RFC: gas .warning "msg" and .error "msg" directives?

RFC: gas .warning "msg" and .error "msg" directives?

[Hans-Peter Nilsson]

How about .error and .warning directives, matching the following
texinfo snippet?  In contrast to .err or just using an undefined
pseudo, this allows the programmer to specify the output,
similar to #warn and #error in cpp.

+`.error STRING'
+     This directive emits an error with the message STRING.  If you
+     don't specify STRING, it defaults to `"(unspecified)"'.
+
+           .error "This code has not been assembled and tested."
+
+`.warning STRING'
+     Similar to the directive `.error', but just emits a warning.
+

The incredibly intricate code would look something like this
(snippets from currently port-local implementation provided for
discussion only, not indented for approval - though preapproval
would of course be welcome):

+  {"error", s_cris_errwarn, 1},
+  {"warning", s_cris_errwarn, 0},

+/* Emit an error or a warning.  */
+
+static void
+s_cris_errwarn (err)
+     int err;
+{
+  char *msg = _("(unspecified)");
+  int len;
+
+  if (!is_it_end_of_statement ())
+    {
+      if (*input_line_pointer != '\"')
+	{
+	  msg = err
+	    ? _(".error argument must be a string")
+	    : _(".warning argument must be a string");
+	  as_bad (msg);
+	  discard_rest_of_line ();
+	  return;
+	}
+
+      msg = demand_copy_C_string (&len);
+      if (msg == NULL)
+	return;
+    }
+
+  if (err)
+    as_bad (msg);
+  else
+    as_warn (msg);
+  demand_empty_rest_of_line ();
+}

If not, I'll pout and keep them local to the CRIS port.

brgds, H-P
PS.  I have a test-suite too, of course. ;-)

Re: RFC: gas .warning "msg" and .error "msg" directives?

[Ken Raeburn]

On Nov 2, 2004, at 22:01, Hans-Peter Nilsson wrote:
> How about .error and .warning directives, matching the following
> texinfo snippet?  In contrast to .err or just using an undefined
> pseudo, this allows the programmer to specify the output,
> similar to #warn and #error in cpp.
>
> +`.error STRING'
> +     This directive emits an error with the message STRING.  If you
> +     don't specify STRING, it defaults to `"(unspecified)"'.

Why not make it default to something that actually describes the 
situation, e.g., ".error directive invoked in source file".  
Conceptually, you could also make the message be required -- for .error 
that just means you print a slightly different message if it's omitted 
("missing .error parameter" for example), the practical difference is 
whether .warning without a message would be an error or not. :-)

Ken

Re: RFC: gas .warning "msg" and .error "msg" directives?

[Hans-Peter Nilsson]

> From: Ken Raeburn <raeburn@raeburn.org>
> Date: Wed, 3 Nov 2004 01:47:10 -0500

> Why not make it default to something that actually describes the 
> situation, e.g., ".error directive invoked in source file".  
> Conceptually, you could also make the message be required -- for .error 
> that just means you print a slightly different message if it's omitted 
> ("missing .error parameter" for example), the practical difference is 
> whether .warning without a message would be an error or not. :-)

All valid points.  Consider them merged with my suggestion, with:
An .error without a string should be compatible with ".err" and
just emit ".error directive invoked in source file".  A .warning
should just warn, with the default message being...you guessed
it ".warning directive invoked in source file".

brgds, H-P

Re: RFC: gas .warning "msg" and .error "msg" directives?

[Nick Clifton]

Hi Hans-Peter,

>>Why not make it default to something that actually describes the 
>>situation, e.g., ".error directive invoked in source file".  
>>Conceptually, you could also make the message be required -- for .error 
>>that just means you print a slightly different message if it's omitted 
>>("missing .error parameter" for example), the practical difference is 
>>whether .warning without a message would be an error or not. :-)
> 
> 
> All valid points.  Consider them merged with my suggestion, with:
> An .error without a string should be compatible with ".err" and
> just emit ".error directive invoked in source file".  A .warning
> should just warn, with the default message being...you guessed
> it ".warning directive invoked in source file".

Ok - if you would care to post a patch like that, along with a case or 
two for the testsuite then I would be happy to review it.

Cheers
   Nick