Chapel: Should lists support slicing by unbounded ranges?
Summary of Problem
It seems reasonable to say myList[lo..] or myList[..hi] (or even myList[..]) yet these are not supported at present and generate a not-particularly-helpful error message. I think we should add support for these list slices, or if not, generate a more useful error message.
Steps to Reproduce
Associated Future Test(s):
test/library/standard/List/this/unboundedListSlice.chpl #16122
Configuration Information
- Output of
chpl --version:chpl version 1.23.0 pre-release (c3edd9e4d3)
bradcray
All 6 comments
@dlongnecke-cray: For your consideration.
bradcray
on 24 Jul 2020
Sure, if we were to support slicing a list by a range, what would we expect the operation to produce? An iterator? A "ListView"?
dlongnecke-cray
on 24 Jul 2020
Oh, do we not already support slicing a list by a range? It seemed to work with bounded ranges, which is why I assumed it was supported/intended. But maybe you're saying that it's only working by promoting the serial list accessor? (which makes the current error message make a bit more sense. I think that definitely complicates this question a little bit and gets into a larger design topic that I've been wondering about recently which is "how good is Chapel at permitting users to create aliasing views of their data structures, and do we need to do more there?"
bradcray
on 24 Jul 2020
I reworded the title of this issue to reflect less certainty now that I understand we didn't put in effort to support slicing.
bradcray
on 24 Jul 2020
Related to #15707 ?
Edit -- Nvm, I was looking at two different issues and got confused
rahulghangas
on 24 Jul 2020
It seems a little bit related to #11222 in that #11222 is arguably requesting string views/borrows and this is arguably requesting list views/borrows.
mppf
on 25 Jul 2020
Related issues
ben-albrecht
路
3Comments
ben-albrecht
路
3Comments
lydia-duncan
路
3Comments
BryantLam
路
3Comments
astatide
路
3Comments
Most helpful comment
Oh, do we not already support slicing a list by a range? It seemed to work with bounded ranges, which is why I assumed it was supported/intended. But maybe you're saying that it's only working by promoting the serial list accessor? (which makes the current error message make a bit more sense. I think that definitely complicates this question a little bit and gets into a larger design topic that I've been wondering about recently which is "how good is Chapel at permitting users to create aliasing views of their data structures, and do we need to do more there?"