add guidelines for testing dicts to devel guide

[kcrisman]

The developer guide doesn't mention that dictionaries can come out platform-dependent or worse in Python. New developers should have a good place to find a good example or two of how to deal with this.

Component: documentation

Reviewer: Jeroen Demeyer, Karl-Dieter Crisman

Issue created by migration from https://trac.sagemath.org/ticket/17362

[kcrisman]

comment:2

E.g. something along the lines of

sage: D = {1:2, 2:3} # possibly random order depending on hashing
sage: E = {2:3, 1:2}
sage: sorted(D.items()) # guaranteed
[(1, 2), (2, 3)]
sage: sorted(E.items()) == sorted(D.items())
True

but perhaps someone might have an actual example that had to be fixed for the Sage doc to test right, I am pretty sure Trac is full of these.

[jdemeyer]

Replying to @kcrisman:

dictionaries can come out platform-dependent or worse in Python.

Isn't this fixed by #16746?

[kcrisman]

comment:4

dictionaries can come out platform-dependent or worse in Python.

Isn't this fixed by #16746?

I didn't know about that aspect of it, only the graphics part. So will doctests automatically fail if people put them in the order that their computer gives and it's not sorted? That could be confusing! So maybe it should still be mentioned in the guide, along with the graphics piece - which is not what the command line gives.

And, more relevant to my immediate concern, what if the keys themselves are dictionary-enough that they are platform-dependent? See #16331 for an explicit example of this. It would be great if you could give some advice for that, because we haven't come to any useful consensus yet.

[jdemeyer]

comment:5

Replying to @kcrisman:

So will doctests automatically fail if people put them in the order that their computer gives and it's not sorted?

There are 2 dict orders: the Python order and the IPython order. The Python order is what you get with print statements, the IPython order is the sorted order. The latter is used by the IPython interface, by sagenb and by doctests. So, I would say that the order in doctests after #16746 is actually the order that "their computer gives".

[jdemeyer]

comment:6

Replying to @kcrisman:

And, more relevant to my immediate concern, what if the keys themselves are dictionary-enough that they are platform-dependent? See #16331 for an explicit example of this. It would be great if you could give some advice for that, because we haven't come to any useful consensus yet.

Well, #16331 is special because you're sorting a list containing integers and strings. The problem is that the output of sorted([3,"foo"]) isn't well-defined.

[kcrisman]

comment:7

And, more relevant to my immediate concern, what if the keys themselves are dictionary-enough that they are platform-dependent? See #16331 for an explicit example of this. It would be great if you could give some advice for that, because we haven't come to any useful consensus yet.

Well, #16331 is special because you're sorting a list containing integers and strings. The problem is that the output of sorted([3,"foo"]) isn't well-defined.

Yikes, and apparently isn't even allowed in Python 3.

the IPython order is the sorted order.

Hmm, although I still think some updating could be done, I guess a ticket just for that or the graphics thing, which is now everywhere, probably isn't that worthwhile.

[jdemeyer]

comment:8

Close as "wontfix"?

[kcrisman]

comment:9

ok

[jdemeyer]

Reviewer: Karl-Dieter Crisman

[kcrisman]

Changed reviewer from Karl-Dieter Crisman to Jeroen Demeyer, Karl-Dieter Crisman