stop exporting HTML and Text from Base

[musm]

Stop exporting HTML and Text from Base.

The Text one is pretty annoying since it's easy to clash with.

[StefanKarpinski]

It might be possible to remove these exports before 2.0 but only if no one in any package is using them.

[stevengj]

The HTML function seems to be used in a bunch of packages. Text is also used, e.g. in Tk.jl.

[KristofferC]

It is used but it is explicitly imported so removing the export wouldn't be breaking.

[stevengj]

There are other packages that use Text without importing, e.g. DiffEqBase.

[StefanKarpinski]

It is used but it is explicitly imported so removing the export wouldn't be breaking.

Seems sneaky and kind of crappy to go from exported—and therefore official API—to unexported, taking advantage of the fact that someone else was a good end-user explicitly and imported it, only to then turn around and break their code by deleting or moving what used to be an official API.

[musm]

Don't we have femtocleaner for precisely these types of trivial substitutions?

[ChrisRackauckas]

The ASCII colors required Text in order to render correctly on I Julia IIRC, but I can double check. Just remove the export and I'll make sure to Base. it ASAP.

[stevengj]

If you take semver seriously, then this is simply not an option until Julia 2.0.

[musm]

I'd also add @text_str to this list.

[tknopp]

If you take semver seriously, then this is simply not an option until Julia 2.0.

I second that. Public packages are not the only Julia code out there.

[StefanKarpinski]

Believe me, I hate that we export this as much as anyone, but we do, now we have to live with it.

[clarkevans]

I didn't grok it at first, but now I think HTML is brilliant -- it's just missing a critical fix #38892 that would enable it to be really useful. Perhaps what is missing is documentation that explains how hypertext processing could be done in Julia? I can work on a pull request for documentation, but it'd need #38892 to be interesting.

Anyway, I use HTML as a central part of HypertextLiteral.jl. I would also note that HTML provides a trait object for content that is intended to be showable(text/html). It's much faster to have dispatches based upon HTML that can be resolved at compile time, rather than having runtime checks for showable, which is an order of magnitude slower. In short, I did feel initially that HTML (and Text) collided with what I wanted to do... till I understood how they worked.

If we want to drop HTML perhaps there's a way to hook into the compilation stage that notices that HTML is used but not Docs.HTML? It'd be easy enough for packages out there to make a gradual change to Docs.HTML. We could start by searching github for usage and ask those package vendors to prefix it. Also the documentation could encourage this.