Re: Handling corner case in building symbol table when "debug_line" includes compilation directory

[Maxim Grigoriev <maxim at tensilica dot com>]

I've attached the patch separately. Now, it should be fine. Sorry for 
inconvenience.

1) test case
extern int printf(const char *fmt,...);

int main()
{
printf("Hello, World !\n");
printf("Let's make DWARF consistent across all the tools !\n");
}

2) output from "readelf --debug-dump"

The section .debug_info contains:

Compilation Unit @ 0:
Length: 241
Version: 2
Abbrev Offset: 0
Pointer Size: 4
<0><b>: Abbrev Number: 1 (DW_TAG_compile_unit)
DW_AT_name : test.c
DW_AT_comp_dir : /home/maxim/W/BadgerPass/PR_14999
DW_AT_producer : xt-xcc for 7.1.2-development -O0 
-OPT:align_instructions=32 -g2
DW_AT_language : 1 (ANSI C)
DW_AT_identifier_case: 0 (case_sensitive)
DW_AT_stmt_list : 0
[ . . . . . ]
Dump of debug contents of section .debug_line:

Length: 101
DWARF Version: 2
Prologue Length: 65
[ . . . . .]

The Directory Table:
/home/maxim/W/BadgerPass/PR_14999

The File Name Table:
Entry Dir Time Size Name
1 1 1173897037 152 test.c

* * * * *

So the compilation directory is listed in the directory table. And, there is
a reference from the file table pointing to this directory.

Normally, we wouldn't see a compilation directory in a line table prologue.
With my native GCC on the same test case I have:

<0><b>: Abbrev Number: 1 (DW_TAG_compile_unit)
DW_AT_stmt_list : 0
DW_AT_high_pc : 0x80483b6
DW_AT_low_pc : 0x8048384
DW_AT_producer : GNU C 4.1.1 20060525 (Red Hat 4.1.1-1)
DW_AT_language : 1 (ANSI C)
DW_AT_name : test.c
DW_AT_comp_dir : /home/maxim/W/BadgerPass/PR_14999
[ . . . . ]
The Directory Table is empty.

The File Name Table:
Entry Dir Time Size Name
1 0 0 0 test.c

So the compilation directory gets always described and referenced 
explicitly.
Thanks for looking into this.

2007-03-15 Maxim Grigoriev <maxim2405@gmail.com>

* buildsym.c (start_subfile): Add handling missing case while
building symbol table for a compilation unit.

< Patch is attached >

Daniel Jacobowitz wrote:
> On Wed, Mar 14, 2007 at 06:54:57PM -0700, Maxim Grigoriev wrote:
>   
> > Hello All,
> >     
>
>   
> > Tensilica compiler seems to provide a unique approach in building DWARF
> > line tables. It confuses GDB algorithms being used to build symbol tables.
> >     
>
>   
> > First in DW_TAG_compile_unit entry
> >     
>
>   
> >   DW_AT_name        is a source file name as it was used in a command line;
> >   DW_AT_comp_dir    is a full path name to a compilation directory;    
> > Then in the statement program prologue, the directory table includes at
> > least one entry, which is again the compilation directory. And, there is
> > a corresponding file table entry referencing to this directory.
> >     
>
>   
> > Assuming that this combination of the names of the directories and files
> > does not contradict DWARF standard, I think GDB has to handle this situation.
> >     
>
>   
> > I suggest the following patch to fix this problem:
> >     
>
> First of all, your mailer ate whitespace in the patch again - very
> hard to read.
>
> Unfortunately, I do not understand your description.  Could you show
> me an example?
>
>   

Index: gdb/buildsym.c
===================================================================
RCS file: /cvs/src/src/gdb/buildsym.c,v
retrieving revision 1.47
diff -u -r1.47 buildsym.c
--- gdb/buildsym.c	27 Feb 2007 22:57:42 -0000	1.47
+++ gdb/buildsym.c	15 Mar 2007 01:14:14 -0000
@@ -549,7 +549,18 @@
 
   for (subfile = subfiles; subfile; subfile = subfile->next)
     {
-      if (FILENAME_CMP (subfile->name, name) == 0)
+      char *subfile_name;
+      if (IS_ABSOLUTE_PATH(name) && !IS_ABSOLUTE_PATH (subfile->name))
+	{
+	  subfile_name = concat (dirname, SLASH_STRING,
+				 subfile->name, (char *)NULL);
+	  make_cleanup (xfree, subfile_name);
+	}
+      else
+	{
+	  subfile_name = subfile->name;
+	}
+      if (FILENAME_CMP (subfile_name, name) == 0)
 	{
 	  current_subfile = subfile;
 	  return;