Re: Count system missing columns

[Ben Pfaff]

On Wed, Dec 23, 2015 at 01:15:30AM +0100, John Darrington wrote:
> On Mon, Dec 21, 2015 at 03:53:58PM -0800, Ben Pfaff wrote:
>      On Mon, Dec 21, 2015 at 04:42:24PM -0600, Alan Mead wrote:
>      > On 12/21/2015 4:10 PM, Ben Pfaff wrote:
>      > > I think that you just want: COUNT n=ALL(SYSMIS). 
>      > 
>      > Brilliant!  But it doesn't wok.  On my datafiles, I get a long list of
>      > errors like:  "V1 and V8 are not of the same type.  All variables in
>      > this variable list must be of the same type. V8 will be omitted from 
> the
>      > list." (where, indeed, V1 and V8 are string variables of different 
> lengths).
>      > 
>      > Finally, it complains that SYSMIS isn't a string and fails to execute
>      > the COUNT: "error: COUNT: Syntax error at `missing': expecting 
> string." 
>      > This happens whether I use MISSING or SYSMIS. 
>      > 
>      > I thought that the above was a bug, and maybe it is, because I should 
> be
>      > able to use SYSMIS/MISSING but it works fine when I specify the columns
>      > to use, so I think it only fails when I specify ALL as the columns.
>      > 
>      > BTW, it would be neat if this worked in PSPP, but SPSS refuses to 
> accept
>      > ALL as a column range.
>      
>      Hmm.
>      
>      We could make COUNT accept this, or at least give better error
>      messages.  But it still won't work any better with SPSS.
>      
>      In the end, the SPSS language isn't very good at this kind of thing.
>      
> My first idea to work around this was to use RECODE.  But it seems that
> RECODE also cannot deal with heterogenous variable types.
> I think it would be good to have a wishlist item to allow COUNT and RECODE
> to do that.

For COUNT, only "MISSING" can be used for both numeric and string
variables, so I think the utility of this feature would be limited.