Re: make-vector/vector-length disagree for large vectors.

[Dirk Herrmann]

On 7 Feb 2001, Rob Browning wrote:

> 
>   guile> (define xxx (make-vector 1000000))
>   guile> (vector-length xxx)
>   1000000
>   guile> (define xxx (make-vector 100000000))
>   guile> (vector-length xxx)                 
>   16113920
>   guile> (version)                           
>   "1.4"

In the current implementation, guile's vectors have a maximum length of
2^24 elements.  It is not difficult to change:  Instead of a single cell,
vectors could be represented using a double cell, which would allow for
2^32 or even 2^64 vector elements.  But, up to now obviously nobody
thought that this would be necessary.

It is a bug, though, that make-vector does not complain if the given
argument is too big.  The only thing that is actually checked is, whether
the given size is a fixnum.  If it is a bignum, you get a "Wrong type
argument" error message, which is also strange:
  Wrong type argument in position 1: 536870912
:-)

The bug with the missing boundary checking is easy to fix.  Whether
vectors should be changed to allow for more than 2^24 elements is up to
the maintainers.  If the maintainers would decide that way, one could
think of unifying all the different vector types within guile:  uniform
vectors, weak and non weak object vectors.

Best regards,
Dirk Herrmann