Hello, The patches I've just pushed lloda-array-support (ending in http://git.savannah.gnu.org/gitweb/?p=guile.git;a=commitdiff;h=64fa67c2f0e08e42a60921db4aa77fa3e52d3fa6) implement the functions (array-from) and (array-amend!), to take or modify array prefix slices. The patches include documentation and tests.
I'd like a version of these patches to be merged before 2.2 is released. These functions do not cover all the cases of multidimensional array subscripting that one may find in other languages, such as Fortran or APL. They are only a foundation. The general subscripting operator implemented in https://gitorious.org/guile-ploy (with the names from / set-from!) is built on top of array-from / array-amend!. I have chosen these particular forms because the implementation is especially simple, nearly identical to that of array-ref. The replacement of make-shared-array pays off in terms of speed. For example, the tests of guile-ploy run in 2.1s with the versions of array-from and array-amend! included in these patches, but take 3.5s if array-from and array-amend! are implemented on top of make-shared-array. Guile/Scheme is still missing an efficient way to iterate over array slices. Logically this should only require bumping a pointer on each iteration. But with the limited array primitives provided by Guile, one must create a shared view on each iteration and manage the loop separately. I think that eventually it will be necessary to expose the array strides and dimensions in Scheme. It may take some effort to do this without making access from C more difficult than it already is. Regards, Daniel
