deprecate print_shortest?

[StefanKarpinski]

This is kind of a weird, old legacy function that we export. Should we deprecate it? Use an IO context flag? Just unexport it?

[vtjnash]

Triage decided the code can remain in Grisu, but it shouldn't be exported anymore

[Keno]

Triage: We should un-export this from base, but keep it available in Grisu if people want to use it directly. Using Grisu directly is sometimes a reasonable thing to do if you want really custom float printing, e.g. the plotting packages use it to print data points with consistent precision.

[strickek]

print_shortest is also used in writedlm(), line 733 of DelimitedFiles in stdlib.

[StefanKarpinski]

As part of the stdlib it's fairly kosher for that to depend on an internal Base function since they will be updated in sync.

[JeffBezanson]

This could become a :shortest flag in IOContext. Grisu already looks at the :compact flag.

[StefanKarpinski]

We could also just get rid of this since the only reason DelimitedFiles calls it is because I put it there and I think various people would prefer that it use more standard short-but-not-necessarily-the-tricksiest-and-very-shortest printing.

[vtjnash]

I've been strongly reluctant to have IOContext be used to pass specific formatting choices recursively. I think if you care about that level of detail in your output, you should [have to] define the recursive structure also.

Not sure why mbauman removed needs-tests – this function does not appear to have any. If it's not simply deleted, it needs them written instead.

[StefanKarpinski]

This issue doesn't add any features so this is not the place to put the "needs tests" label. If you want to open a separate issue about testing print_shortest that would be fine, but that seems premature since we may decide to delete it.

[ararslan]

Now that we have printstyled, we could fold shortest into a keyword argument to that function.

[StefanKarpinski]

Frankly, I'm just not sure this is all that necessary of a function. But sure, that's an option.

[strickek]

I think the magic of print_shortest in writedlm() is not necessary. It lose the type differentation for floats with '*.0' because print_shortest write them as '0', so afterwards you dont know wether this element was originaly an Int or an Float.