Make GUI-only Windows binary more discoverable

[tleedjarv]

Another fix for #778.

The following normally text-only output is now handled differently by the GUI-only binary:

• -version output is displayed in a GUI message box.
• -doc produces a generic message explaining the GUI-only binary.
• -ui text produces a generic message explaining the GUI-only binary.
• -help produces a generic message explaining the GUI-only binary.
• command line parsing errors are displayed in a GUI message box.
• GTK init errors are displayed in a GUI message box.

@rivy could you also test and see how this fits into your workflow.

[acolomb]

My only concern here is about errors involving the GUI system itself, such as when GTK cannot be initialized. I think those are dumped to the console either way, and obviously a GUI window for those is impossible.

Could we just pop up a console window in that last case? I'd assume it's something like GApplication failing to initialize or other error return codes from early GTK functions.

[tleedjarv]

My only concern here is about errors involving the GUI system itself, such as when GTK cannot be initialized. I think those are dumped to the console either way, and obviously a GUI window for those is impossible.

Could we just pop up a console window in that last case? I'd assume it's something like GApplication failing to initialize or other error return codes from early GTK functions.

Will give it a try. Don't know yet how to test that, though.

I don't think it's going to be too useful for a user because there will be no meaningful error message. lablgtk uses the gtk_init_check() function https://docs.gtk.org/gtk3/func.init_check.html and this only returns FALSE on errors, without actual error message. But it might still be better than just "silently crashing". At least the user knows that gtk init failed, even though it's impossible to tell why.

[tleedjarv]

Now GTK init errors are also displayed in a message box.

[acolomb]

Message box provided by GTK? How is that supposed to work when the first initialization fails?

Could we instead detect when something is sent to the standard output stream, and open a console window unconditionally of what part produced the output?

As for testing GTK errors, you might have luck setting some invalid configuration via environment variables. GDK_BACKEND comes to mind for example, which should make it impossible to open any GUI when using an invalid value. See here for some hints: https://jan.newmarch.name/Wayland/GTK/

[tleedjarv]

Message box provided by GTK? How is that supposed to work when the first initialization fails?

No, of course not. None of this PR uses GTK. If that was feasible or possible then we wouldn't need this PR in the first place. This is Windows only, it uses the Windows native GUI (just see in the code, it's very short).

Could we instead detect when something is sent to the standard output stream, and open a console window unconditionally of what part produced the output?

Not feasible.

[acolomb]

Alright, that sounds fair. Sorry I didn't have a chance to look at the code changes while traveling.

[tleedjarv]

As for testing GTK errors, you might have luck setting some invalid configuration via environment variables. GDK_BACKEND comes to mind for example, which should make it impossible to open any GUI when using an invalid value.

Thanks for the hint @acolomb. This worked (to make GTK init fail). Now, a native Windows message box is displayed with the error. Note that this is strictly an improvement because previously (since always) the error would have been displayed in the console window but the console window would be closed immediately with application exit (unless run from command line) and the user wouldn't see anything.