hi,
there are two issues here, small regressions.
I think in tmpenv we should create the variable unconditionally (concerning
the namerefs), and should not follow the namerefs on checking if the variable
exists:
diff --git a/variables.c b/variables.c
index 1877b17..f149b66 100644
--- a/variables.c
+++ b/variables.c
@@ -3076,7 +3076,7 @@ assign_in_env (word, flags)
aflags |= ASS_APPEND;
}
- var = find_variable (name);
+ var = find_variable_noref(name);
if (var == 0)
var = find_variable_last_nameref (name, 1);
@@ -3088,14 +3088,6 @@ assign_in_env (word, flags)
return (0);
}
temp = name + offset + 1;
- if (var && nameref_p (var) && valid_nameref_value (temp, 0) == 0)
- {
- /* If we're assigning a value to a nameref variable in the temp
- environment that's an invalid name, flag an error. */
- sh_invalidid (temp);
- free (name);
- return (0);
- }
value = expand_assignment_string_to_string (temp, 0);
cheers,
pg
On 13 May 2016, at 02:58, Grisha Levit wrote:
> The 20160506 snapshot changes assign_in_env to do some nameref-related checks
> on the variable name, but these don’t really seem to make sense since
> assignments in the temporary env just create regular variables and don’t
> follow the nameref chain otherwise.
>
> i.e. bash doesn’t actually care that ref is a nameref at the outer scope for
> the purposes of creating the temp env:
>
> $ unset var; declare -n ref=var
> $ ref=xxx declare -p ref var # the chain is not followed for assignment
> (makes sense)
> declare -x ref="xxx"
> bash: declare: var: not found
>
> $ var= ref=5 declare -p ref
> declare -x ref="5"
>
> $ ref=5 declare -p ref
> bash: `5': not a valid identifier # this shouldn't matter?
> declare -n ref="var"
>
> Same thing for the check to see if ref points to a readonly variable.
>
