ncurses: fixed batch mode initialization for unset or broken $TERM

 * If TERM is set but wrong, it might still fail. Errors are reported now.
 * If $TERM is unset or empty, we refuse to start the interactive mode.
   Unfortunately, we cannot easily let curses decide this since
   newterm() does not work as expected.
 * However this solution should be quite good: Either $TERM is empty - then
   batch mode will work but not interactive mode.
   Or it is not empty, then it is checked by the first call to newterm()
   when initializing batch mode.
   The chance of starting up interactive mode with a wrong $TERM
   or in head-less environments is now quite low.
 * This is most likely responsible for the PPA build issues.

1 files changed, 56 insertions, 10 deletions

diff --git a/src/interface-curses.cpp b/src/interface-curses.cpp
index d12aed4..5f8a00c 100644
--- a/src/interface-curses.cpp
+++ b/src/interface-curses.cpp
@@ -178,6 +178,23 @@ InterfaceCurses::init_interactive(void)
 void
 InterfaceCurses::init_batch(void)
 {
+	const gchar *term = g_getenv("TERM");
+
+	/*
+	 * In headless or broken environments,
+	 * $TERM may be unset or empty.
+	 * Still batch-mode operation is supposed
+	 * to work.
+	 * Therefore we initialize Curses with
+	 * a terminal type that ensures it will at least
+	 * start up ("ansi" is in ncurses-base).
+	 * We do not always use "ansi" since
+	 * it is also not guaranteed to work and we cannot
+	 * change it later on.
+	 */
+	if (!term || !*term)
+		term = "ansi";
+
 	/*
 	 * This sets stdscr to a new screen associated
 	 * with /dev/null.
@@ -186,26 +203,55 @@ InterfaceCurses::init_batch(void)
 	 * Curses (required by Scintilla and thus for
 	 * batch processing).
 	 */
-	screen_tty = g_fopen("/dev/null", "r+b");
+	screen_tty = g_fopen("/dev/null", "r+");
+	/* should never fail */
 	g_assert(screen_tty != NULL);
+
 	setvbuf(screen_tty, NULL, _IOFBF, 0);
-	screen = newterm(NULL, screen_tty, screen_tty);
+
+	screen = newterm(term, screen_tty, screen_tty);
+	if (!screen) {
+		/* $TERM may be set but to a wrong value */
+		g_fprintf(stderr, "Error initializing batch mode. "
+		                  "$TERM may be incorrent.\n"
+		                  "Try unsetting it if you do not need "
+		                  "the interactive mode.\n");
+		exit(EXIT_FAILURE);
+	}
+
 	def_shell_mode();
 }
 
 void
 InterfaceCurses::init_interactive(void)
 {
+	const gchar *term = g_getenv("TERM");
+
+	/* at least try to report a broken $TERM */
+	if (!term || !*term) {
+		g_fprintf(stderr, "Error initializing interactive mode: "
+		                  "$TERM is unset or empty.\n");
+		exit(EXIT_FAILURE);
+	}
+
 	/*
-	 * To initialize interactive mode, we
-	 * reopen the file opened for Curses in init_batch()
-	 * with the current controlling terminal device
-	 * (see tty(4)).
-	 * Should also work with newterm(), but this
-	 * seems more elegant.
+	 * Reopen screen_tty on the real terminal
+	 * device.
+	 * NOTE: It would be better to create a new
+	 * terminal with newterm() since the current
+	 * terminal might still be configured as
+	 * "ansi" to get the batch mode working.
+	 * If this is the case because we are in a head-less
+	 * or broken environment NOW would be the time
+	 * to tell the user.
+	 * However, I cannot get ncurses to switch
+	 * to a new terminal. Perhaps existing windows are
+	 * somehow "bound" to the current screen.
+	 * Perhaps this only works if we delete and recreate
+	 * ALL windows...
 	 */
-	if (!g_freopen("/dev/tty", "r+b", screen_tty)) {
-		/* no controlling terminal, process detached? */
+	if (!g_freopen("/dev/tty", "r+", screen_tty)) {
+		/* no controlling terminal? */
 		g_fprintf(stderr, "Error initializing interactice mode: %s\n",
 		          g_strerror(errno));
 		exit(EXIT_FAILURE);