Add option to run REPL after running a program

[s2maki]

Currently, running julia program.jl causes Julia to exit after the program finishes. For program development, it frequently makes sense to stay in the REPL after the program finishes. This pull request adds the -I option (--force-repl long option) to do this.

[jakebolewski]

How is this different than julia -i -L test.jl? I don't think we want to start introducing flags that just combine additional functionality.

[s2maki]

There are several reasons.

The -L option is meant to load modules since it calls require(), whereas the program argument is passed to include(). If this wasn't an important distinction, there wouldn't be a need for a -L option, either. It's also not possible to have multiple -L options, so if you needed -L and a program and to end in the REPL, there's presently no option for that.

The -L option is processed before almost any other option, including the load of .juliarc.jl and the creation of additional procs with -p or --machinefile. A program may need those resources.

Plus, it's more generally useful. It can be used in conjunction with --eval or --print, and therefore eliminates the need for -P, because -I -e <expr> could replace -P <expr>.

[JeffBezanson]

It sounds like -L is a bit broken; the help for it says the file is loaded on all processors. .juliarc.jl should definitely be loaded as early as possible. The order in client.jl doesn't seem right to me.

I agree the -e, -E, -P, and -L options are too complicated. All you actually need are -e and this PR's -I. Instead of debating whether to use require or include, you should use -e to get whichever you want.

[JeffBezanson]

I see --startup-file=yes also changes the order in which .juliarc is loaded. That's not at all obvious. Since our current way of handling options does not execute them left to right, you can't control the order so it should always be the same.

[jakebolewski]

It would be great to remove some of these redundant options.

[kmsquire]

#10713

[lstagner]

Couldn't there be a function that opens up the REPL from within a program. Something like repl()

[josefsachsconning]

Could someone please clarify what are the obstacles to merging this PR?

[ihnorton]

It would be best to fix the existing options rather than add a new one.

[josefsachsconning]

What exactly is the broken option? Should -i cause the REPL to be run? Is that what "Force isinteractive() to be true" in the --help output means? Is there an existing issue for that?

[ihnorton]

Right now -i just means "make isinteractive() return true" (see #6765). I think -i should mean "run-to-REPL" by default, like Python and bash. I don't believe this would conflict with IJulia or Gtk.jl/Julietta usage because they take over the event loop, but if it is a problem we could probably check whether isatty to decide.