Make --quiet and --banner independent (fix #23380)

[StefanKarpinski]

This allows --banner={yes,no} to be set independently from --quiet, which, without an explicit --banner flag implies --banner=no. Adds tests for both options and their interaction.

[yuyichao]

I'm fine with either way but I'm not really convinced that we really need the --banner=* option that does basically the same thing as -q 99% of the time.......

[StefanKarpinski]

You may be right. Now we have both options available to merge.

[JeffBezanson]

I think a good option is to remove the terminal warning; then these two command-line options collapse to one.

[yuyichao]

remove the terminal warning

What's the likelihood that we'll have other REPL initialization warnnings? In any case though, since this is basically a REPL only option so the easiness to type in a terminal matters more (compare to other options) and so I would prefer an option with a shortform. -q naturally fill in that role thought sth like -bn can also work, if not a little bit strange....

[HarrisonGrodin]

This strategy seems logical, also remaining consistent with other languages. For example, IPython uses very similar flags, although their --no-banner is not inferred by --quiet. (I am in favor of that inference, as is included in this PR.)

--no-banner
    Don't display a banner upon starting IPython.
--quiet
    set log level to logging.CRITICAL (minimize logging output)

If logging is added into Base (JuliaLang/Juleps#30), the --quiet flag could become something similar to an alias for --logging=error, using the assumption that the banner is of level info or debug.

[ViralBShah]

Having -bn be a short form for --banner=no would be useful, as I often use -q myself. As @HarrisonGrodin says, freeing up -q and --quiet for future usage could be valuable too.

[StefanKarpinski]

Two-letter short options are a non-starter. With this PR, you can use -q to suppress the banner (and the one rare REPL warning). If --quiet is used in the future to suppress more than that, we could consider adding an abbreviation for turning the REPL off. I'm going to move on from this issue and someone else can decide which of the two perfectly viable directions we should go here.

[ViralBShah]

Since this does make -q work and keeps the --banner options, seems perfectly fine to merge this to me. For log levels, we can always use --logging= in full form, if that ever happens.

[JeffBezanson]

Ok, I'm fine with this. I buy the -q == "no banner plus log level" argument.

[ararslan]

Unless I'm missing something, did this PR make banner suppression the default?

[13:03:09] alex:julia git:(master) $ ./julia
julia> 

:/

[StefanKarpinski]

Oops. That was not the intention. Will fix.

[ararslan]

This might fix it:

diff --git a/base/client.jl b/base/client.jl
index 3f2f434bf9..b3600669d8 100644
--- a/base/client.jl
+++ b/base/client.jl
@@ -394,7 +394,7 @@ function _start()
                 term = Terminals.TTYTerminal(get(ENV, "TERM", @static Sys.iswindows() ? "" : "dumb"), STDIN, STDOUT, STDERR)
                 global is_interactive = true
                 color_set || (global have_color = Terminals.hascolor(term))
-                banner && REPL.banner(term,term)
+                (banner || is_interactive) && REPL.banner(term,term)
                 if term.term_type == "dumb"
                     active_repl = REPL.BasicREPL(term)
                     quiet || warn("Terminal not fully functional")

[StefanKarpinski]

Yeah, that probably works too.

[ararslan]

I think what you have in your PR is more correct. Thanks for jumping on that!

[StefanKarpinski]

Well, I did break it after all. We can't have the banner not printing. There has always been a banner.