Re: gdbinit.in

[david carlton <carlton@math.Stanford.EDU>]

On 26 Jul 2002 12:17:47 -0500, Jim Blandy <jimb@redhat.com> said:
> david carlton <carlton@math.stanford.edu> writes:

>> When debugging GDB, if I'm looking at a function in corefile.c, it
>> finds the one in bfd/corefile.c rather than the one in
>> gdb/corefile.c.

> The directory list maintained by the `dir' command only helps GDB
> find source files for code listings.

My apologies for being unclear;  that's what I was referring to.  (I
should know better than to submit such a vague bug report.)

If, from within the GDB source directory, I do

$ gdb gdb
(top-gdb) dir ./../bfd
(top-gdb) b read_memory
(top-gdb) run gdb

I get a bunch of output and then

Breakpoint 3, read_memory (memaddr=135224950, myaddr=0x828b92c "", len=1)
    at corefile.c:236
Line number 236 out of range; corefile.c has 105 lines.

Whereas if I do

$ gdb gdb
(top-gdb) dir .
(top-gdb) b read_memory
(top-gdb) run gdb

I get a bunch of output and then

Breakpoint 3, read_memory (memaddr=135224950, myaddr=0x828b92c "", len=1)
    at corefile.c:236
236	  status = target_read_memory (memaddr, myaddr, len);


Now I've investigated this some more, and I'm even more confused.

* Either a binary contains complete paths for its source files in its
  debugging information, or it doesn't.

* If it does, then using 'dir' is rarely likely to be a good idea.
  The binary contains the correct information; dir might tell GDB to
  mistakenly look in the wrong directory.  This occurs in the first
  example above.  (I'm not sure whether or not I think it's a good
  idea for GDB to look in location listed in the binary before
  searching the directories specified by 'dir', but that seems to me
  to be worth considering.  Given the info node (gdb)Source Path,
  though, I suspect that idea has been considered and rejected.)

* On my computer, it would seem that the debugging information _does_
  contain full pathnames.  So the best situation for me personally is
  to remove all of the dir commands from .gdbinit (and from
  gdbinit.in).

* If, on most people's computers, the debugging information does
  contain full pathnames, then I think we should remove all of the dir
  command from gdbinit.in.  (Patch below my signature.)  Otherwise, if
  configure can auto-detect whether the debugging information is rich
  enough, then the dir commands should only be included conditionally.

* In the unfortunate situation where debugging information doesn't
  contain full pathnames, the dir lines are useful.  In this case, the
  question is: if a file name occurs in both the GDB directory and
  another directory, are people debugging GDB more likely to encounter
  code in the file in the GDB directory or the other directory?  My
  guess is that the answer is "the GDB directory"; in that case, the
  command 'dir @srcdir@' should be moved further down in gdbinit.in,
  as my previous patch suggested.

* For what it's worth,

    for file in *.c *.h
      do ls ../{mmalloc,libiberty,bfd}/$file
    done 2>&1| grep -v 'No such file or directory'

  from within the GDB directory turns up

    ../bfd/corefile.c
    ../bfd/init.c
    ../bfd/config.h
    ../libiberty/config.h
    ../bfd/version.h

  So there aren't a whole lot of clashes.

* And yes, I do realize that, as bugs go, this isn't a particularly
  serious one...

David Carlton
carlton@math.stanford.edu

Index: gdbinit.in
===================================================================
RCS file: /cvs/src/src/gdb/gdbinit.in,v
retrieving revision 1.1.1.2
diff -c -p -r1.1.1.2 gdbinit.in
*** gdbinit.in	16 Aug 1999 19:52:40 -0000	1.1.1.2
--- gdbinit.in	26 Jul 2002 17:59:56 -0000
*************** commands
*** 10,18 ****
  	return
  end
  
- dir @srcdir@
- dir .
- dir @srcdir@/../mmalloc
- dir @srcdir@/../libiberty
- dir @srcdir@/../bfd
  set prompt (top-gdb) 
--- 10,13 ----

2002-07-26  david carlton  <carlton@math.stanford.edu>

	* gdbinit.in: Removed all dir commands.