search

HolyLab / BlockRegistrationScheduler.jl

Multi-core image registration scheduler
1 stars 0 forks source link

Fixes for julia-0.5 #28

Closed timholy closed 7 years ago

timholy commented 7 years ago

I had forgotten about needing to update this package. Requires a fix to julia-0.5 that's only present on cannon. :frowning:

Cody-G commented 7 years ago

Whoops I didn't realize. Is there a public Julia branch that has the fix in the meantime?

timholy commented 7 years ago

You can see the change with a git diff from /usr/local/julia/julia-0.5. It's also https://github.com/JuliaLang/julia/pull/20255. But we don't yet have a 0.5.1, even, and with 0.6 coming pretty soon I'm beginning to wonder if there will ever be one.

Cody-G commented 7 years ago

Cool, just one line. Am I understanding correctly that the move to 0.6 won't be so painful anyway?

timholy commented 7 years ago

Unclear. There are a lot of deprecations for vectorized functions---almost everything that used to be foo(A) (where foo operates elementwise) will now be foo.(A). But I'm not sure how much of our code this will affect.

Of course there is also the type-system overhaul, which may catch some places where we relied on broken dispatch rules. 0.6 seems much more robust about this, but it has also---for the moment---introduced some new bugs (e.g., ColorTypes doesn't work). I'm not going to worry about that yet, though, I suspect the bugs in 0.6 will get ironed out in the next week or two.

If I had to take a wild guess, I'd say it will be on par with previous version-migrations. But hopefully it will be the last bad one for a long time.