Deprecate DevNull, STDIN, STDOUT, and STDERR to lowercase

[ararslan]

EDIT: Initially this PR was just to capitalize DevNull but the intent has since changed. The text below summarizes the initial state of the PR.

---

This is consistent with the other stream constants, i.e. STDOUT, STDIN, and STDERR, and indeed this was a violation of the typical style of all-caps constants.

Fixes #25786.

[vtjnash]

was a violation of the typical style of all-caps constants.

In that case, shouldn't we make all of these lower-caps instead? (that's how they are in C, for example: stdin / stdout / stderr / stdio / devnull).

EDIT: clarification: these aren't constants

[ararslan]

In that case, shouldn't we make all of these lower-caps instead?

I don't understand what you mean. We don't do that for anything else, and these aren't just for C compatibility.

[JeffBezanson]

Should move this to the constants block alongside STDOUT etc.

[StefanKarpinski]

Most of our global singletons like nothing and missing are lowercase. There's also the fact that this looks quite suggestive of what it does:

pipeline(`echo Hello`, stdout=stderr)

i.e. temporarily replace stdout with the value of stderr. I'm in favor of changing these all to lowercase: stdin, stdout, stderr, devnull.

[ararslan]

In the particular example of pipeline, you run into the issue where the keyword argument name shadows a global:

julia> const x = 1
1

julia> f(x=4, y=x) = (x, y)
f (generic function with 3 methods)

julia> f()
(4, 4)

Note that the keyword argument value of x is used rather than the global x. In pipeline, it's possible to specify both stdout and stderr, so without constants that can be differentiated by name from those, everything will get all messed up.

[JeffBezanson]

That's only an issue for the method definition, not call sites. In the method definition these default to nothing, so there should be no problem.

[JeffBezanson]

I think there's still some controversy here over whether we want lowercase or uppercase.

[ararslan]

Per triage consensus, this now deprecates DevNull, STDIN, STDOUT, and STDERR to devnull, stdin, stdout, and stderr, respectively. I did this in three commits: one to change uses of the lowercase versions as local variables, one to deprecate DevNull, and one to deprecate the others.

[StefanKarpinski]

I find this to be rather pleasant. Thanks for doing it despite dissenting, Alex! In 1.0 we could rename DevNullStream to DevNull. Not sure if that’s even a public name.

[ararslan]

Hm, looks like there's an issue building the docs. I'm not sure what the deal is yet. EDIT: I think it's a Documenter thing.

[ararslan]

Okay, doc building is apparently still an issue, even after updating Documenter.

[mortenpi]

In case you already didn't come to the same conclusion: withoutput crashes when evaluating the following REPL block:

```@repl
versioninfo()
```

It works fine on master though, so it seems like something in this PR.

Edit: here's the error (I think)

ERROR: LoadError: MethodError: no method matching redirect_stdout(::Core.CoreSTDOUT)
Closest candidates are:
  redirect_stdout() at stream.jl:1083
  redirect_stdout(!Matched::Union{IOStream, Base.LibuvStream}) at stream.jl:1077
  redirect_stdout(!Matched::Function, !Matched::Any) at stream.jl:1131
Stacktrace:
 [1] withoutput(::getfield(Documenter.Expanders, Symbol("##11#13")){Documenter.Documents.Page}) at /home/mortenpi/.julia/v0.7/Documenter/src/Utilities/Utilities.jl:597
 [2] runner(::Type{Documenter.Expanders.REPLBlocks}, ::Markdown.Code, ::Documenter.Documents.Page, ::Documenter.Documents.Document) at /home/mortenpi/.julia/v0.7/Documenter/src/Expanders.jl:502
 [3] dispatch(::Type{Documenter.Expanders.ExpanderPipeline}, ::Markdown.Code, ::Vararg{Any,N} where N) at /home/mortenpi/.julia/v0.7/Documenter/src/Selectors.jl:167
 [4] expand(::Documenter.Documents.Document) at /home/mortenpi/.julia/v0.7/Documenter/src/Expanders.jl:30
 [5] runner(::Type{Documenter.Builder.ExpandTemplates}, ::Documenter.Documents.Document) at /home/mortenpi/.julia/v0.7/Documenter/src/Builder.jl:178
 [6] dispatch(::Type{Documenter.Builder.DocumentPipeline}, ::Documenter.Documents.Document, ::Vararg{Documenter.Documents.Document,N} where N) at /home/mortenpi/.julia/v0.7/Documenter/src/Selectors.jl:167
 [7] #2 at /home/mortenpi/.julia/v0.7/Documenter/src/Documenter.jl:203 [inlined]
 [8] cd(::getfield(Documenter, Symbol("##2#3")){Documenter.Documents.Document}, ::String) at ./file.jl:70
 [9] #makedocs#1 at /home/mortenpi/.julia/v0.7/Documenter/src/Documenter.jl:202 [inlined]
 [10] (::getfield(Documenter, Symbol("#kw##makedocs")))(::NamedTuple{(:modules, :clean, :format, :assets, :sitename, :authors, :analytics, :linkcheck, :pages, :html_prettyurls, :html_canonical),Tuple{Array{Module,1},Bool,Symbol,Array{String,1},String,String,String,Bool,Array{Pair{String,Any},1},Bool,String}}, ::typeof(makedocs)) at ./<missing>:0
 [11] top-level scope
 [12] include at ./boot.jl:295 [inlined]
 [13] include_relative(::Module, ::String) at ./loading.jl:1060
 [14] include(::Module, ::String) at ./sysimg.jl:27
 [15] exec_options(::Base.JLOptions) at ./client.jl:332
 [16] _start() at ./client.jl:448
in expression starting at /home/mortenpi/Julia/Documenter/docs/make.jl:3

[ararslan]

Hm, somewhere it's using Core.stdout where it should be using Base.stdout, which is a LibuvStream and thus can be passed to redirect_stdout.

[JeffBezanson]

This is because binding deprecations only work with constants. Base.stdout changes at some point, and Base.STDOUT continues to point to the value it had when deprecated.jl ran.

The simplest fix is to also update the old names in reinit_stdio in libuv.jl. But of course that won't handle any other cases of changing the variables.

[JeffBezanson]

They are changed to acquire more functionality after bootstrap, and they're changed by the redirect_ functions.

[JeffBezanson]

Are you close here or should I make the remaining change?

[ararslan]

Sorry, been distracted with other things. If you know what needs to happen then go for it.

[JeffBezanson]

Will do.

[fredrikekre]

julia> print(STDOUT, "foo")
WARNING: Base.STDOUT is deprecatednothing
 in module Main
foo

[stevengj]

(Unfortunately, this PR is going to cause an huge number of deprecation warnings, and just switching to lowercase and doing using Compat will not be sufficient. People will need to put @compat decorators everywhere they use the uppercase names.)

[StefanKarpinski]

Why can't Compat export stdin, stdout, stderr and devnull as aliases? None of those names are exported or defined by Base in any earlier version of Julia. Compat does not generally provide the ability to use the old names of things on newer versions of Julia but rather the other way around.

[KristofferC]

Because functions like redirect_stdout rebinds the variables.

[StefanKarpinski]

That is annoying. I'm not sure what we can do about it though.

[KristofferC]

I thought about using getproperty-overloading but that is not available on 0.6 :(

[StefanKarpinski]

How about making stdout et al. mutable wrappers around an io object that just proxies IO calls through. Then the binding can be const and to redirect it, you just put a different IO object inside the wrapper. That definition could be put in Base and copied in Compat.

[mbauman]

Seems like we could have Compat override the Base._redirect_* methods that actually do the rebindings. They're totally untyped so we might even be able to do it without a warning with a ::IO annotation.

[ararslan]

This has the critically unfortunate consequence of conflicting with StatsBase.stderr.

[stevengj]

Note that the SciPy stats.stderr function was deprecated in favor of stats.sem. sem is pretty short, but it doesn't seem terrible to do some sort of renaming in StatsBase. I suppose it makes the deprecation process more painful, however.

[sbromberger]

Sorry for jumping in here so late, but what about Random.GLOBAL_RNG?

[KristofferC]

That is a constant so how does the discussion here apply?

[sbromberger]

ah, I see:

const GLOBAL_RNG = MersenneTwister(0)

Thanks.