Brock Pytlik wrote:
> I understand all that. However, the gate tip throws an exception during 
> plan_creation, and Bart's wad changes that behavior. I want to 
> understand why that was changed, as I think I have with the issue of 
> error = 1 not being set in the except block when something is added to 
> illegal_fmri. I don't object to the change to check the return value. I 
> do want to make sure we understand why test suite behavior is changing 
> when it's not expected to.

I can't comment on the error = 1 part because I don't quite follow 
everything being changed here.

However, it seems wrong to me that *any* consumer of the api would 
ignore the return value of plan_install(), etc. if those are normally 
supposed to be check given that plan_execute() doesn't check for 
nothingtodo() at the beginning.

> That means this can only blow up in the event that it's called with 
> packages that are already up to date or cannot be installed. My personal 
> belief is that for this testing setting, those tests shouldn't use 
> _do_install and should instead do plan_install manually and explicitly 
> check the nothingtodo flag in the imageplan. So, my slight preference 
> would be for undoing all the changes in this module. I don't feel that 
> strongly about it, so if others want to make the change on lines 
> 152-154, I won't object.

I see both sides of this and I'm not certain which is the right answer.

One change I do think is missing here is that plan_execute() should 
either return if nothingtodo() or have an assert not nothingtodo().

It seems wrong that plan_execute() would blithely go ahead and execute 
if there is nothingtodo().

Cheers,
-- 
Shawn Walker
_______________________________________________
pkg-discuss mailing list
[email protected]
http://mail.opensolaris.org/mailman/listinfo/pkg-discuss

Reply via email to