setenv's broken case when value is nil

setenv's broken case when value is nil

[Noah Friedman]

lorentey@elte.hu changed env.el on 2007-08-29 to define setenv-internal and
for setenv always to keep the variable name in process-environment, even if
value is nil.  In other words, he removed the distinction between the empty
string and nil, and therefore you can't remove variables from the
environment using setenv anymore.

Since all kinds of environment variables have meaning when they are
defined, even if they are just set to the empty string, this is a screw.

Re: setenv's broken case when value is nil

[Noah Friedman]

>lorentey@elte.hu changed env.el on 2007-08-29 to define setenv-internal and
>for setenv always to keep the variable name in process-environment, even if
>value is nil.  In other words, he removed the distinction between the empty
>string and nil, and therefore you can't remove variables from the
>environment using setenv anymore.
>
>Since all kinds of environment variables have meaning when they are
>defined, even if they are just set to the empty string, this is a screw.

My proposed change is this simple:

--- /export/src/emacs/build/2010-02-03--06-16-34--99438/lisp/env.el.~1~ 2010-02-17 18:00:41.968005373 -0800
+++ /export/src/emacs/build/2010-02-03--06-16-34--99438/lisp/env.el 2010-02-17 18:00:41.968005373 -0800
@@ -178,7 +178,7 @@
   (if (string-equal "TZ" variable)
       (set-time-zone-rule value))
   (setq process-environment (setenv-internal process-environment
-                                             variable value t))
+                                             variable value))
   value)
 
 (defun getenv (variable &optional frame)

Re: setenv's broken case when value is nil

[Stefan Monnier]

>> lorentey@elte.hu changed env.el on 2007-08-29 to define setenv-internal and
>> for setenv always to keep the variable name in process-environment, even if
>> value is nil.  In other words, he removed the distinction between the empty
>> string and nil, and therefore you can't remove variables from the
>> environment using setenv anymore.

That would be a bug, so bug-gnu-emacs would have been a better place to
send it.
This said, I don't see this bug: after (setenv "COLORTERM" nil),
a M-! printenv RET indeed doesn't show COLORTERM any more.

`process-environment' still has an entry for COLORTERM but this entry
has the form "COLORTERM" rather than "COLORTERM=" which distinguishes
the empty-string-envvars from the the absent-envvars.


        Stefan

Re: setenv's broken case when value is nil

[Giorgos Keramidas]

On Wed, 17 Feb 2010 17:56:16 -0800 (PST), Noah Friedman <noah@splode.com> wrote:
> lorentey@elte.hu changed env.el on 2007-08-29 to define setenv-internal and
> for setenv always to keep the variable name in process-environment, even if
> value is nil.  In other words, he removed the distinction between the empty
> string and nil, and therefore you can't remove variables from the
> environment using setenv anymore.
>
> Since all kinds of environment variables have meaning when they are
> defined, even if they are just set to the empty string, this is a screw.

It's not a very common usage pattern of environment variables, but you
are right.  There are all sorts of environment variables that work (for
some value of 'work') even if they are set to empty strings.