Add some tests.

regress/cmd-client-argv.sh

@@ -0,0 +1,50 @@
+#!/bin/sh
+
+# Client argv parsing and the start-server boundary.
+#
+# The command line given to the client is parsed and invoked as one sequence.
+# This checks: a multi-command argv runs every command and starts the server
+# (new-session carries CMD_STARTSERVER); a command without start-server fails
+# cleanly against no server; and a parse error leaves no stray server behind.
+
+PATH=/bin:/usr/bin
+TERM=screen
+
+[ -z "$TEST_TMUX" ] && TEST_TMUX=$(readlink -f ../tmux)
+TMUX="$TEST_TMUX -Ltest -f/dev/null"
+$TMUX kill-server 2>/dev/null
+
+# Multi-command argv: starts the server and runs all commands in order.
+$TMUX new-session -d -swork -nbase \; set -g @a one \; new-window -dn W2 \; \
+	set -g @b two || exit 1
+[ "$($TMUX show -gv @a)" = one ] || { echo "argv cmd 2 did not run" >&2; exit 1; }
+[ "$($TMUX show -gv @b)" = two ] || { echo "argv cmd 4 did not run" >&2; exit 1; }
+got=$($TMUX list-windows -t work -F '#{window_name}' | tr '\n' ',')
+[ "$got" = "base,W2," ] || { echo "argv windows: got [$got]" >&2; exit 1; }
+# kill-server is asynchronous; wait for the server to actually exit before the
+# checks below that require no server to be running.
+$TMUX kill-server 2>/dev/null
+i=0
+while $TMUX has-session 2>/dev/null; do
+	sleep 0.05
+	i=$((i + 1))
+	[ $i -gt 100 ] && break
+done
+
+# A command without start-server fails cleanly when no server is running and
+# does not fork one.
+$TMUX new-window -dn ZZ 2>/dev/null && { echo "new-window unexpectedly succeeded" >&2; exit 1; }
+if $TMUX has-session 2>/dev/null; then
+	echo "non-start-server command left a stray server" >&2
+	exit 1
+fi
+
+# A parse error with no server running starts no server.
+$TMUX 'if-shell true {' 2>/dev/null && { echo "parse error did not fail" >&2; exit 1; }
+if $TMUX has-session 2>/dev/null; then
+	echo "parse error left a stray server" >&2
+	exit 1
+fi
+
+$TMUX kill-server 2>/dev/null
+exit 0