New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
fix&test push/pull recarrays #2034
Conversation
which is totally wrong for recarrays
man you're fast!! This only seems to happen on the pull. I can access things on the engines once they're pushed. |
This is less efficient that what we used to do for simple arrays, as sending the array is ~200B instead of ~3, but that's not very much for sending a real array, and it should be much more general. |
Test results for commit d8f2b33 merged into master
Not available for testing: python3.1 |
Hm, unpickling dtypes doesn't seem to work in numpy 1.5. Need to work on it for a moment more. |
Okay, I think I got it for real this time. testing again... |
Code looks clean, running test_pr right now. Thanks for catching this one, @wkerzendorf, and as usual record-setting response time, @minrk :) |
No need to thank me, @minrk did all the work (and really fast), well if now I only have to get my own code working ;-) |
'str' would fail with zero-size in numpy < 1.6
Test results for commit fc2de03 merged into master
Not available for testing: python2.6, python3.1 |
rerunning test_pr in case b54e738 helps with the failures reported above... |
slow down - there's one more commit after b54. |
ok, restarted run with bdf7984. |
Hm, while I'm pretty sure bdf should work, I might take one more stab at sending not the dtype object itself. |
Nevermind - I'll stick with the full dtype object since I'm not super confident in my understanding of dtype objects. I'll take a stab at cleaning it up when I do my general serialization improvement pass for 0.14. |
Sounds good. Let's be conservative here, and we can ping Travis O. or Mark Wiebe in Austin for a bit of wisdom on possible improvements we could make here while remaining 100% correct. |
Test results for commit bdf7984 merged into master
Not available for testing: python3.1 |
It's relatively minor - with this change, we are sending a few hundred bytes of metadata when ~99% of cases are covered with less than five. I'm pretty sure the right answer is: dtype = array.dtype.descr if array.dtype.shape else array.dtype.str But I don't want to make a change like that immediately before release. |
Test results for commit bdf7984 merged into master
Not available for testing: python2.6, python3.1 |
Conservative but safe, and code looks clean; thanks for the added tests! Merging now, glad we caught this before 0.13! |
fix&test push/pull recarrays: We were only pushing the dtype str for some reason, which is totally wrong for recarrays. closes #2033
Perhaps we should leave an issue open on this, or is it 'in cache' enough that you want to just open the 0.14 PR with that change? |
I'll add a note to #1504, which is for the more general serialization adjustments (and for which I already have a months-old branch in-progress) |
Perfect, thanks. |
fix&test push/pull recarrays: We were only pushing the dtype str for some reason, which is totally wrong for recarrays. closes ipython#2033
We were only pushing the dtype str for some reason,
which is totally wrong for recarrays.
closes #2033