gmane.comp.debugging.dwarves

Re: .debug_types again

Arnaldo Carvalho de Melo — 2012-04-10T18:58:27

Em Mon, Apr 09, 2012 at 07:56:36AM -0600, Tom Tromey escreveu:

I saw them, was in a hurry working on slides for SF Collab Summit, will
process them shortly, thanks!

--
To unsubscribe from this list: send the line "unsubscribe dwarves" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA< at >public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

.debug_types again

Tom Tromey — 2012-04-09T13:56:36

I wrote some patches to change dwarves to support .debug_types (new from
DWARF 4).

I sent these as pull requests, but Mark W. asked me to post a note here
as well.

Here's the pull requests in question:

https://github.com/acmel/dwarves/pulls

Tom

--
To unsubscribe from this list: send the line "unsubscribe dwarves" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA< at >public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

failure with -gdwarf-4

Tom Tromey — 2011-07-01T18:40:04

pahole doesn't understand the new DWARF 4 .debug_types section.

I'm using the F15 GCC, but really any recent-enough GCC will do.
I compile this:

    struct s
    {
      int x;
      int y : 5;
      int z;
    };
    struct s i;

    int main()
    {
    }

Like so:

    gcc -gdwarf-4 -g -o d d.c

Running pahole gives an error:

    $ pahole --class_name=s ./d
    tag__recode_dwarf_type: couldn't find 0x1d type for 0x51 (variable)!

It would be nice to have support for this feature in dwarves, as
.debug_types can greatly reduce the size of the debuginfo in the final
executable.

Tom

--
To unsubscribe from this list: send the line "unsubscribe dwarves" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA< at >public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [patch] Don't die when encountering unknown tags, just warn

Arnaldo Carvalho de Melo — 2011-05-18T16:13:48

Em Wed, May 18, 2011 at 05:07:00PM +0200, Mark Wielaard escreveu:

Seems ok, I'll test it and merge if actual test matches visual
inspection.

Sorry for the delay in providing feedback, I broke a finger weeks ago
and was unable to type.

- Arnaldo
--
To unsubscribe from this list: send the line "unsubscribe dwarves" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA< at >public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [patch] Don't die when encountering unknown tags, just warn

Mark Wielaard — 2011-05-18T15:07:00

Hi,

I would still like to see this patch applied if possible.
Please let me know if this is the right way to fix the issue or whether
you like some other solution.

Thanks,

Mark

On Sat, 2011-03-26 at 22:23 +0100, Mark Wielaard wrote:


--
To unsubscribe from this list: send the line "unsubscribe dwarves" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA< at >public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[patch] Don't die when encountering unknown tags, just warn

Mark Wielaard — 2011-03-26T21:23:06

Hi,

I was playing with pahole against a binary compiled with an unreleased
gcc. This version outputs some DW_TAGs that pahole doesn't know about.
These tags have been added to elftuils git, but aren't yet in any
release.

The problem was that all errors from die__process_tag () are treated the
same (it returns NULL and then all callers interpret that as out of
memory). So this patch adds an explicit UNKNOWN_TAG return value so
callers can decide to ignore it and just carry on instead of terminating
the program where appropriate.

With this patch in place pahole will just print a warning when
encountering an unknown tag and ignore that die.

Cheers,

Mark

Re: [PATCH] pfunct: Introduce '-u'/'--all_units'

Rakesh Pandit — 2010-05-25T09:47:54

ping

Re: dwarf_loader.c: Fixed leaked bogon

Rakesh Pandit — 2010-05-10T11:03:26

Finally had a look at this case again and I am able to reproduce it.
Kernel kills the command:
[rakesh< at >simu build]$ ./pfunct -l -P
~/testrpms/webkitgtk-debuginfo-1.1.12-1.fc12.x86_64/usr/lib/debug/usr/lib64/libwebkit-1.0.so.2.debug
with message:
May 10 12:19:20 simu kernel: Out of memory: kill process 2956 (bash)
score 496818 or a child
May 10 12:19:20 simu kernel: Killed process 5758 (pfunct)

On 6 January 2010 22:14, Arnaldo Carvalho de Melo  wrote:

I am using -l ( along with -P & one more option also) among the above.
ELF file is webkitgtk-debuginfo-1.1.12-1.fc12.x86_64/usr/lib/debug/usr/lib64/libwebkit-1.0.so.2.debug
which is essentially a symlink to libwebkit-1.0.so.2.8.1.debug in same
debuginfo rpm.

Link to debuginfo rpm is:
http://kojipkgs.fedoraproject.org/packages/webkitgtk/1.1.12/1.fc12/x86_64/webkitgtk-debuginfo-1.1.12-1.fc12.x86_64.rpm

It is 290 MB ELF file.


Looking into where this leaks happens and what would be better way to
deal with it.

[PATCH] pfunct: Introduce '-u'/'--all_units'

Rakesh Pandit — 2010-05-06T10:27:29

Link: http://rakesh.fedorapeople.org/misc/0001-pfunct-Introduce-u-all_units.patch

Reason I am not reusing cus__for_each_cu which in turn uses
cu_function_iterator is because it would have in my view been more
unclean way of doing it. To keep things simple and clean I am rather
using a static function print_all_matches.

May this be reviewed ?

Regards,

Re: option for printing all matching function names

Rakesh Pandit — 2010-05-05T14:30:28

[..]

No problem, I will try to get some patches for review ready.


Exactly,

I would consider it as a two step problem. First add --all-units which
works as you specified in previous reply. Later on work on what you
suggested in here (on giving output somewhat more sensible C++ way),
following the very algo you mentioned.

I will first work on former first (--all-units) and then later move to C++ :)

Between thanks for your suggestions.

Regards,

Re: option for printing all matching function names

Arnaldo Carvalho de Melo — 2010-05-05T13:12:30

Em Wed, May 05, 2010 at 09:44:01AM -0300, Arnaldo Carvalho de Melo escreveu:

But that is not exactly what you want, as you more precisely described
here:

https://bugzilla.redhat.com/show_bug.cgi?id=495606

<quote>
Actually this does not results in correct match because corresponding
symbool name which we provide it with function arg is mangled. In case
we want to get a perfect matched we need to provide it de-mangled actual
function name.

But for C++ say (even in C) we can have more then one function with same
name.  In that case right now it will just print the first match.
</quote>

So what you want is that we traverse all the C++ classes finding which
ones have methods with that name, which is way more sensible than what
pfunct, being mostly oriented, so far, to C, does.

I don't have time right now to devote to this, but what you want is to:

1. go on CU by CU looking at all structs
2. look at each namespace in them looking for functions with the desired
   name, printing it in some C++ sensible way,

Re: option for printing all matching function names

Arnaldo Carvalho de Melo — 2010-05-05T12:44:01

Em Wed, May 05, 2010 at 05:15:20PM +0530, Rakesh Pandit escreveu:

well, if you say something like:

pfunct --function do_read --all-units

Meaning that it will print something like:

obj_foo: void do_read(char buffer);
obj_bar: int do_read(char bf, int size);
obj_baz: size_t do_read(int fd, char buf, int ln);

Its useful, send the patch :-)

- Arnaldo
--
To unsubscribe from this list: send the line "unsubscribe dwarves" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA< at >public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

option for printing all matching function names

Rakesh Pandit — 2010-05-05T11:45:20

Hello,

If we are using pfunct right with option --function, it prints the
first matched function name. This does not satisfy those cases where
object file has more then one functions with same name. I would like
it to print all matching cases not just the first one. Would it be
acceptable ? Or if it is not acceptable as it may break few scripts
written by users which are consuming the tool. An alternate
implementation could be if supplied with an extra option print all
matches and by default just print first one.

Suggestions ? If any of above to is acceptable I can straight right
down a patch accordingly.

Thanks,

Re: [PATCH] pfunct: Introduced --all/-A

Rakesh Pandit — 2010-05-05T03:49:43

[..]

Thanks for adding this option.

Re: [PATCH] pfunct: Introduced --all/-A

Arnaldo Carvalho de Melo — 2010-05-04T16:18:03

Em Tue, May 04, 2010 at 12:26:52PM +0530, Rakesh Pandit escreveu:

Isn't:

pfunct -V ~/bin/perf | grep \;$

equivalent?

Perhaps have a new flag to ask that no C comment like lines be issued?

so it would be:

pfunct -V --no-comments objfile

But ok, I'll add a '-P'/'--prototypes' option for that.

- Arnaldo
--
To unsubscribe from this list: send the line "unsubscribe dwarves" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA< at >public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH] pfunct: Introduced --all/-A

Rakesh Pandit — 2010-05-04T06:56:52

ping for review ?

Regards,

[PATCH] pfunct: Introduced --all/-A

Rakesh Pandit — 2010-04-05T07:31:58

Hello,

I have a request for enhancement (new option in pfunct) via which I
want pfunct to print signatures for all symbols in one go. To add to
it and as expected I want source code information per symbol if -l
option is specified.

Use can for me was: I wanted signature for all symbols from an ELF
file. Repeatedly asking pfunct for all symbols using names had a cost
for me. It used to take substantial time per symbol for uploading ELF
file into memory and searching for sym name and printing required
information. This option will upload it once and print all information
for me.

Can this be considered for inclusion ? If yes, may inline patch be reviewed ?

Thanks,

Re: dwarf_loader.c: Fixed leaked bogon

Rakesh Pandit — 2010-04-05T07:27:15

Sorry for delay in responding. Hardly had any time to look at this
one. Will reply in few days.

Re: [PATCH] scncopy: Fix minor problems with --usage.

Arnaldo Carvalho de Melo — 2010-01-12T18:38:02

Em Tue, Jan 12, 2010 at 01:17:18PM -0500, Peter Jones escreveu:

Please add a s-o-b line in the next patches.

Thanks, applied and pushed out!

- Arnaldo
--
To unsubscribe from this list: send the line "unsubscribe dwarves" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA< at >public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH] scncopy: Fix minor problems with --usage.

Peter Jones — 2010-01-12T18:17:18

- "-s section 0" doesn't really read the same as "-s section0"
- "--help" is something we should allow
- usage should say "scncopy" not "pjoc"
---
 scncopy.c |    6 ++++--
 1 files changed, 4 insertions(+), 2 deletions(-)

diff --git a/scncopy.c b/scncopy.c
index 38aaa9a..9acc16d 100644
--- a/scncopy.c
+++ b/scncopy.c
< at >< at > -65,8 +65,10 < at >< at > int main(int argc, char *argv[])
 n++;
 outfile = argv[n];
 continue;
-} else if (!strcmp(argv[n], "-?") || !strcmp(argv[n],"--usage")) {
-printf("usage: pjoc -s section 0 [[-s section1] ... -s sectionN] -o outfile infile\n");
+} else if (!strcmp(argv[n], "-?") ||
+!strcmp(argv[n], "--help") ||
+!strcmp(argv[n], "--usage")) {
+printf("usage: scncopy [-s section0 [[-s section1] ... -s sectionN] | -a ] -o outfile infile\n");
 return 0;
 } else if (n == argc-1) {
 infile = argv[n];

Re: [PATCH] scncopy: Fix a typo and add "-a" in --help.

Peter Jones — 2010-01-12T18:15:52

Er, yes ;)