Now that we have thousands of modelines in the xdummy config (r16160), the server can take too long to startup. ie: r20970 This is more of a problem on low-power devices: #1777
This was recorded in ticket:1557#comment:11: This makes the Xdummy server a little bit slower to start, so r16252 increases the timeout used in the automated tests.
Looking at the Xorg server code, there is a 3-level nested loop on modelines in xf86ValidateModes
.
The middle loop reads: If the supplied mode names don't produce a valid mode, scan through unconsidered modePool members until one survives validation. This is done in decreasing order by mode pixel area.
The innermost loop, never actually triggers anything, its docstring says: Deal with the case where this mode wasn't considered because of a builtin mode of the same name
With a thousand modeline entries, we can end up running the strcmp
millions of times...
We should be able to re-work that code to achieve the same results with a lot less CPU time. On a decent CPU this can take multiple seconds, it should be doable under 100ms.
This also explains why the unit tests have been failing occasionally.
Re-scheduling.
Less of a problem recently? Did this get fixed or improved already?
See also #1467
Less of a problem as Debian builds have reverted to Xvfb: #2834
this ticket has been moved to: https://github.com/Xpra-org/xpra/issues/2091