangeLog for your entry.

> In the meantime, I noticed that debian woody has g++ 3.0.4. After
> So maybe you want to just recommend that alternative? I don't know who

No, we support 2.2 kernels still (in theory) so such old systems won't
have more up to date stuff. Of course, it's doubtful there's actually
any users, but there's little cost right now to supporting the
configuration.

regards
john