Re: Executing a file only on later passes

[Eric Sorenson]

On Mon, 3 Nov 2003, Chip Seraphine wrote:

> Had I some sort of 'laterpasses' thingie, I could ensure that cf.finish 
> always executed last.  I agree that this could cause some danger in that 
> the contents of cf.finish would not benefit from a multiple-pass 
> approach, but that should be a risk that the user should be willing to 
> shoulder if he goes out of his way to do so.  I.e., it is my foot so it 
> is my own fault if I shoot it on purpose; cfengine should only be 
> worried about preventing me from shooting it accidentally :-)

Hi Chip. 

I thought this was exactly what the "AddInstallable" control: definition
was for -- if C was in AddInstallable, and your cf.finish stanza referenced
C, cfengine would know that although C might not be defined at first parse,
it might be defined by the end of the run, and would defer the execution
of the relevant actions until the second pass, after those actions with
'define=C' parameters have had a chance to run.

Unless I'm totally wrong about that... Mark?
 
> Hmm.  Is it just me, or is all this talk of ordering starting tomake me 
> sound like Steve Traugott? :-)

By the end of last week, even Steve came to agree that order mattering was
a sign of some weakness in the toolset, not an intrisic property of 
all repeatable configuration actions. It's hard to read Luke's paper 
objectively (setting aside 'religious' bias) and not come to the same
conclusion.