Closed andrewjradcliffe closed 2 years ago
Thanks!
Do you know whether the test failures on nightly are new?
The replacement of _maybetail
with safe_tail
is truly 1-for-1 -- the code of safe_tail
is exactly identical to _maybetail
. The test failures on nightly are not due to this particular change.
Inspection of the details indicates uncolon
changes in Base
; looks like
uncolon(inds::Tuple{}, I::Tuple{Colon, Vararg{Any}}) = Slice(OneTo(1))
uncolon(inds::Tuple, I::Tuple{Colon, Vararg{Any}}) = Slice(inds[1])
in v1.8 becomes
uncolon(::Tuple{}) = Slice(OneTo(1))
uncolon(inds::Tuple) = Slice(inds[1])
in v1.9
I am not sufficiently confident that I know how to fix the uncolon
change (well, replicating the v1.8 code within selector.jl
always works, but there might be a better way).
Ok. Thanks for having a look. Let's merge this then, and return to the rest.
This was motivated by the disappearance of
_maybetail
fromBase
, which causesAxisKeys
to not be compatible with v1.9. As it happens,safe_tail
is identical to_maybetail
, thus, it would seem that the latter was eliminated due to redundancy. Inspection of the v1.6 branch of JuliaLang confirms thatsafe_tail
exists in its same form, thus, given the[compat]
for thisAxisKeys
, this seems like a harmless change which enables compatibility withmaster
.