[PEAK] A few issues and suggestions

[Phillip J. Eby]

I don't see from this example why .c's initialization *shouldn't* be 
undone, unless you're saying that the make() isn't also being 
undone...  in which case the test should be about that.

It seems to me that the real issue here is that CellAttribute and 
TodoProperty don't undo-log their addition of a cell to a Component's 
__cells__.  I didn't originally do this because then the Cells() 
addon would need to support locking to be able to be threadsafe in 
the future.  But I suppose that bridge can be crossed later.

Is there any other reason besides this inconsistency, that the 
component __init__ shouldn't be undone?

(Btw, the same problem occurs with cells in addons and services, and 
won't be fixed by undo-logging the cell creation.  It's a big open 
issue at the moment, since even avoiding an __init__ doesn't 
help.  In essence, undo logging and "write-until-read" consistency 
simply don't mix, and  Addons, services, and the Cells() map all run 
off the write-until-read model.)

At 05:09 AM 11/4/2008 +0200, Sergey Schetinin wrote:
>Here's the test case:
>
>class C(Component):
>     x = attr(0)
>     def __init__(self):
>         self.x = 10
>         @on_undo
>         def undo():
>             print 'UNDO'
>
>
>class TC(Component):
>     c = make(C)
>     x = attr(0)
>     y = attr(0)
>
>     @maintain
>     def test(self):
>         if self.x:
>             self.c
>             self.y
>
>     @maintain
>     def write(self):
>         self.y = self.x
>
>
>tc = TC()
>
>@atomically
>def test():
>     tc.x = 1
>     tc.test
>
>print tc.c.x
>
># prints UNDO \ 0
>
>
>
>
>
>On Tue, Nov 4, 2008 at 04:51, Sergey Schetinin <maluke at gmail.com> wrote:
> >> Thanks for your help identifying and fixing the problems.  I've checked in
> >> fixes, as described here:
> >>
> >> http://svn.eby-sarna.com?rev=2595&view=rev
> >
> > Thanks a lot. However, I see that some problems, that I thought was
> > related to my monkeypatching, didn't go away, what happens in effect
> > is that changes done Component.__init__ get rolled back. I'm working
> > on a test case for this, and I've noticed something I didn't think
> > about before. In the following example __init__ reads data it
> > shouldn't. This isn't a problem normally, but something to keep in
> > mind when writing components' __init__
> >
> > class TC(Component):
> >    y = attr(0)
> >
> >    def __init__(self):
> >        print self.read # 0
> >
> >    @compute
> >    def read(self):
> >        return self.y
> >
> >    @maintain
> >    def write(self):
> >        self.y = 5
> >
>
>
>
>--
>Best Regards,
>Sergey Schetinin
>
>http://s3bk.com/ -- S3 Backup
>http://word-to-html.com/ -- Word to HTML Converter