Please add FOREIGN-ARRAY-TO-LISP-SPECIALIZED

[Michał "phoe" Herda]

CFFI's FOREIGN-ARRAY-TO-LISP always allocates vectors of element type T, even where it has all the required information to allocate more specialized arrays. This means that client code needs to do costly coercions if it wants to use CFFI converted vectors inside code that requires, for instance, octet vectors, even if the array type provided to FOREIGN-ARRAY-TO-LISP specifically states that the data is a vector of UNSIGNED-CHARs.

Backwards compatibility requires that FOREIGN-ARRAY-TO-LISP returns vectors of type T. Therefore, I request that a new function is added, that I roughly called FOREIGN-ARRAY-TO-LISP-SPECIALIZED, that automatically creates and returns arrays of properly specialized types.

[Luís Oliveira (luismbo)]

How about adding &REST MAKE-ARRAY-ARGUMENTS to FOREIGN-ARRAY-TO-LISP arglist then applying MAKE-ARRAY to that?

[Michał "phoe" Herda (phoe-krk)]

@luismbo Your proposition seems like a backwards compatible change, and also much cleaner than mine. I am in favor of that.

[Luís Oliveira (luismbo)]

Feel like opening a pull request with that change, a test and a manual update? :-)

[Michał "phoe" Herda (phoe-krk)]

I can try.

What kind of test suite should I run once I have made the change?

[Luís Oliveira (luismbo)]

IIRC these array functions are not being tested nor documented in the manual, so you'll have to create a new test in the cffi-tests suite. A simple test might be to use foreign-array-to-lisp to create a displaced array (or an adjustable array?) and then test that indeed such an array has been created. (I don't think it's easy to portably test with specialized arrays because not all Lisps support the same kind of specialized arrays.)

[Michał "phoe" Herda (phoe-krk)]

PR created at https://github.com/cffi/cffi/pull/128

Test results:

6 out of 322 total tests failed: FUNCALL.VARARGS.DOUBLE,
   DEFCFUN.VARARGS.FLOAT, DEFCFUN.VARARGS.DOUBLE, FSBV.WFO,
   FSBV.MAKEPAIR.1, FSBV.MAKEPAIR.2.
3 unexpected failures: FUNCALL.VARARGS.DOUBLE, DEFCFUN.VARARGS.FLOAT,
   DEFCFUN.VARARGS.DOUBLE.

[Luís Oliveira (luismbo)]

PR accepted. The unexpected failures happen because those 3 tests are sensitive to locale settings. :-(