[PATCH] m4_set_foreach: pop macro definition

The current implementation of m4_set_foreach leaves its macro variable
defined to whatever the final element of the set was.  This differs
from m4_foreach, which pushes & pops the definition in its expansion.

Since this macro is often used with really short macro names (the
manual has an example with m4_set_foreach([a], [i], ...)), this can
lead to some incredibly surprising behaviour.  For example:

  % cat >test.m4 <<'EOF'
  m4_init
  m4_define([_m4_divert(BODY)], [0])
  m4_set_add([a], [not_i])
  m4_set_foreach([a], [i])
  m4_divert_text([BODY], [i])
EOF

  % autom4te -l m4sugar test.m4
  not_i

Turns out the implementation already had a pushdef, and it looks like
the popdef was simply forgotten.  Add it now.

* lib/m4sugar/m4sugar.m4 (m4_set_foreach): pop macro definition.
---
 lib/m4sugar/m4sugar.m4 | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/lib/m4sugar/m4sugar.m4 b/lib/m4sugar/m4sugar.m4
index a93b4392e1c5..5421ed4894b0 100644
--- a/lib/m4sugar/m4sugar.m4
+++ b/lib/m4sugar/m4sugar.m4
@@ -3126,7 +3126,8 @@ m4_define([m4_set_empty],
 # guaranteed.  This is faster than the corresponding m4_foreach([VAR],
 #   m4_indir([m4_dquote]m4_set_listc([SET])), [ACTION])
 m4_define([m4_set_foreach],
-[m4_pushdef([$2])m4_set_map_sep([$1], [m4_define([$2],], [)$3])])
+[m4_pushdef([$2])m4_set_map_sep([$1], [m4_define([$2],], [)$3])dnl
+m4_popdef([$2])])
 
 # m4_set_intersection(SET1, SET2)
 # -------------------------------
-- 
2.2.2