Redefining built-in commands.

Redefining built-in commands.

[Don Howard]

I've run across a bug involving redefinition of built-in commands:

If the command has an alias, the alias is not redefined. Instead, the
alias is left in an inconsistent state and causes a crash when invoked.

Should gdb allow users to redefine built-in commands?   If so, should the 
orignial alias continue to behave like the original built-in, or should 
it's behavior be modified also?

-- 
dhoward@redhat.com
gdb engineering

Re: Redefining built-in commands.

[Don Howard]

On Thu, 30 May 2002, Don Howard wrote:

> 
> 
> I've run across a bug involving redefinition of built-in commands:
> 
> If the command has an alias, the alias is not redefined. Instead, the
> alias is left in an inconsistent state and causes a crash when invoked.
> 
> Should gdb allow users to redefine built-in commands?   If so, should the 
> orignial alias continue to behave like the original built-in, or should 
> it's behavior be modified also?
> 


An example:

(gdb) define next
Really redefine built-in command "next"? (y or n) y
Type commands for definition of "next".
End with a line saying just "end".
>echo A silly example - redefine next\n
>end
(gdb) next
A silly example - redefine next
(gdb) n
Segmentation fault


-- 
dhoward@redhat.com
gdb engineering

Re: Redefining built-in commands.

[Eli Zaretskii]

> Date: Thu, 30 May 2002 17:24:05 -0700 (PDT)
> From: Don Howard <dhoward@redhat.com>
> 
> Should gdb allow users to redefine built-in commands?

I think it should, unless we cannot do it safely.

> If so, should the 
> orignial alias continue to behave like the original built-in, or should 
> it's behavior be modified also?

I'm not 100% sure, but I tend to think the aliases that are mere
abbreviations, like `n' for `next', should be redefined to the new
command, while aliases which aren't abbreviations should be left to
point to the original command.

But if the above cannot be done, we should either redefine the aliases
to the new command, or disable the aliases entirely.  Crashing is not
an option, obviously.

Re: Redefining built-in commands.

[Don Howard]

On Fri, 31 May 2002, Eli Zaretskii wrote:

> > Date: Thu, 30 May 2002 17:24:05 -0700 (PDT)
> > From: Don Howard <dhoward@redhat.com>
> > 
> > Should gdb allow users to redefine built-in commands?
> 
> I think it should, unless we cannot do it safely.
> 
> > If so, should the 
> > orignial alias continue to behave like the original built-in, or should 
> > it's behavior be modified also?
> 
> I'm not 100% sure, but I tend to think the aliases that are mere
> abbreviations, like `n' for `next', should be redefined to the new
> command, while aliases which aren't abbreviations should be left to
> point to the original command.
> 
> But if the above cannot be done, we should either redefine the aliases
> to the new command, or disable the aliases entirely.  Crashing is not
> an option, obviously.
> 


Ignoring the alias issue, I don't see any benefit to being able to
redefine built-in commands.

When a built-in command is redefined, it's original functionality is lost
for the remainder of that debugging session.  For example, say the user
redefines the 'step' command -- there is no longer a way to do
source-level stepping.  

(gdb) define step
> do something special
> step
> end
(gdb)

The above example ends up "doing something special" until gdb hits it's
recursion limit, rather than doing something special and then stepping.  
The example *does* work just fine if the user picks anything other than a
built-in name:

(gdb) define special_step
> do something special
> step
> end
(gdb)

I think the ability to redefine built-in commands is a misfeature.

Users can always choose an alternate name for their custom commands.



-- 
dhoward@redhat.com
gdb engineering

Re: Redefining built-in commands.

[Tom Tromey]

>>>>> "Don" == Don Howard <dhoward@redhat.com> writes:

Don> (gdb) define step
Don> do something special
Don> step
Don> end
Don> (gdb)

Don> The above example ends up "doing something special" until gdb hits it's
Don> recursion limit, rather than doing something special and then stepping.  
Don> The example *does* work just fine if the user picks anything other than a
Don> built-in name:

Bash solves this problem by adding a `command' command, which invokes
the original command.  So in the above you could use `command step' to
invoke the real step.  Perhaps this would be useful in gdb.

Tom