[BUG] emulate sh: arith assignment assigns variable type

[BUG] emulate sh: arith assignment assigns variable type

[Martijn Dekker]

When an assignment is done to an unset variable using an arithmetic
expression, zsh assigns a numerical or arithmetic type to that variable,
causing subsequent normal shell assignments to be interpreted as
arithmetic expressions.

This causes an incompatibility with other shells under 'emulate sh',
because the POSIX shell is supposed to be a typeless language. In sh
emulation mode, arithmetic assignment should not assign a variable type.

Steps to reproduce:

$ emulate sh
$ X=a:b:c
$ : $((X=1))	# no type assignment as variable was already et
$ X=a:b:c
$ unset X
$ : $((X=1))	# variable now restricted to arith expressions
$ X=a:b:c
zsh: bad math expression: ':' without '?'

Expected behaviour: the last assignment should succeed in sh emulation
mode (and does on all other shells).

Confirmed in zsh 4.1.1 through current.

Thanks, and happy new year.

- Martijn

Re: [BUG] emulate sh: arith assignment assigns variable type

[Peter Stephenson]

On Fri, 01 Jan 2016 21:03:33 +0100
Martijn Dekker <martijn@inlv.org> wrote:
> When an assignment is done to an unset variable using an arithmetic
> expression, zsh assigns a numerical or arithmetic type to that variable,
> causing subsequent normal shell assignments to be interpreted as
> arithmetic expressions.

Yes, it does.

Does this need a special option?  The linkage is a bit tenouous.

diff --git a/Doc/Zsh/options.yo b/Doc/Zsh/options.yo
index f377539..b5e9100 100644
--- a/Doc/Zsh/options.yo
+++ b/Doc/Zsh/options.yo
@@ -2098,6 +2098,12 @@ When it is unset, zsh allows expressions of the form tt($#)var(name)
 to refer to the length of tt($)var(name), even for special variables,
 for example in expressions such as tt($#-) and tt($#*).
 
+Another difference is that with the option set assignment to an
+unset variable in artihmetic context causes the variable to be created
+as a scalar rather than a numeric type.  So after `tt(unset t; (( t = 3
+)))'. without tt(POSIX_IDENTIFIERS) set tt(t) has integer type and with
+it set it has scalar type.
+
 When the option is unset and multibyte character support is enabled (i.e. it
 is compiled in and the option tt(MULTIBYTE) is set), then additionally any
 alphanumeric characters in the local character set may be used in
diff --git a/Src/params.c b/Src/params.c
index 8cab969..054fb1f 100644
--- a/Src/params.c
+++ b/Src/params.c
@@ -3061,6 +3061,7 @@ setnparam(char *s, mnumber val)
 	if (ss)
 	    *ss = '\0';
 	pm = createparam(t, ss ? PM_ARRAY :
+			 isset(POSIXIDENTIFIERS) ? PM_SCALAR :
 			 (val.type & MN_INTEGER) ? PM_INTEGER : PM_FFLOAT);
 	if (!pm)
 	    pm = (Param) paramtab->getnode(paramtab, t);
diff --git a/Test/C01arith.ztst b/Test/C01arith.ztst
index c7bd81f..61da763 100644
--- a/Test/C01arith.ztst
+++ b/Test/C01arith.ztst
@@ -409,3 +409,14 @@
 >2
 >(eval):6: bad math expression: unexpected ')'
 >(eval):7: bad math expression: unexpected ')'
+
+  unset number
+  (( number = 3 ))
+  print ${(t)number}
+  unset number
+  (setopt posix_identifiers
+  (( number = 3 ))
+  print ${(t)number})
+0:type of variable when created in arithmetic context
+>integer
+>scalar

Re: [BUG] emulate sh: arith assignment assigns variable type

[Mikael Magnusson]

On Sat, Jan 2, 2016 at 7:11 PM, Peter Stephenson
<p.w.stephenson@ntlworld.com> wrote:
> On Fri, 01 Jan 2016 21:03:33 +0100
> Martijn Dekker <martijn@inlv.org> wrote:
>> When an assignment is done to an unset variable using an arithmetic
>> expression, zsh assigns a numerical or arithmetic type to that variable,
>> causing subsequent normal shell assignments to be interpreted as
>> arithmetic expressions.
>
> Yes, it does.
>
> Does this need a special option?  The linkage is a bit tenouous.
>
> diff --git a/Doc/Zsh/options.yo b/Doc/Zsh/options.yo
> index f377539..b5e9100 100644
> --- a/Doc/Zsh/options.yo
> +++ b/Doc/Zsh/options.yo
> @@ -2098,6 +2098,12 @@ When it is unset, zsh allows expressions of the form tt($#)var(name)
>  to refer to the length of tt($)var(name), even for special variables,
>  for example in expressions such as tt($#-) and tt($#*).
>
> +Another difference is that with the option set assignment to an
> +unset variable in artihmetic context causes the variable to be created

Typo here on artihmetic. I would also put a comma after 'with the option set'?

-- 
Mikael Magnusson

Re: [BUG] emulate sh: arith assignment assigns variable type

[Bart Schaefer]

On Jan 2,  6:11pm, Peter Stephenson wrote:
}
} Does this need a special option?  The linkage is a bit tenouous.

Are there other assignment-time differences?  Maybe a POSIX_ASSIGNMENT
option is in order, though I dislike moving functionality from one
option to another.

On the other hand, is this --

}  	pm = createparam(t, ss ? PM_ARRAY :
} +			 isset(POSIXIDENTIFIERS) ? PM_SCALAR :
}  			 (val.type & MN_INTEGER) ? PM_INTEGER : PM_FFLOAT);

-- really the correct fix?  I was thinking more along the lines of
creating it as an integer but then changing the type back to scalar
at the time of assignment if the assigned string did not parse as an
arithmetic expression.

Re: [BUG] emulate sh: arith assignment assigns variable type

[Peter Stephenson]

On Sat, 2 Jan 2016 12:46:59 -0800
Bart Schaefer <schaefer@brasslantern.com> wrote:
> On the other hand, is this --
> 
> }  	pm = createparam(t, ss ? PM_ARRAY :
> } +			 isset(POSIXIDENTIFIERS) ? PM_SCALAR :
> }  			 (val.type & MN_INTEGER) ? PM_INTEGER : PM_FFLOAT);
> 
> -- really the correct fix?  I was thinking more along the lines of
> creating it as an integer but then changing the type back to scalar
> at the time of assignment if the assigned string did not parse as an
> arithmetic expression.

That's obviously not what other POSIX shells do, so I don't see how that
can be right.  It's forcing arithmetic evaaluation in cases those shells
wouldn't.  The POSIX* options aren't supposed to be for anything more
than strict(ish) compatibility.

pws

Re: [BUG] emulate sh: arith assignment assigns variable type

[Martijn Dekker]

Peter Stephenson schreef op 02-01-16 om 19:11:
> On Fri, 01 Jan 2016 21:03:33 +0100
> Martijn Dekker <martijn@inlv.org> wrote:
>> > When an assignment is done to an unset variable using an arithmetic
>> > expression, zsh assigns a numerical or arithmetic type to that variable,
>> > causing subsequent normal shell assignments to be interpreted as
>> > arithmetic expressions.
> Yes, it does.
> 
> Does this need a special option?  The linkage is a bit tenouous.

It's working fine in any case. Thanks.

I noticed the fix has made it into the git repository, but without a
corresponding ChangeLog entry. Thought I'd mention it in case that was
an unintentional omission.

- M.