Use serial_terminal to get a stable ghostscript

[Zaoliang]

select_serial_terminal should be stable and resolves the sporadic
typing issue.
see https://progress.opensuse.org/issues/87674
verification:
http://10.160.64.152/tests/52#step/ghostscript/

[okurz]

This won't scale. How do you envision our tests should look like if you sprinkle in type_string_slow in a specific case but not in general? I doubt this approach is sustainable.

If you see "performance problems" on specific workers then debug the specific workers. If you see stability problems in specific test modules then the test module should be designed accordingly. If you see a problem with the overall test flow or scenario then stability should be ensured on a higher level. For well-known infrastructure related limitations VNC_TYPING_LIMIT can be used.

[Zaoliang]

VNC_TYPING_LIMIT is set already for 6, and I checked 4, it works though, but I don't want slow down whole tests.
DVD-installation has less performance, this is what experienced in past.

[mdoucha]

Running the setup commands in serial terminal and switching to VNC only to run gv and check needles would be much more reliable solution. Call $self->select_serial_terminal; to select the best available serial backend.

[okurz]

VNC_TYPING_LIMIT is set already for 6, and I checked 4, it works though, but I don't want slow down whole tests.
DVD-installation has less performance, this is what experienced in past.

I did not suggest to use VNC_TYPING_LIMIT in this specific case. Either find out what is causing trouble in the specific case and fix that or follow the advice from @mdoucha

[mdoucha]

Some comment cleanup and new verification run would be nice. Other than that, looks good to me.

[Zaoliang]

this is not working yet:

http://10.160.64.152/tests/29#step/ghostscript/34

[mdoucha]

Oh, right. The test PDF file was written to /root/ but gv is looking for it in ~bernhard/. Replacing the typestring "exit\n"; on line 45 with $self->select_serial_terminal(0); will fix it.

[Zaoliang]

$self->select_serial_terminal(0); failed :
see http://10.160.64.152/tests/35/file/autoinst-log.txt

�[0m[2021-01-25T14:06:53.271 CET] [debug] tests/x11/ghostscript.pm:45 called opensusebasetest::select_serial_terminal -> lib/opensusebasetest.pm:1243 called testapi::select_console
[2021-01-25T14:06:53.271 CET] [debug] <<< testapi::select_console(testapi_console="virtio-terminal")
console virtio-terminal does not exist at /usr/lib/os-autoinst/backend/driver.pm line 86.
�[33m[2021-01-25T14:06:53.319 CET] [info] ::: basetest::runtest: # Test died: Can't call method "select" on an undefined value at /usr/lib/os-autoinst/backend/baseclass.pm line 667.

[mdoucha]

Oh, great, somebody forgot to define a non-root virtio console... Never mind, you can just switch directory to '~'.$testapi::username instead. That'll also place the test PDF in the right directory.

[okurz]

I still suggest to run tests itself after the preparation as the non-priviledged user and keep that part as before.

[mdoucha]

I still suggest to run tests itself after the preparation as the non-priviledged user and keep that part as before.

Yes, that would be nice, but it's currently not possible without workarounds like su bernhard because non-root virtio terminal is missing.

[okurz]

sure, I would be fine with su $username, can we do that? Maybe even in a separate bash process within

[mdoucha]

sure, I would be fine with su $username, can we do that? Maybe even in a separate bash process within

That pointless user switch is now messing up command prompt detection and slowing down the test to a crawl:
https://openqa.suse.de/tests/5350686#step/ghostscript/13

[Zaoliang]

sure, I would be fine with su $username, can we do that? Maybe even in a separate bash process within

That pointless user switch is now messing up command prompt detection and slowing down the test to a crawl:
https://openqa.suse.de/tests/5350686#step/ghostscript/13

please provide suggestion here, if you are not happy with these changes from Oliver or yourself for user switch, I really don't understand your guys what you're expecting.

[mdoucha]

please provide suggestion here, if you are not happy with these changes from Oliver or yourself for user switch, I really don't understand your guys what you're expecting.

Sorry, my last comment was for Oliver. The test will work perfectly fine with just changing the directory and generating the PDF as root. Otherwise LGTM.

[foursixnine]

Following @mdoucha's last suggestion:

Suggested change

	type_string "su $username\n";
	type_string "cd --\n";
	my $working_directory = script_output("mktemp -d -t GHOSTSCRIPTXXX");
	type_string "$working_directory\n";

[foursixnine]

Later before changing to x11, do:

script_run "chmod -R $username $working_directory"

[foursixnine]

once xterm is open, cd $working_directory

[mdoucha]

That's quite a lot of overkill. Simple assert_script_run('cd ~' . $testapi::username); will suffice. The rm -f at the end will remove files owned by root just fine. Once the missing virtio terminal gets added, we can simply swap the cd for select_user_serial_terminal().

[Zaoliang]

thanks for your input!