Thanks. But could you please also add those lua tests from the previous PR

Those failures are most likely a result of unexpected defaults in the Vim tests. E.g. Vim doesn't show a statusbar by default, so that's where [No Name] comes from.

Should be OK to :set laststatus=1 in testdir/setup.vim . We didn't run into this until now because Vim has few screen tests.

@janlazo The cause of #9338 (comment) seems to be that msg_use_printf() returns true in headless mode. In my environment, applying the following patch will pass that test. However, other tests have not been verified, so there may be something wrong.

diff --git a/src/nvim/message.c b/src/nvim/message.c
index b4aa333a4..93c2675fd 100644
--- a/src/nvim/message.c
+++ b/src/nvim/message.c
@@ -2289,7 +2289,7 @@ static void t_puts(int *t_col, const char_u *t_s, const char_u *s, int attr)
 //    - no UI and not embedded
 int msg_use_printf(void)
 {
-  return !embedded_mode && !ui_active();
+  return !embedded_mode && !headless_mode && !ui_active();
 }

 /// Print a message when there is no valid screen.

I read :h --headless, the #9338 (comment) change is not desirable. Therefore, do not use headless mode with oldtest, or in the case of the headless mode, the :echo output needs to be changed to the screen and stderr.

Therefore, do not use headless mode with oldtest, or in the case of the headless mode, the :echo output needs to be changed to the screen and stderr.

@erw7 what do you mean by "changed to the screen and stderr"?

--headless outputs to stderr:

$ 2>/dev/null nvim --headless +'echo "foo"' +q
$ 1>/dev/null nvim --headless +'echo "foo"' +q
foo

what do you mean by "changed to the screen and stderr"?

@justinmk It is to change it like the following patch.

diff --git a/src/nvim/message.c b/src/nvim/message.c
index a83c9cf05..20d0807f4 100644
--- a/src/nvim/message.c
+++ b/src/nvim/message.c
@@ -1810,8 +1810,13 @@ void msg_puts_attr_len(const char *const str, const ptrdiff_t len, int attr)
   // different, e.g. for Win32 console) or we just don't know where the
   // cursor is.
   if (msg_use_printf()) {
+    int saved_msg_col = msg_col;
     msg_puts_printf(str, len);
-  } else {
+    if (headless_mode) {
+      msg_col = saved_msg_col;
+    }
+  }
+  if (!msg_use_printf() || headless_mode) {
     msg_puts_display((const char_u *)str, len, attr, false);
   }
 }

Ah, I think I understand: in order to make screenchar() work, headless-mode must call msg_puts_display() so that the internal screen is updated.

We could try that...

Updated after #10109 . Hoping to merge after CI.