Re: Long-name/short-name complexity

[Ben Pfaff]

John Darrington <address@hidden> writes:

> The short name has to exist only when writing
> sysfiles/portable files.  Its scope can be limited to
> the routines for doing that.  However, I think it will be necessary to
> keep the mapping from long to short names in static memory somewhere
> otherwise the mapping could change unexpectedly.  Consider the
> following scenario:
>
> * foo.sav contains a variable called ABCDEFGHIJ (short name ABCDEF_1)
> GET /FILE='foo.sav'.
>
> AGGREGATE /OUTFILE=* /ABCDEFGHI=MAX(ABCDEFGHIJ).
>
> SAVE /OUTFILE='foo.sav'.
> * foo.sav now contains 2 variables:
> * ABCDEFGHI  (short name ABCDEF_1)
> * ABCDEFGHIJ (short name ABCDEF_2)

It might be nice to do so, but I think there are good reasons not
to do so:

        1. I doubt SPSS does anything similar.  Have you tested
           it?  (I have not.)  I suspect that you would get
           similar behavior to the above using SPSS.

        2. It would only matter for transferring files between
           new versions of PSPP and old versions of SPSS.  Old
           versions of PSPP will reject files with long names, I
           believe; new versions of PSPP and new versions of SPSS
           understand long names.

        3. It will increase complexity.

I'd prefer to retain compatibility, especially where it actually
simplifies our code, hence the preference not to retain name
mappings.

Comments?

Oh, one more thing--do you have an idea of how SPSS does short
name mappings?  The SPSS 12.0 manual suggests it uses a base-10
suffix with a underscore separator.  The "pseudo base-27" idea is
cute but if possible I'd rather be compatible.
-- 
"How could this be a problem in a country
 where we have Intel and Microsoft?"
--Al Gore on Y2K