From: Stuart Henderson <stu@spacehopper.org>
Subject: Re: make: refuse empty variable names
To: tech@openbsd.org
Date: Thu, 18 Jun 2026 16:09:23 +0100

This has been through mkr and a ports bulk build on i386 without problem.
It's ok with me.

On 2026/06/16 12:11, Marc Espie wrote:
> On Sun, Jun 14, 2026 at 11:01:47AM +0200, Marc Espie wrote:
> > Found by accident while looking at www/dufs/crates.inc
> > 
> > gmake refuses empty variable names, I think we should too.
> > 
> > Especially if we make typos and end up expanding empty stuff.
> > 
> > (currently, stuff that starts with an = will be treated as a comment)
> > 
> > This needs to get through a full release, obviously...
> 
> Of course first version was wrong. Namely the check needs to happen
> *after* the parser has determined we are in a variable assignment.
> 
> Index: parsevar.c
> ===================================================================
> RCS file: /home/espie/cvs/src/usr.bin/make/parsevar.c,v
> diff -u -p -r1.18 parsevar.c
> --- parsevar.c	18 Jun 2024 02:11:03 -0000	1.18
> +++ parsevar.c	16 Jun 2026 08:00:51 -0000
> @@ -133,6 +133,8 @@ parse_variable_assignment(const char *li
>  		}
>  	}
>  
> +	if (name.s == name.e)
> +		Parse_Error(PARSE_FATAL, "empty variable name");
>  	arg++;
>  	while (ISSPACE(*arg))
>  		arg++;
>