gdb_assert when resetting breakpoints

gdb_assert when resetting breakpoints

[Roland Schwingel]

Hi...

When using current gdb cvs head sources I encounter a gdb_assert() in 
breakpoint.c line 7451.
This is in function breakpoint_re_set_one().
Let me tell you what I am doing to get it (reproduceably). I am not a 
guru to gdb's internals so
please forgive me if I conclude something wrong.

I am having a mixed C and Objective C application. I am running on 
windows using GNUstep
for my ObjC Foundation classes. GNUstep-base is a shared library loaded 
lazy.
Due to the nature of this bug I assume that this will also appear on 
other platforms like linux
(according to my analysis). So it is not windows specific.

I start gdb and set a breakpoint to main() using "b main". Then I run my 
app using "r".
A few seconds later I get the assert in breakpoint.c:7451 
gdb_assert(sals.nelts==1).
When I debug into it sals.nelts == 2. The assert happens when the 
gnustep-base shared
library was loaded.

What has happenend: In breakpoint.c line 7417 you find:
sals = decode_line_1 (&s, 1, (struct symtab *) NULL, 0, (char ***) 
NULL,not_found_ptr);
s points to "main" at that moment as found in the breakpoint structure 
supplied to breakpoint_re_set_one().

sals (returned from decode_line_1) contains 2 entries.
Sals entry 0 points to the ObjectiveC METHOD main of the foundation base 
class NSThread
(written in ObjC notation -[NSThread main]). This is a real ObjC 
baseclass not something
from my own code. It has also nothing to do with an application's 
entrypoint. It also perfectly
legal in ObjectiveC to name methods like C functions. They do not collide.
Sals entry 1 points to the function "main()" of my application. This is 
the entry which I think is
the correct one to use here.

Obviously decode_line_1 returns every symbol matching a certain string 
independant of
its language. As far as I understand gdb at the moment this appears to 
be ok.

So what is the correct behaviour here to fix this problem? I assume that 
there might be more
areas of symbol duplicates that can occur not only main() vs -[NSThread 
main].

Maybe the assert should be replaced by a loop over the sals entries 
matching the infos from the
breakpoint structure passed to breakpoint_re_set_one() and the 
individual symtab entry of one
sals entry by filename(?) matching?

Thanks in advance for your help,

Roland

Re: gdb_assert when resetting breakpoints

[Paul Pluzhnikov]

On Tue, Mar 10, 2009 at 7:02 AM, Roland Schwingel
<roland.schwingel@onevision.de> wrote:

> When using current gdb cvs head sources I encounter a gdb_assert() in
> breakpoint.c line 7451.

How current? It appears that this may have been fixed yesterday :-)
Could you check if this problem is still present in current source?

Thanks,
-- 
Paul Pluzhnikov

Re: gdb_assert when resetting breakpoints

[Kai Tietz]

gdb-owner@sourceware.org wrote on 10.03.2009 15:02:40:

> Hi...
> 
> When using current gdb cvs head sources I encounter a gdb_assert() in 
> breakpoint.c line 7451.
> This is in function breakpoint_re_set_one().
> Let me tell you what I am doing to get it (reproduceably). I am not a 
> guru to gdb's internals so
> please forgive me if I conclude something wrong.
> 
> I am having a mixed C and Objective C application. I am running on 
> windows using GNUstep
> for my ObjC Foundation classes. GNUstep-base is a shared library loaded 
> lazy.
> Due to the nature of this bug I assume that this will also appear on 
> other platforms like linux
> (according to my analysis). So it is not windows specific.
> 
> I start gdb and set a breakpoint to main() using "b main". Then I run my 

> app using "r".
> A few seconds later I get the assert in breakpoint.c:7451 
> gdb_assert(sals.nelts==1).
> When I debug into it sals.nelts == 2. The assert happens when the 
> gnustep-base shared
> library was loaded.
> 
> What has happenend: In breakpoint.c line 7417 you find:
> sals = decode_line_1 (&s, 1, (struct symtab *) NULL, 0, (char ***) 
> NULL,not_found_ptr);
> s points to "main" at that moment as found in the breakpoint structure 
> supplied to breakpoint_re_set_one().
> 
> sals (returned from decode_line_1) contains 2 entries.
> Sals entry 0 points to the ObjectiveC METHOD main of the foundation base 

> class NSThread
> (written in ObjC notation -[NSThread main]). This is a real ObjC 
> baseclass not something
> from my own code. It has also nothing to do with an application's 
> entrypoint. It also perfectly
> legal in ObjectiveC to name methods like C functions. They do not 
collide.
> Sals entry 1 points to the function "main()" of my application. This is 
> the entry which I think is
> the correct one to use here.
> 
> Obviously decode_line_1 returns every symbol matching a certain string 
> independant of
> its language. As far as I understand gdb at the moment this appears to 
> be ok.
> 
> So what is the correct behaviour here to fix this problem? I assume that 

> there might be more
> areas of symbol duplicates that can occur not only main() vs -[NSThread 
> main].
> 
> Maybe the assert should be replaced by a loop over the sals entries 
> matching the infos from the
> breakpoint structure passed to breakpoint_re_set_one() and the 
> individual symtab entry of one
> sals entry by filename(?) matching?
> 
> Thanks in advance for your help,
> 
> Roland
> 
> 

Hi,

I prepared a patch for this issue, which should solve the described 
problem of Roland.

Changelog:

2009-03-17  Kai Tietz  <Kai.Tietz@onevision.com>

        * breakpoint.c: take care about multiple breakpoints
        possibilities when resetting breakpoints.


Is this patch ok for apply?

Cheers,
Kai

|  (\_/)  This is Bunny. Copy and paste Bunny
| (='.'=) into your signature to help him gain
| (")_(") world domination.

Patch:

Index: src/gdb/breakpoint.c
===================================================================
--- src.orig/gdb/breakpoint.c   2009-03-17 11:10:45.000000000 +0100
+++ src/gdb/breakpoint.c        2009-03-17 11:19:01.890880000 +0100
@@ -57,6 +57,7 @@
 #include "top.h"
 #include "wrapper.h"
 #include "valprint.h"
+#include "filenames.h"
 
 #include "mi/mi-common.h"
 
@@ -7392,6 +7393,7 @@
   struct breakpoint *b = (struct breakpoint *) bint;
   struct value *mark;
   int i;
+  int sals_index = -1;
   int not_found = 0;
   int *not_found_ptr = &not_found;
   struct symtabs_and_lines sals = {};
@@ -7455,20 +7457,59 @@
       if (not_found)
        break;
 
-      gdb_assert (sals.nelts == 1);
-      resolve_sal_pc (&sals.sals[0]);
-      if (b->condition_not_parsed && s && s[0])
+      /* If only one sals entry is found use it directly
+        elsewise cycle over all sals and match breakpoint
+        against all returned sals by checking filenames and
+        linenumbers. */
+      if (sals.nelts == 1)
+       sals_index = 0;
+      else
+       {
+         for (i=0;(i < sals.nelts) && (sals_index == -1);i++)
+           {
+             int symtab_linenum = sals.sals[i].line;
+             int bpoint_linenum = b->line_number;
+
+             if (symtab_linenum == bpoint_linenum)
+               {
+                 struct symtab *s = sals.sals[i].symtab;
+
+                 if (s)
+                   {
+                     const char *symtab_filename = s->filename;
+                     const char *bpoint_filename = b->source_file;
+
+                     if (symtab_filename)
+                       symtab_filename = lbasename(symtab_filename);
+
+                     if (bpoint_filename)
+                       bpoint_filename = lbasename(bpoint_filename);
+
+                     if (bpoint_filename && symtab_filename
+                         && 
(FILENAME_CMP(bpoint_filename,symtab_filename)
+                             == 0))
+                       sals_index = i;
+                   }
+               }
+           }
+       }
+      gdb_assert(sals_index != -1);
+      if (sals_index == -1)
+       break;
+
+      resolve_sal_pc (&sals.sals[sals_index]);
+      if (b->condition_not_parsed && s && s[sals_index])
        {
          char *cond_string = 0;
          int thread = -1;
-         find_condition_and_thread (s, sals.sals[0].pc, 
+         find_condition_and_thread (s, sals.sals[sals_index].pc, 
                                     &cond_string, &thread);
          if (cond_string)
            b->cond_string = cond_string;
          b->thread = thread;
          b->condition_not_parsed = 0;
        }
-      expanded = expand_line_sal_maybe (sals.sals[0]);
+      expanded = expand_line_sal_maybe (sals.sals[sals_index]);
       cleanups = make_cleanup (xfree, sals.sals);
       update_breakpoint_locations (b, expanded);
       do_cleanups (cleanups);

Re: gdb_assert when resetting breakpoints

[Kai Tietz]

Hello all,

any reply would be kind

Cheers,
Kai

2009/3/17 Kai Tietz <Kai.Tietz@onevision.com>:
> gdb-owner@sourceware.org wrote on 10.03.2009 15:02:40:
>
>> Hi...
>>
>> When using current gdb cvs head sources I encounter a gdb_assert() in
>> breakpoint.c line 7451.
>> This is in function breakpoint_re_set_one().
>> Let me tell you what I am doing to get it (reproduceably). I am not a
>> guru to gdb's internals so
>> please forgive me if I conclude something wrong.
>>
>> I am having a mixed C and Objective C application. I am running on
>> windows using GNUstep
>> for my ObjC Foundation classes. GNUstep-base is a shared library loaded
>> lazy.
>> Due to the nature of this bug I assume that this will also appear on
>> other platforms like linux
>> (according to my analysis). So it is not windows specific.
>>
>> I start gdb and set a breakpoint to main() using "b main". Then I run my
>
>> app using "r".
>> A few seconds later I get the assert in breakpoint.c:7451
>> gdb_assert(sals.nelts==1).
>> When I debug into it sals.nelts == 2. The assert happens when the
>> gnustep-base shared
>> library was loaded.
>>
>> What has happenend: In breakpoint.c line 7417 you find:
>> sals = decode_line_1 (&s, 1, (struct symtab *) NULL, 0, (char ***)
>> NULL,not_found_ptr);
>> s points to "main" at that moment as found in the breakpoint structure
>> supplied to breakpoint_re_set_one().
>>
>> sals (returned from decode_line_1) contains 2 entries.
>> Sals entry 0 points to the ObjectiveC METHOD main of the foundation base
>
>> class NSThread
>> (written in ObjC notation -[NSThread main]). This is a real ObjC
>> baseclass not something
>> from my own code. It has also nothing to do with an application's
>> entrypoint. It also perfectly
>> legal in ObjectiveC to name methods like C functions. They do not
> collide.
>> Sals entry 1 points to the function "main()" of my application. This is
>> the entry which I think is
>> the correct one to use here.
>>
>> Obviously decode_line_1 returns every symbol matching a certain string
>> independant of
>> its language. As far as I understand gdb at the moment this appears to
>> be ok.
>>
>> So what is the correct behaviour here to fix this problem? I assume that
>
>> there might be more
>> areas of symbol duplicates that can occur not only main() vs -[NSThread
>> main].
>>
>> Maybe the assert should be replaced by a loop over the sals entries
>> matching the infos from the
>> breakpoint structure passed to breakpoint_re_set_one() and the
>> individual symtab entry of one
>> sals entry by filename(?) matching?
>>
>> Thanks in advance for your help,
>>
>> Roland
>>
>>
>
> Hi,
>
> I prepared a patch for this issue, which should solve the described
> problem of Roland.
>
> Changelog:
>
> 2009-03-17  Kai Tietz  <Kai.Tietz@onevision.com>
>
>        * breakpoint.c: take care about multiple breakpoints
>        possibilities when resetting breakpoints.
>
>
> Is this patch ok for apply?
>
> Cheers,
> Kai
>
> |  (\_/)  This is Bunny. Copy and paste Bunny
> | (='.'=) into your signature to help him gain
> | (")_(") world domination.
>
> Patch:
>
> Index: src/gdb/breakpoint.c
> ===================================================================
> --- src.orig/gdb/breakpoint.c   2009-03-17 11:10:45.000000000 +0100
> +++ src/gdb/breakpoint.c        2009-03-17 11:19:01.890880000 +0100
> @@ -57,6 +57,7 @@
>  #include "top.h"
>  #include "wrapper.h"
>  #include "valprint.h"
> +#include "filenames.h"
>
>  #include "mi/mi-common.h"
>
> @@ -7392,6 +7393,7 @@
>   struct breakpoint *b = (struct breakpoint *) bint;
>   struct value *mark;
>   int i;
> +  int sals_index = -1;
>   int not_found = 0;
>   int *not_found_ptr = &not_found;
>   struct symtabs_and_lines sals = {};
> @@ -7455,20 +7457,59 @@
>       if (not_found)
>        break;
>
> -      gdb_assert (sals.nelts == 1);
> -      resolve_sal_pc (&sals.sals[0]);
> -      if (b->condition_not_parsed && s && s[0])
> +      /* If only one sals entry is found use it directly
> +        elsewise cycle over all sals and match breakpoint
> +        against all returned sals by checking filenames and
> +        linenumbers. */
> +      if (sals.nelts == 1)
> +       sals_index = 0;
> +      else
> +       {
> +         for (i=0;(i < sals.nelts) && (sals_index == -1);i++)
> +           {
> +             int symtab_linenum = sals.sals[i].line;
> +             int bpoint_linenum = b->line_number;
> +
> +             if (symtab_linenum == bpoint_linenum)
> +               {
> +                 struct symtab *s = sals.sals[i].symtab;
> +
> +                 if (s)
> +                   {
> +                     const char *symtab_filename = s->filename;
> +                     const char *bpoint_filename = b->source_file;
> +
> +                     if (symtab_filename)
> +                       symtab_filename = lbasename(symtab_filename);
> +
> +                     if (bpoint_filename)
> +                       bpoint_filename = lbasename(bpoint_filename);
> +
> +                     if (bpoint_filename && symtab_filename
> +                         &&
> (FILENAME_CMP(bpoint_filename,symtab_filename)
> +                             == 0))
> +                       sals_index = i;
> +                   }
> +               }
> +           }
> +       }
> +      gdb_assert(sals_index != -1);
> +      if (sals_index == -1)
> +       break;
> +
> +      resolve_sal_pc (&sals.sals[sals_index]);
> +      if (b->condition_not_parsed && s && s[sals_index])
>        {
>          char *cond_string = 0;
>          int thread = -1;
> -         find_condition_and_thread (s, sals.sals[0].pc,
> +         find_condition_and_thread (s, sals.sals[sals_index].pc,
>                                     &cond_string, &thread);
>          if (cond_string)
>            b->cond_string = cond_string;
>          b->thread = thread;
>          b->condition_not_parsed = 0;
>        }
> -      expanded = expand_line_sal_maybe (sals.sals[0]);
> +      expanded = expand_line_sal_maybe (sals.sals[sals_index]);
>       cleanups = make_cleanup (xfree, sals.sals);
>       update_breakpoint_locations (b, expanded);
>       do_cleanups (cleanups);
>



-- 
|  (\_/) This is Bunny. Copy and paste
| (='.'=) Bunny into your signature to help
| (")_(") him gain world domination

Re: gdb_assert when resetting breakpoints

[Joel Brobecker]

> >> What has happenend: In breakpoint.c line 7417 you find:
> >> sals = decode_line_1 (&s, 1, (struct symtab *) NULL, 0, (char ***)
> >> NULL,not_found_ptr);
> >> s points to "main" at that moment as found in the breakpoint structure
> >> supplied to breakpoint_re_set_one().
> >>
> >> sals (returned from decode_line_1) contains 2 entries.  Sals entry
> >> 0 points to the ObjectiveC METHOD main of the foundation base class
> >> NSThread (written in ObjC notation -[NSThread main]).

The problem, at this point, seems to be the fact that decode_line_1
should NOT be returning a match for this ObjC class method, since
the breakpoint language was C.  I think we need to find out why
this method was selected as a match, and prevent it from happening.

-- 
Joel

Re: gdb_assert when resetting breakpoints

[Roland Schwingel]

Hi Paul,

Thanks for your reply.

Paul Pluzhnikov <ppluzhnikov@google.com> wrote on 10.03.2009 16:39:09:
 > On Tue, Mar 10, 2009 at 7:02 AM, Roland Schwingel
 > <roland.schwingel@onevision.de> wrote:
 >
 > > When using current gdb cvs head sources I encounter a gdb_assert() in
 > > breakpoint.c line 7451.
 >
 > How current? It appears that this may have been fixed yesterday :-)
 > Could you check if this problem is still present in current source?
Just verfied it. I am using the most recent code from CVS.

In the meantime I have hacked myself the matching of sals entries
and infos from breakpoint. It appears to work (for me). I am
matching linenumber and filenames in case of more than one
sals hit. The question for me is: Is this a good solution? I would
prefer a fix from someone knowing gdb's code much better than
me at any time...

Roland

Re: gdb_assert when resetting breakpoints

[Roland Schwingel]

Hi Kai,

Kai Tietz wrote on 17.03.2009 11:24:25:
 > Roland Schwingel wrote on 10.03.2009 15:02:40:
 >
 > > Hi...
 > >
 > > When using current gdb cvs head sources I encounter a gdb_assert() in
 > > breakpoint.c line 7451.
 > > This is in function breakpoint_re_set_one().
 > >
 > > ...
 > >
 > > Thanks in advance for your help,
 > >
 > > Roland
 >
 > Hi,
 >
 > I prepared a patch for this issue, which should solve the described
 > problem of Roland.
 >
 > Changelog:
 >
 > 2009-03-17  Kai Tietz  <Kai.Tietz@onevision.com>
 >
 >         * breakpoint.c: take care about multiple breakpoints
 >         possibilities when resetting breakpoints.
 >
 >
 > Is this patch ok for apply?
 >
 > Cheers,
 > Kai

Thanks for your patch and your work. I just tried it and it works
without trouble and solves the problems in resetting the breakpoints for me.
It looks quite similar to what I did for myself. I couldn't find a 
problem upto
now with your patch. I think it would be could if this could make it to 
gdb's
mainline. Others might also profit from that.

Thank you very much,

Roland