fix(github_actions): Fix clients lingering into subsequent tests

On GitHub Actions, clients created in some tests take too long to close
and still exist when the next test has started. This can affect certain
tests that rely on a specific client count or simply produce a race
condition within `scan` in `awesome.c`.

Signed-off-by: Lucas Schwiderski <lucas@lschwiderski.de>

tests/run.sh

@@ -185,6 +185,20 @@ fi
 start_awesome() {
     cd "$build_dir"
 
+    # On some systems clients from a test may still linger for a while until they
+    # are fully killed. Since this can affect susequent tests, we wait until all
+    # of them are gone.
+    # The check command needs to call `xlsclients` twice:
+    # 1. Test the stdout of `xlsclients` and make sure there are no clients left.
+    # 2. Check and fail if `xlsclients` itself fails. Print its stderr, if any.
+    # This is due to the fact that, if `xlsclient` return non-0, the `wc` would
+    # still report `0`. The order of commands makes sure that `xlsclient` is only
+    # called twice when needed. If there are still clients left, the `[` will
+    # shotcurcuit.
+    wait_until_success \
+        'wait for X clients from previous test to close' \
+        "[ \$(DISPLAY="$D" xlsclients | wc -l) -eq 0 ] && DISPLAY="$D" xlsclients > /dev/null"
+
     # Kill awesome after $TEST_TIMEOUT seconds (e.g. for errors during test setup).
     # SOURCE_DIRECTORY is used by .luacov.
     DISPLAY="$D" SOURCE_DIRECTORY="$source_dir" \

tests/test-leak-client.lua

@@ -85,12 +85,25 @@ local steps = {
             -- Test that we have a client and that it's invalid (tostring()
             -- causes an "invalid object" error)
             local success, msg = pcall(function() tostring(objs[1]) end)
-            assert(not success)
+            assert(not success, msg)
             assert(msg:find("invalid object"), msg)
 
             -- Check that it is garbage-collectable
             collectgarbage("collect")
-            assert(#objs == 0)
+
+            -- On GitHub Actions, it can take a while for clients to be killed
+            -- properly.
+            local tries = 0
+            while (#objs > 0 and tries < 60) do
+                os.execute("sleep 1")
+                tries = tries + 1
+            end
+
+            if tries > 0 then
+                print("Took approx. " .. tries .. " seconds to clean leaked client")
+            end
+
+            assert(#objs == 0, "still clients left after garbage collect")
             return true
         end
     end,