On Mon, Feb 09, 2026 at 04:40:46PM +0100, Marc Espie wrote:
> Over the years, these lines have changed a bit, but the error message
> is still misleading
> 
> "null definition" would refer to what you put in the macro (IMO) whereas
> in actuality, it's the macro NAME that's empty.
> 
> Using two slightly different error messages will grow the executable slightly
> but it's still the right thing to do ?
> 
> Index: eval.c
> ===================================================================
> RCS file: /vide/cvs/src/usr.bin/m4/eval.c,v
> diff -u -p -r1.79 eval.c
> --- eval.c	26 Dec 2022 19:16:01 -0000	1.79
> +++ eval.c	9 Feb 2026 15:35:15 -0000
> @@ -606,7 +606,7 @@ void
>  dodefine(const char *name, const char *defn)
>  {
>  	if (!*name && !mimic_gnu)
> -		m4errx(1, "null definition.");
> +		m4errx(1, "attempting to define macro with empty name.");
>  	else
>  		macro_define(name, defn);
>  }
> @@ -643,7 +643,7 @@ static void
>  dopushdef(const char *name, const char *defn)
>  {
>  	if (!*name && !mimic_gnu)
> -		m4errx(1, "null definition.");
> +		m4errx(1, "attempting to pushdef macro with empty name.");
>  	else
>  		macro_pushdef(name, defn);
>  }

I was feeling whimsical, so I went all the way into NetBSD's tree, those
obscure error messages have been with us since at least 4.4-lite.

I have a suspicion the code was in serv.c in 386bsd-0.1, but somehow I
can't seem to find a valid copy of serv.c anywhere.