Re: hbreak.exp: Test hardware breakpoints

["Maciej W. Rozycki" <macro at mips dot com>]

On Sat, 29 Sep 2007, Daniel Jacobowitz wrote:

> Maciej posted a patch to test hardware breakpoints.  It passes on
> i386-linux, using current HEAD GDB.  I wanted to see what would happen
> when the test was run against gdbserver, which does not support
> hardware breakpoints on i386 (yet).  We encountered the long-standing
> bug that nm-i386.h overrides target.h using #define; this means that
> a native i386 GDB incorrectly tries to use ptrace even when connected
> to a remote target.
> 
> I saw two ways to test: build a cross debugger or fix the bug.  I
> fixed the bug.  Unfortunately that broke the test:
> 
> (gdb) hbreak main
> No hardware breakpoint support in the target.
> UNSUPPORTED: gdb.base/hbreak.exp: hardware breakpoints
> 
> This is "target exec", since we have not yet started the program.
> While we were using #define to override GDB's defaults, this worked;
> every target, even exec, claimed to support watchpoints.  Now that
> only "target child" supports hardware watchpoints, you can't insert
> hardware breakpoints before the program is running.
> 
> This is the status quo on other targets too.
> 
> How do you all feel about leaving the hbreak-before-run problem for
> another day, and adjusting the test to start the program running
> before trying hbreak?

 Have we reached any conclusion on this matter?  I cannot seem to find 
any...

 I am certainly fine to adjust the test script, at least for now -- no 
problem.  The question is how problematic, if at all, the change is to 
people in actual debugging.  For example, how often is a hardware 
breakpoint used locally before a program is started:

(gdb) hbreak foo
(gdb) run

How about:

(gdb) break main
(gdb) run
[main() hit]
(gdb) hbreak foo
(gdb) continue
[foo() hit]
(gdb) run

And similarly for remote targets -- presumably having to define a 
breakpoint before a connection is opened to the target is a non-issue, but 
how about:

(gdb) target bar
(gdb) load
(gdb) hbreak foo
(gdb) run
[foo() hit]
(gdb) target bar

Not that these cases bother me terribly much, but they should be noted and 
dismissed if not important enough.

  Maciej