Hmm, the test is failing now.

Maybe it wasn't running before.

https://github.com/vim/vim/actions/runs/11827146010/job/32954690492?pr=16047#step:18:3418

Failing with ioctl?

Consider how to identify the problem.

I added call ch_logfile('Test_getcellpixels.log', 'w') to the test code to check if ioctl was failing.

" Test for getcellwidth()
" Pixel size of a cell is terminal-dependent, so in the test, only the list and size 2 are checked.
func Test_getcellpixels()
" Not yet Windows-compatible
CheckNotMSWindows
call ch_logfile('Test_getcellpixels.log', 'w')
let cellpixels = getcellpixels()
call assert_equal(2, len(cellpixels))
endfunc

As a result, it was found that ioctl failed.

$ head -n 2 testdir/Test_getcellpixels.log
==== start log session Wed Nov 13 23:47:25 2024 ====
0.000002764 : ioctl(TIOCGWINSZ) failed

In Vim's test, ioctl doesn't work?

How do we test this?

RunVimInTerminal?

Oh, out of habit I was running it inside the Vim terminal.
I'll try running it again outside Vim.

Sorry, What mean RunVimInTerminal?

You can see Test_state() for a reference :)

Thank you. I maybe understand.

RunVimInTerminal?

RunVimInTerminal is an interesting situation because you are testing two things here. In addition to getcellpixels() in the Vim you are testing, you are also testing that the parent Vim is correctly passing the size info to the child Vim process to be tested.

In particular, the code here is doing the work of passing that info in to the Vim embedded terminal (and is what would end up being tested):

    int
mch_report_winsize(int fd, int rows, int cols)
{
    //...
    // calcurate and set tty pixel size
    struct cellsize cs;
    mch_calc_cell_size(&cs);
    ws.ws_xpixel = cols * cs.cs_xpixel;
    ws.ws_ypixel = rows * cs.cs_ypixel;

    retval = ioctl(tty_fd, TIOCSWINSZ, &ws);
    //...
}

It gets called to report the window size info to the embedded terminal in Vim's terminal.c (fd here is the fd for the embedded tty). Before the last PR, it used to be hard-coded to report a 5x10 pixel size which was not accurate. Now it's doing the right thing and passing the correct values along.

However, in this situation when the ioctl call in the parent Vim fails, it is going to set cs_xpixel to -1, which actually leads to kind of funky results in the above code as you get cols * -1 and rows * -1 which leads to some weird negative numbers. One way to fix it is to use a pre-canned 5x10 in this situation, or just use 0.

You can see this behavior just running Vim within Vim and then calling echo getcellsize() inside the child Vim.

@ychin Thank you for your comment.

One way to fix it is to use a pre-canned 5x10 in this situation, or just use 0.

Using a value of 5x10 when returning the value could lead the user to mistakenly believe that the result was correct, so we thought it would be better to change it to 0x0 (returning [0, 0]).

I will fix it.

diff --git a/src/os_unix.c b/src/os_unix.c
index 56bbc868d..80ca0ce26 100644
--- a/src/os_unix.c
+++ b/src/os_unix.c
@@ -4434,8 +4434,18 @@ mch_report_winsize(int fd, int rows, int cols)
     // calcurate and set tty pixel size
     struct cellsize cs;
     mch_calc_cell_size(&cs);
-    ws.ws_xpixel = cols * cs.cs_xpixel;
-    ws.ws_ypixel = rows * cs.cs_ypixel;
+
+    if (cs.cs_xpixel == -1)
+    {
+        // failed get pixel size.
+        ws.ws_xpixel = 0;
+        ws.ws_ypixel = 0;
+    }
+    else
+    {
+        ws.ws_xpixel = cols * cs.cs_xpixel;
+        ws.ws_ypixel = rows * cs.cs_ypixel;
+    }
-
     retval = ioctl(tty_fd, TIOCSWINSZ, &ws);
     ch_log(NULL, "ioctl(TIOCSWINSZ) %s", retval == 0 ? "success" : "failed");