Swift: Expand models for UnsafePointer and friends #15052
Conversation
|
There's also a test failure in a test I hadn't run locally. I'll fix that, then start a DCA run... |
|
... we lose a result in the path injection test because we're modelling collection content in these pointers correctly now, but the I could fix this instance of the problem now by creating duplicate pointer models that don't specify |
Yeah, I agree. Let's just accept this regression for now and come up with a better plan for how to address this. |
|
DCA LGTM. I was seeing consistency check failures in one test on CI for Linux (only). Tried merging in latest main, now there are two tests with consistency check failures. Hmm. :( |
|
Right - it appears this branch has been the victim of some kind of extractor bug, which causes it to fail on some swift code (we don't know what yet) on Linux. An issue has been created for that. The consistency failure in the The consistency failure in the |
geoffw0
force-pushed
the
pointermodels
branch
2 times, most recently
from
2f0964a
to
e506096
Compare
geoffw0
force-pushed
the
pointermodels
branch
from
95d9012
to
36d0148
Compare
|
I've narrowed down the lines of tests that trigger the extractor issue and commented them out. CI now passes. I have created an issue to repair the test after the underlying extractor issue is fixed. |
Expand the models for the 8 core Swift pointer classes (
UnsafePointerand friends), plusSlice. These models were very fragmented and mostly just consisted of the 'closure methods' we've made an effort on + a few things we happened to use in tests of other functionality. Now they're now more-or-less completely modelled.There are quite a large number of MAD rows in this PR because, as far as I can tell, there's not much class hierarchy between the 8 pointer classes. There are lots of similarities / duplication, but also lots of (possibly justified) minor differences. I've only created a random scatter of test cases, we'll just have to trust that the documentation + my models are accurate ... or alternatively, trust that the impact of any mistakes is likely to be quite low compared to the impact of not having these models at all!