Add CI | Fix '(array (signed-byte 8)) is not of type (array (signed-byte *))'

[digikar99]

As mentioned here and evident with this commit, #6 was not good!

Here's another attempt - tests included this time!

I was also playing around with OpenCV via py4cl2 and the current fixes seem okay. Check the tests! CCL fails on (signed-byte 64) case though. (See this.)

[marcoheisig]

I have incorporated your changes (after simplifying the test-suite a bit). But I am not sure I want to use the Github CI. Not because I don't want to have CI, but because I don't want to have a .github directory in my repository.

I will think some more about how this CI issue can be solved gracefully.

Thanks for writing a test suite!

[digikar99]

Hmm, the given test suite turns out to be non-exhaustive. I had added this part because the floats are actually encoded as (unsigned-byte 32) rather than (signed-byte 32). But the same logic should be applicable to (complex single-float) and (complex double-float).

And indeed, this fails

(store-array (make-array 1 :initial-element #c(-1.0e15 0)
                         :element-type '(complex single-float))
             "/tmp/nff")

So, the overall logic for stream-element-type should be:

(if (or (member element-type '(single-float double-float
                               (complex single-float)
                               (complex double-float))
                :test #'equal) ; might require type= some other time
        (subtypep element-type '(unsigned-byte *)))
    `(unsigned-byte ,chunk-size)
    `(signed-byte ,chunk-size))

Alternatively, the tests pass (they shouldn't pass!) even if stream-element-type was given solely by

(if (subtypep element-type '(unsigned-byte *))
    `(unsigned-byte ,chunk-size)
    `(signed-byte ,chunk-size))

So, the tests for complex/floats should involve numbers like -1e15 and -1d15 that are not of type (signed-byte 32) and (signed-byte 64) respectively.

---

Could add something that says this in a comment:

Intuitively, it feels as if a simpler logic should be applicable, but can't think of any at the moment. Prime reasoning seems to be

• Even if the element-type is not unsigned, we need the stream to be unsigned in the case of complex/floats. So, an approach that simply checks for (not (subtypep element-type '(unsigned-byte *))) fails, since this does not account for element-type being a complex/float.
• Another approach that'd check for (subtypep element-type '(signed-byte *)) also fails because (unsigned-byte *) is a subtype of (signed-byte *), so this path would be taken regardless of whether element-type is a signed-byte or unsigned-byte.

And, TIL uiop:with-temporary-file!