Hi,
I know this patch is a bit hard to swallow given the number of trivial changes in the test suite, but it would be nice to get feedback, at least on the overall approach. As fas as I can tell, the bug is still present in git master.

Thanks, Meador! No, I don't.

Sorry for the ping, but this patch has been marked "accepted and ready to land", and I can't see it on master. Is anything needed on my part?

Hi,
Is there anything else you want me to do on this patch?
Thank you

Updated patch folds the change into raw_ostream::operator<<(double N).

Updated against git master.

I think we'd want APFloat::convertFromDecimalString with rmNearestTiesToEven seeing as how we don't expect to see -nan and the like.

(by the way, a related submission awaiting review is at http://reviews.llvm.org/D15375)

Note the test I added to AsmWriterTest.cpp may not be acceptable (as it changes the locale), but it showcases the issue for review.

Ping again :-) This is bitting our users on a regular basis: https://github.com/numba/numba/issues/1656

Just a ping. I'm not sure there's anything holding this patch?

Here is an updated patch without the locale-changing test.

Thanks for committing this, and sorry for the leak!

Here is a new patch without the optimization settings APIs.

I see this issue has generated several changesets. The latest one seems to be rL222897, where only the Large code model avoids truncation. Am I correct in understanding that the JITDefault code model still generates truncations?