Re: remote/1832: spaces in directory names

Re: remote/1832: spaces in directory names

[Brendan Kehoe]

This change makes GDB a little more willing to debug an application on a
remote target when the file or its path happens to have one or more
spaces in it.  Among other things, this makes paths like
   C:/Documents and Settings/Joe Smith/My Documents/hack/mytest
usable when you want to debug mytest on a remote target via 'load'.

Two bits I'll try to bring up elsewhere but want to write down here:
  - I think it'd be better to pass the ARGS part of the on_load callback
function as an array of pointers similar to ARGV, so you don't have to
try to tokenize in places it might matter; and
  - I'm curious about the mystical second argument to the LOAD command
in GDB; it's only mentioned in the docs related to the Sparlet, and even
then only in passing.

Hope this helps,
B

2006-02-13  Brendan Kehoe  <brendan@zen.org>

    * symfile.c (generic_load): Adjust the handling of finding any
    second/offset argument so we can properly use filenames and paths
    that use spaces, like "/home/foo/my hacks/code".  Since the
    LOAD_OFFSET is already initialized to 0 at the beginning, we
    don't need an else stmt.


--- gdb/symfile.c.~1~    2005-08-31 22:07:33.000000000 +0100
+++ gdb/symfile.c    2006-02-13 17:21:16.000000000 +0000
@@ -1626,6 +1626,23 @@ generic_load (char *args, int from_tty)
   old_cleanups = make_cleanup (xfree, filename);
   strcpy (filename, args);
-  offptr = strchr (filename, ' ');
+  offptr = strrchr (filename, ' '); /* find the last space char */
   if (offptr != NULL)
+  {
+    /* Note where the space was, since that's effectively the end
+       of the filename if we have to insert the null char.
+       This implies we won't grok filenames that end in a space. :) */
+    char* end_of_filename = offptr++;
+
+    /* As per ISO 9899 (C99) $7.20.1.4/3 and POSIX 1003.1-2001,
+       strtoul expects a string meeting this sort of expression
+       for an integer-constant:
+          [\s]*[+\-]?(0[xX][0-9a-fA-F]*|[0-9]*)
+       This will be true when its third argument, BASE, is zero.
+       So having skipped the whitespace, we either have to have
+       a '+' or '-', or a valid digit character.  Anything else
+       has to be considered part of the filename.
+     */
+    if ((offptr[0] == '+' || offptr[0] == '-')
+    || (isdigit (offptr[0])))
     {
       char *endptr;
@@ -1634,8 +1651,7 @@ generic_load (char *args, int from_tty)
       if (offptr == endptr)
     error (_("Invalid download offset:%s."), offptr);
-      *offptr = '\0';
+      *end_of_filename = '\0';
     }
-  else
-    cbdata.load_offset = 0;
+  }
 
   /* Open the file for loading. */

Re: remote/1832: spaces in directory names

[Andrew STUBBS]

Brendan Kehoe wrote:
> This change makes GDB a little more willing to debug an application on a
> remote target when the file or its path happens to have one or more
> spaces in it.  Among other things, this makes paths like
>    C:/Documents and Settings/Joe Smith/My Documents/hack/mytest
> usable when you want to debug mytest on a remote target via 'load'.

I already have a patch for this awaiting approval.

See http://sources.redhat.com/ml/gdb-patches/2005-12/msg00127.html

Mine allows/required proper quoting and supports tilde expansion.

I also have two related patches waiting:

directory command patch:
http://sources.redhat.com/ml/gdb-patches/2005-12/msg00104.html

add-symbol-file command patch:
http://sources.redhat.com/ml/gdb-patches/2005-12/msg00106.html

Andrew Stubbs

Re: remote/1832: spaces in directory names

[Michael Snyder]

Brendan Kehoe wrote:

>   - I'm curious about the mystical second argument to the LOAD command
> in GDB; it's only mentioned in the docs related to the Sparlet, and even
> then only in passing.

It's ancient, and mostly not used any more.
One place that does still use it is monitor.c::monitor_load (q.v.)

The second argument was a base address or offset where
you wanted the load to be relocated.  This was basically
a hold-over from the days of a.out (in which the sections
were assumed to start at zero).

There was once a GNU extension called b.out, which was
a.out plus a section load offset.

Everything since coff has allowed an individual offset
per section to be specified.