Is delwin(stdscr) legal?

That's multiple questions:

• If it is legal, should I call this always when exiting from ncurses?
• Or is it automatically called from endwin()?
• What are the consequences of doing delwin(stdscr)?

The first two are equivalent (it's not necessary in any implementation of curses, and not something that endwin does). Call that two questions, then. The first asks how delwin and endwin are related, and the other asks about specific behavior of delwin. The latter is interesting.

Actually it's probably implementation-dependent.

Limiting it to ncurses (per tag),

• delwin calls
• _nc_freewin, which calls
• remove_window_from_screen, which makes a special check for each of curscr, newscr and stdscr (setting the value of those global variables to NULL if the window to be deleted happens to match, before deallocating their respective data).

So it works for ncurses. I added the feature in December 1996:

    + modify _nc_freewin() to reset globals curscr/newscr/stdscr when
      freeing the corresponding WINDOW (found using Purify).

Regarding portability:

• PDcurses does not do this.
• NetBSD does (but see this — from 2006, likely imitating ncurses)
• Solaris SVr4 curses does not, nor does its X/Open (xpg4) flavor.

There were two questions. The other regards endwin. That never deletes a window, because it is always possible to follow a call to endwin with a call to wrefresh. It's in the manual page:

   Calling refresh(3x) or doupdate(3x) after a temporary escape causes the
   program to resume visual mode.

However, if you did a delwin(stdscr) between the endwin() and wrefresh(stdscr) calls, ncurses would return an error for the latter. NetBSD (read the code to see that it will dereference a null pointer) and Solaris would both dump core.