RFA: dbxread patch

RFA: dbxread patch

[J. Johnston]

[-- Attachment #1: Type: text/plain, Size: 862 bytes --]

The following patch fixes a problem when the stabs debug info
for a file does not contain sline entries.  This happens, for
example with the glibc syscall code which is generated assembler
code.  In process_one_symbol in dbxread.c, a line 0 is recorded for the
end of the function.  When there are no other line numbers, we end up finding this
line 0 as our alt_symtab in symtab:find_pc_sect_line() and we end up
subtracting one resulting in line -1.  This problem manifests itself
when we attempt to break at the generated glibc functions such as
kill.

The patch changes process_one_symbol() to only record line 0 if sline info
has been found for the function.

Ok to commit?

-- Jeff J.


2003-02-24  Jeff Johnston  <jjohnstn@redhat.com>

	* dbxread.c (process_one_symbol): Only record line 0 if one or more sline
	entries have been seen for the function.




[-- Attachment #2: dbxread.patch --]
[-- Type: text/plain, Size: 532 bytes --]

Index: dbxread.c
===================================================================
RCS file: /cvs/src/src/gdb/dbxread.c,v
retrieving revision 1.43
diff -u -r1.43 dbxread.c
--- dbxread.c	20 Feb 2003 18:31:14 -0000	1.43
+++ dbxread.c	24 Feb 2003 23:39:52 -0000
@@ -2831,7 +2831,9 @@
  	      break;
  	    }
 
-	  record_line (current_subfile, 0, last_function_start + valu);
+	  if (sline_found_in_function)
+	    record_line (current_subfile, 0, last_function_start + valu);
+
 	  within_function = 0;
 	  new = pop_context ();
 

Re: RFA: dbxread patch

[Jim Blandy]

Yes, looks good.

Please add a comment briefly explaining why the check is needed
(hand-generated stabs that contain N_FUN stabs but no N_SLINE stabs).

Elena suggested the change, right?  She should be credited in the
ChangeLog entry.

"J. Johnston" <jjohnstn@redhat.com> writes:

> The following patch fixes a problem when the stabs debug info
> for a file does not contain sline entries.  This happens, for
> example with the glibc syscall code which is generated assembler
> code.  In process_one_symbol in dbxread.c, a line 0 is recorded for the
> end of the function.  When there are no other line numbers, we end up finding this
> line 0 as our alt_symtab in symtab:find_pc_sect_line() and we end up
> subtracting one resulting in line -1.  This problem manifests itself
> when we attempt to break at the generated glibc functions such as
> kill.
> 
> The patch changes process_one_symbol() to only record line 0 if sline info
> has been found for the function.
> 
> Ok to commit?
> 
> -- Jeff J.
> 
> 
> 2003-02-24  Jeff Johnston  <jjohnstn@redhat.com>
> 
> 	* dbxread.c (process_one_symbol): Only record line 0 if one or more sline
> 	entries have been seen for the function.
> 
> 
> 
> Index: dbxread.c
> ===================================================================
> RCS file: /cvs/src/src/gdb/dbxread.c,v
> retrieving revision 1.43
> diff -u -r1.43 dbxread.c
> --- dbxread.c	20 Feb 2003 18:31:14 -0000	1.43
> +++ dbxread.c	24 Feb 2003 23:39:52 -0000
> @@ -2831,7 +2831,9 @@
>   	      break;
>   	    }
>  
> -	  record_line (current_subfile, 0, last_function_start + valu);
> +	  if (sline_found_in_function)
> +	    record_line (current_subfile, 0, last_function_start + valu);
> +
>  	  within_function = 0;
>  	  new = pop_context ();
>  

Re: RFA: dbxread patch

[J. Johnston]

[-- Attachment #1: Type: text/plain, Size: 1386 bytes --]

Jim Blandy wrote:
> Yes, looks good.
> 
> Please add a comment briefly explaining why the check is needed
> (hand-generated stabs that contain N_FUN stabs but no N_SLINE stabs).
> 

I have attached a new patch with a comment.

> Elena suggested the change, right?  She should be credited in the
> ChangeLog entry.
> 

You are correct - thanks for spotting this oversight.

2003-02-25  Jeff Johnston  <jjohnstn@redhat.com>

	From Elena Zannoni  <ezannoni@redhat.com>
	* dbxread.c (process_one_symbol): Only record line 0 if one or more sline
	entries have been seen for the function.


Ok now?

-- Jeff J.

> "J. Johnston" <jjohnstn@redhat.com> writes:
> 
> 
>>The following patch fixes a problem when the stabs debug info
>>for a file does not contain sline entries.  This happens, for
>>example with the glibc syscall code which is generated assembler
>>code.  In process_one_symbol in dbxread.c, a line 0 is recorded for the
>>end of the function.  When there are no other line numbers, we end up finding this
>>line 0 as our alt_symtab in symtab:find_pc_sect_line() and we end up
>>subtracting one resulting in line -1.  This problem manifests itself
>>when we attempt to break at the generated glibc functions such as
>>kill.
>>
>>The patch changes process_one_symbol() to only record line 0 if sline info
>>has been found for the function.
>>
>>Ok to commit?
>>
>>-- Jeff J.
>>
>>

[-- Attachment #2: dbxread.patch --]
[-- Type: text/plain, Size: 754 bytes --]

Index: dbxread.c
===================================================================
RCS file: /cvs/src/src/gdb/dbxread.c,v
retrieving revision 1.43
diff -u -r1.43 dbxread.c
--- dbxread.c	20 Feb 2003 18:31:14 -0000	1.43
+++ dbxread.c	25 Feb 2003 23:12:07 -0000
@@ -2831,7 +2831,13 @@
  	      break;
  	    }
 
-	  record_line (current_subfile, 0, last_function_start + valu);
+	  /* The following check is added before recording line 0 at
+	     end of function so as to handle hand-generated stabs
+	     which may have an N_FUN stabs at the end of the function, but
+	     no N_SLINE stabs.  */
+	  if (sline_found_in_function)
+	    record_line (current_subfile, 0, last_function_start + valu);
+
 	  within_function = 0;
 	  new = pop_context ();
 

Re: RFA: dbxread patch

[Elena Zannoni]

J. Johnston writes:
 > Jim Blandy wrote:
 > > Yes, looks good.
 > > 
 > > Please add a comment briefly explaining why the check is needed
 > > (hand-generated stabs that contain N_FUN stabs but no N_SLINE stabs).
 > > 
 > 
 > I have attached a new patch with a comment.
 > 
 > > Elena suggested the change, right?  She should be credited in the
 > > ChangeLog entry.
 > > 
 > 
 > You are correct - thanks for spotting this oversight.
 > 
 > 2003-02-25  Jeff Johnston  <jjohnstn@redhat.com>
 > 
 > 	From Elena Zannoni  <ezannoni@redhat.com>
 > 	* dbxread.c (process_one_symbol): Only record line 0 if one or more sline
 > 	entries have been seen for the function.
 > 
 > 
 > Ok now?


Yes.
Elena


 > 
 > -- Jeff J.
 > 
 > > "J. Johnston" <jjohnstn@redhat.com> writes:
 > > 
 > > 
 > >>The following patch fixes a problem when the stabs debug info
 > >>for a file does not contain sline entries.  This happens, for
 > >>example with the glibc syscall code which is generated assembler
 > >>code.  In process_one_symbol in dbxread.c, a line 0 is recorded for the
 > >>end of the function.  When there are no other line numbers, we end up finding this
 > >>line 0 as our alt_symtab in symtab:find_pc_sect_line() and we end up
 > >>subtracting one resulting in line -1.  This problem manifests itself
 > >>when we attempt to break at the generated glibc functions such as
 > >>kill.
 > >>
 > >>The patch changes process_one_symbol() to only record line 0 if sline info
 > >>has been found for the function.
 > >>
 > >>Ok to commit?
 > >>
 > >>-- Jeff J.
 > >>
 > >>
 > Index: dbxread.c
 > ===================================================================
 > RCS file: /cvs/src/src/gdb/dbxread.c,v
 > retrieving revision 1.43
 > diff -u -r1.43 dbxread.c
 > --- dbxread.c	20 Feb 2003 18:31:14 -0000	1.43
 > +++ dbxread.c	25 Feb 2003 23:12:07 -0000
 > @@ -2831,7 +2831,13 @@
 >   	      break;
 >   	    }
 >  
 > -	  record_line (current_subfile, 0, last_function_start + valu);
 > +	  /* The following check is added before recording line 0 at
 > +	     end of function so as to handle hand-generated stabs
 > +	     which may have an N_FUN stabs at the end of the function, but
 > +	     no N_SLINE stabs.  */
 > +	  if (sline_found_in_function)
 > +	    record_line (current_subfile, 0, last_function_start + valu);
 > +
 >  	  within_function = 0;
 >  	  new = pop_context ();
 >  

Re: RFA: dbxread patch

[J. Johnston]

Patch applied.  Thanks.

Elena Zannoni wrote:
> J. Johnston writes:
>  > Jim Blandy wrote:
>  > > Yes, looks good.
>  > > 
>  > > Please add a comment briefly explaining why the check is needed
>  > > (hand-generated stabs that contain N_FUN stabs but no N_SLINE stabs).
>  > > 
>  > 
>  > I have attached a new patch with a comment.
>  > 
>  > > Elena suggested the change, right?  She should be credited in the
>  > > ChangeLog entry.
>  > > 
>  > 
>  > You are correct - thanks for spotting this oversight.
>  > 
>  > 2003-02-25  Jeff Johnston  <jjohnstn@redhat.com>
>  > 
>  > 	From Elena Zannoni  <ezannoni@redhat.com>
>  > 	* dbxread.c (process_one_symbol): Only record line 0 if one or more sline
>  > 	entries have been seen for the function.
>  > 
>  > 
>  > Ok now?
> 
> 
> Yes.
> Elena
> 
> 
>  > 
>  > -- Jeff J.
>  > 
>  > > "J. Johnston" <jjohnstn@redhat.com> writes:
>  > > 
>  > > 
>  > >>The following patch fixes a problem when the stabs debug info
>  > >>for a file does not contain sline entries.  This happens, for
>  > >>example with the glibc syscall code which is generated assembler
>  > >>code.  In process_one_symbol in dbxread.c, a line 0 is recorded for the
>  > >>end of the function.  When there are no other line numbers, we end up finding this
>  > >>line 0 as our alt_symtab in symtab:find_pc_sect_line() and we end up
>  > >>subtracting one resulting in line -1.  This problem manifests itself
>  > >>when we attempt to break at the generated glibc functions such as
>  > >>kill.
>  > >>
>  > >>The patch changes process_one_symbol() to only record line 0 if sline info
>  > >>has been found for the function.
>  > >>
>  > >>Ok to commit?
>  > >>
>  > >>-- Jeff J.
>  > >>
>  > >>
>  > Index: dbxread.c
>  > ===================================================================
>  > RCS file: /cvs/src/src/gdb/dbxread.c,v
>  > retrieving revision 1.43
>  > diff -u -r1.43 dbxread.c
>  > --- dbxread.c	20 Feb 2003 18:31:14 -0000	1.43
>  > +++ dbxread.c	25 Feb 2003 23:12:07 -0000
>  > @@ -2831,7 +2831,13 @@
>  >   	      break;
>  >   	    }
>  >  
>  > -	  record_line (current_subfile, 0, last_function_start + valu);
>  > +	  /* The following check is added before recording line 0 at
>  > +	     end of function so as to handle hand-generated stabs
>  > +	     which may have an N_FUN stabs at the end of the function, but
>  > +	     no N_SLINE stabs.  */
>  > +	  if (sline_found_in_function)
>  > +	    record_line (current_subfile, 0, last_function_start + valu);
>  > +
>  >  	  within_function = 0;
>  >  	  new = pop_context ();
>  >