On Mon, Jan 25, 2010 at 05:13:40PM -0800, Danek Duvall wrote:
> [email protected] wrote:
> 
> > In general this looks good.  I have a meta-comment about the code in
> > imageplan around line 641.  I'm assuming that the action order doesn't
> > have to be deterministic, but it just cannot change during the remainder
> > of the evaluate step.  While this isn't a problem now, we might want to
> > observe that adding install actions, or changing the order of the actions
> > in the package plan after this point will lead to very confusing and hard
> > to debug results.
> 
> Very true.  The code here has been pretty stable, so I'm not too worried
> about it, and it already has been the case that order mattered (with the
> original_name handling).  On the other hand, with the solutions Bart's
> talking about to reduce the ginormous memory consumption seen with
> operations on redistributable, this may eventually become an issue.
> 
> Are you suggesting I add a comment towards the top warning folks that
> maintaining order in the package plans is crucial to correct operations?
> Or are you suggesting that I should find a better way to handle the
> removals?  I tried decorating the actions with the plans they belonged to,
> but that was a nightmare in terms of memory usage.  I imagine that other
> object decoration methods would suffer similarly.

Mostly I'm suggesting the former.  A comment with a dire warning would
be sufficient.

I wondered if you could assert that a package plan's actions have not
changed during this step, but if there's no easy way to do this then I
wouldn't worry about it.

-j
_______________________________________________
pkg-discuss mailing list
[email protected]
http://mail.opensolaris.org/mailman/listinfo/pkg-discuss

Reply via email to