Use of /sys/class/tty/console/active breaks OLPC splash

[Daniel Drake]

On Mon, Apr 9, 2012 at 12:41 PM, Ray Strode <halfline at gmail.com> wrote:
>> If I'm reading add_consoles_from_file() correctly, if
>> /sys/class/tty/console/active has any contents at all, the graphical
>> splash will be disabled due to the following line:
>>      state->should_force_details = true;
> Yea there's a bug there. We used to say "if there's any console= lines
> on kernel command line force details".
> When I copied that code over to read /sys/class/tty/console/active I
> didn't update the logic to account for
> /sys/class/tty/console/active always being available.
>
>> Also, there is a bug in the loop here:
>>
>>      console_length = strlen (console);
>>      ...
>>      remaining_command_line += console_length;
>>
>>
>> As strlen doesn't include the null-byte, remaining_command_line is not
>> fast-forwarded enough, meaning that the "tty0" console above is never
>> read.
> Interesting. It's more like we aren't skipping over the spaces between
> consoles.  But this whole fucntion suffers from being copy-and-pasted
> from the kernel command line reading function and not being properly
> fixed up afterward.
>
> I've pushed a few fixes, but...
>
>> However things are still not right - no graphical splash appears.
>> The fact that 2 text displays have been registered means that when we get here:
>>
>>  if (should_add_displays && ply_list_get_length (state->text_displays) == 0)
>>    add_default_displays_and_keyboard (state);
>>
>> since we have 2  text_displays we dont bother calling
>> add_default_displays_and_keyboard, which means that we never add any
>> pixel displays --> no splash.
>> Bit lost in the logic here; how is this supposed to work?
> That's a problem. You basically want this bug fixed:
>
> https://bugs.freedesktop.org/show_bug.cgi?id=22239

I may be missing something in the code, but I don't see how the
behaviour differs if there is only one non-serial console: one text
display (tty0) would be added from /sys/class/tty/console/active and
that means that add_default_displays_and_keyboard() is never called as
per the logic pasted above.

Anyway, going back to that bug entry, is there any chance you can work
on that soon, or can we look for alternative solutions? e.g. asking
plymouth not to look at the consoles file, or when working with the
limitation of either-details-or-graphical, base the decision on which
one to use on the primary console specified (i.e. the last one in the
active file)?
It is unfortunate that these changes have broken graphical splash for OLPC.

Thanks
Daniel