I don't think it is reasonable to blame me for encountering a bug in their code, the problem is in their code. Please read the success story - the person advocating for it describes plenty of time spent compensating for it for reasons much more diverse than this one example. This is a common experience - mentioned in several comments here too.

Well, it simply isn't a bug. Most sort algorithms aren't stable, and the only thing a generic _sort_ algorithm is supposed to guarantee is a sorted output according to the comparison function you supply it with. If however the method was called "stableSort", the situation would be very different...

The default Array.sort stability is inconsistent across targets _because_ guaranteeing it would prevent reusing implementations already available in the targets. And differently than what you claimed, many languages (and most Haxe targets) don't guarantee sort stability in their default "sort" APIs: C, C++, C#, JS, PHP and Neko.

Also, any remark placed in the official api docs for "Array" or in that very module, just next to sort method, can't be said to be "buried" in the docs.

It is not important we agree on this being a bug or not. If you have had a smooth experience you don't have to look further than the linked success story to find a counter-example.

Respectfully, it is important to understand whether sort stability is as bug or not. Particularly since you seem to have experienced exceptions on Android, and assuming that the problem was in the sort stability may have mislead you.

It's still reproducible just by substituting the top function for the stable alternative we use: http://pastebin.com/D3S6SYBM. Haxe 3.1.3 on OSX.

Do you have the same problem with a deterministic comparison function?

I'm not dismissing the existence of a bug just because of this (although the issue may be related to it), but non-deterministic comparison functions are problematic with many sort _algorithms_. You might experience (other) problems when using implementation independent "sort" APIs, since many algorithms need to rerun the comparison function several times and expect consistent behaviour; I would expect certain algorithms to, for instance, never complete.

The haxe people are the ones you need to discuss it further with. My pastebin expires in a week so you might want to clone it before you link to it.

Hm... No.

I don't want to attach myself to further discussion of it, but it is still a crash reasonably isolated to their standard library and I took the time to document it so that my feedback could be helpful. If you don't pass it on they will miss this opportunity to improve their platform.