Cairo exports and more (was Re: [cairo] Re: Munging header files for export (and other) attributes)

[Bill Spitzak]

Alexander Larsson wrote:

> This method is called double checked locking, and it doesn't in general
> work for all hardware architectures. See e.g. 
> http://www.cs.umd.edu/~pugh/java/memoryModel/DoubleCheckedLocking.html
> 
> glib uses this on architectures where it works. Looking at its
> configure.in it seems it doesn't work on old sparcs, alpha, ppc and
> ia64.

If it works on all platforms that are physically capable of compiling 
and using Cairo then this is ok. Or you can ifdef it out and force the 
lock on all tests if it does not work. Of since this only has to be in 
the backend_surface_create function it is probably ok if it slows things 
down by locking every time it is called (it is likely it has to lock 
something anyway...).

I hope this does not detract from my initial reason for posting this, 
which is:

Please do not put "initialization" into the Cairo API! There is an 
obvious reason. If you do, all that is going to happen is that people 
like me will erroneously try to call it with the wrong locking, and 
screw up. The fact that my code is technically wrong should be PROOF 
that you should not provide a mechanism to screw up like this.