Comment 5 for bug 309125
Revision history for this message
| Christophe Rhodes (csr21-cantab) wrote | #5 |
© 2004
Canonical Ltd.
•
Terms of use
•
Data privacy
•
Contact Launchpad Support
•
Blog
•
Careers
•
System status
•
0f7b58d
(Get the code!)
This has been making me sad.
Firstly, updated patch is attached; it should apply, but it is hand-edited.
Secondly, some of the test cases fail, in surprising ways. In particular, one which fails is
(defun foo (x)
(declare (optimize speed (safety 1)) (type (and simple-bit-vector (satisfies undefined)) x)
(elt x 5))
(foo #*101010) ; see :bug-316078 in tests/compiler.
where what we assume happens is that the hairy type test is weakened by the speed-3/safety-1 policy, and this no longer happens.
The semantics of mixing type and policy declarations is not 100% clear, but the way ANSI explains type declarations as being equivalent to inserting THE forms in the body (including at the head of the body) suggests to me that the test is correct and the new semantics are wrong, for this case. I have not been able to adjust the compiler to implement what I think are the correct semantics, however.