Deprecate compute-in-repr behavior.

[kwmsmith]

Start process to replace Expr.repr with a more standard
implementation.

Adds a peek method to do what __repr__ used to do.

@llllllllll For the life of me I can't figure out how to configure the warning filters so that the warning only shows up once -- do you have any guidance here?

[llllllllll]

we should also make the _repr_html_ not implicitly compute when in jupyter.

[llllllllll]

we are actually using a deprecation warning here

[llllllllll]

Other than the comment this looks good. Regarding the visibility of the warning: I think that we should not be forcing warnings to be more visible than they are by default. If users do not want to see deprecation warnings then that is on them.

[kwmsmith]

@llllllllll thanks for the feedback. I addressed your comment, updated whatsnew, and tweaked the __repr__ implementations for Symbol, _Data, and Expr to be more consistent:

In [1]: import blaze as bz

In [2]: t = bz.symbol('t', dshape='var * {name: string, id: int}')

In [3]: t
Out[3]: <`t` symbol; dshape='var * {name: string, id: int32}'>

In [4]: d = bz.data([['Alice', 1], ['Bob', 2]], dshape='var * {name: string, id: int32}', name='d')

In [5]: d
Out[5]: <'list' data; _name='d', dshape='var * {name: string, id: int32}'>

In [6]: d.name
Out[6]:
    name
0  Alice
1    Bob

In [7]: bz.interactive.use_new_repr = True

In [8]: d.name
Out[8]: <`Field` expression; dshape='var * string'>

In [9]: t.id ** 2
Out[9]: <`Pow` expression; dshape='var * int64'>

In [10]: str(t.id ** 2)
Out[10]: 't.id ** 2'

In [11]: d.name.peek()
Out[11]:
    name
0  Alice
1    Bob

Things to point out:

• The __repr__s are all boring reprs now that show the dshape along with the type of the thing.
• Long dshapes are truncated with a ... at the end.
• The string form of an expression--whether compound or atomic--uses the underlying _name of the leaf terms to keep the string form looking nice.

[kwmsmith]

I'm still looking into the best way to address the doctest failures.

[llllllllll]

The doctest failures look like they are just relying on the new symbol repr, I think it is safe to just change them. I think we might want to include a bit more data in the reprs as we go, but we can do that after merging this. One immediate thing I am thinking of is including the field name in Field repr.

[kwmsmith]

The doctest failures look like they are just relying on the new symbol repr, I think it is safe to just change them.

+1 -- although I didn't expect you to say that :)

I think we might want to include a bit more data in the reprs as we go, but we can do that after merging this. One immediate thing I am thinking of is including the field name in Field repr.

+1.

[llllllllll]

I'm really glad to see this feature merged, thanks!