Pass ndimx = 7 to dat_shape

[grahambell]

This module's tests fail when being built with a version of Starlink
including HDS-v5. Passing ndimx > DAT__MXDIM to dat_shape is giving a
segmentation fault because hds_select's datShape is writing beyond
the size of a temporary array allocated by the fortran interface's
DAT_SHAPE.

[timj]

I'm a bit confused. I thought that the second argument to dat_shape was indicating the amount of space to allocate and that HDS would only fill in up to 7 because it couldn't take more dimensions (and I am guessing that DAT__MXDIM wasn't easily available in Perl so I just picked a large number). I wish I had written those perl APIs to return the answer rather than having return arguments.

[timj]

Shall I move the repo to Starlink org?

[grahambell]

I think HDS would only fill up to 7, but new code in hds_select copies up to the specified number of values from its temporary array. The problem is really the fortran interface which allocates a 7-entry array but then passes on the given size, i.e. potentially claiming that the array it's passing is more than 7 entries long. There are lots of functions in the fortran interface which do this kind of thing:

   hdsdim cdims[DAT__MXDIM];
   /* ... */
   datShape( locator_c, *ndimx, cdims, ndim, status );

(so they will fail if *ndimx > DAT__MXDIM and the underlying code trusts the given ndimx).

Moving it to the Starlink organization sounds a good idea. I would like to also make a new CPAN release so that we can easily install the module here. (I think the SCUBA-2 quick-look uses it.)

[timj]

@grahambell I think you now have control of this PR.