[maemo-developers] Automatic optification (was: Re: Proposal: Karma-Whores protection mailing list)

From: Graham Cobb g+770 at cobb.uk.net
Date: Tue Jan 5 00:41:55 EET 2010
On Monday 04 January 2010 12:11:46 Marius Vollmer wrote:
> ext Ed Bartosh <bartosh at gmail.com> writes:
> > I'll definitely find a time to do whatever is needed. Moreover, I was
> > asking couple of times already if it's time to enable optification by
> > default in autobuilder. I was given an answer that some testing is
> > still needed. I think Marius should know the latest status.
> I still have to do something about the Python optification.  I will do
> that in the next few days.  

Thanks.  We really need this in order to turn on optification by default.

> The 'something' will likely be some way to 
> detect the relevant packages and to stop optifying them in auto mode.
> (Indirect dependencies are a bit expensive to follow, so my current idea
> is that I go with a list of direct dependencies instead.)

The sooner we can try it, the sooner we can find out if it works!

> My current idea is that we will have a rule that takes in ratios: You
> need to have 20 times as many bytes (uncompressed) on the eMMC than on
> the OneNAND.  The idea with this is that when all packages conform to
> this, your will most probably run out of eMMC before you run out of
> OneNAND.

Can you explain that in more detail?  Do you mean that you will optify (move) 
smaller and smaller files until you get to the right ratio and then stop?  
What if you can't get to the right ratio -- is that a failure?  Are you still 
planning to allow the developer to override decisions?

Do we actually need that level of complexity?  I really don't like the idea of 
the optification changing between releases just because one file has changed 
size -- for example, I wouldn't want the developer to find that one of their 
files (e.g. a binary) has suddenly been moved because a different file (e.g. 
a text file) has changed size.  That is asking for creating unexpected bugs.

I would prefer to keep the algorithm the same as we have now, as we have got 
some experience with that, and turn on automatic optification with that 
algorithm.  We can look at improving it in the future but otherwise it will 
be another 6 months before we are ready to turn on automatic 
optification!  "The best is the enemy of the good"!


More information about the maemo-developers mailing list