About the new compiler, Part I

Philippe Marschall philippe.marschall at gmail.com
Mon Jan 21 06:00:53 UTC 2008

2008/1/20, Keith Hodges <keith_hodges at yahoo.co.uk>:
> > So.The Compiler is slower to compile code (factor ca. 4). But then,
> > this is not that of a problem. Most people
> > will now think about the slowness of loading code with MC, but that
> > has multiple causes, the compiler is not
> > the slowest thing here... (in 3.9, we managed to speed up code loading
> > with MC by a factor of two just by caching
> > the class category in the "category" variable of the class objec).
> MC will get much much faster in coming versions.
> Currently MC does not support Atomic loading. MC1.5 is slower than MC1
> because it iterates through all of the changes to be applied in several
> passes.
> MC1.5 has code in place ready to use SystemEditor to apply changes.
> Firstly This does not need to iterate through the changes as many times,
> and secondly the code is much simpler overall. This will be enabled in MC1.6
> I would not be at all surprised if we cannot gain back a factor of 4 or
> more from this.

The main problem not MC but PackageInfo. This is what kills MC
performance for bigger packages.


> I would like to extend MC to support binary loading in the future (1.7?)
> Keith

More information about the Squeak-dev mailing list