Deprecate showcompact()

[nalimilan]

There is no obvious reason to provide a special function for the :compact IOContext property
but not for other properties. showcompact used to be useful to print a single-line representation of an array at the REPL, but now show does the same thing. So it should only be needed for collections to print their elements when implementing show, and for tests (most of the changes here).

Also improve a few docstrings.

[helgee]

That's the wrong way around, isn't it?

[nalimilan]

Thanks. I had originally written "is deprecated in favor of", but I changed it because it's not really a separate deprecation, mostly an advice. Will fix in the next round.

[nalimilan]

Fixed.

[JeffBezanson]

The downside is this has to allocate a new IOContext on every call, but I guess that's life.

[vtjnash]

Specifying a Pair does happen to work, but only the usage case of passing an IO was what I had really originally intended (and found to be common).

I think that we maybe instead should suggest the usage of anonymous functions as a general way of passing a configuration set of arbitrarily many new properties:

showcompact = (io, v) -> show(IOContext(io, :compact => true), v)
sprint(showcompact, 66.66666)

[nalimilan]

OK. I assumed it was intended since the signature doesn't use ::IOContext.

I actually started with the anonymous function approach, but it's quite verbose and much less convenient than passing a pair for common cases (have a look at the diff to get and idea). So I'd say we'd better continue to support passing a pair.

[vtjnash]

I don't intend to prevent passing a Pair, but it simply doesn't generalize as well. Although I suppose it also quickly becomes easier just to write it out:

buf = IOBuffer(sizehint = sizehint)
io = IOContext(buf, :compact => true)
show(io, 66.66666)
return String(take!(buf))

[nalimilan]

My thought was that passing a single pair is the most common case, and for other situations it's not too hard to create a full IOContext.

[JeffBezanson]

I agree it's very nice to be able to pass a single pair.

[ararslan]

If people are concerned about the IOContext part, perhaps we could add a compact keyword argument to printstyled?

[JeffBezanson]

Triage is in favor of removing showcompact, and improving the documentation about how to implement it correctly via show.

[nalimilan]

I've rebased and added a short presentation of IOContext and :compact in the "Custom pretty-printing" section of the manual. Let me know whether it's OK.

[JeffBezanson]

LGTM!