Re: improvements for parallel makes ?

[Alexey Neyman]

On Friday 21 April 2006 12:11, Boris Kolpackov wrote:
> "Paul D. Smith" <address@hidden> writes:
> 
> > In order to really get at what .WAIT is intended to do you'd have 
to
> > convert the original to this:
> >
> >     foobar: a b c D E F
> >
> >     D: X
> >     E: Y
> >     F: Z
> >
> >     D E F X Y Z : | a b c
> >
> > and so on for the entire dependency tree rooted at D, E, and F.  
> > Which could be done, but is not that trivial and may not be the
> > best way to proceed in terms of efficiency etc.
> 
> After reading this fragment, I had the following idea: imagine we
> had: (1) support for guile, (2) support for accessing/modifying
> dependency tree from guile, (3) support for truly order-only
> prerequisites. Then the .WAIT feature could be implemented in
> "user-space" ;-). 

Almost. The only thing is that we need to apply implicit rules early 
in order to decide, whether a target is going to be built at all. GNU 
make applies them "lazily", that is, delaying the search for implicit 
rule until it really needs it.

BTW, I thought about a less intrusive (from the point of code changes) 
approach. update_goal_chain() can first perform a "dry run", looking 
for implicit rules and marking the targets as "scheduled to be made". 
Then, it will perform a "normal" run and update_file_1() can then 
wait only for those prerequisites that are to be made.

Regards,
Alexey.

-- 
A quoi ca sert d'etre sur la terre
Si c'est pout faire nos vies a genoux?
                        -- Les rois du monde