Re: Improvement on parallel make

[Christophe LYON]

Alexey,


> 1. This patch is subject to the same deficiencies as I mentioned at 
> http://lists.gnu.org/archive/html/help-make/2006-04/msg00152.html, 
> but in other way: in that example, when one runs "make -j b a", the 
> jobs updating "x1" and "x2" are started in parallel. The rule in 
> makefile, however, requests serialization of "x1" and "x2".

I don't think so:
x2 will wait for x1 to complete, and x3 will wait for x2.

You may have noticed that my patch differs from yours in that the .wait 
property, although first tracked on the deps list, is later propagated 
at the file level.

> 2. The .WAIT semantics you implemented differs from its origins in BSD 
> make: it waits for *all* dependencies preceding .WAIT, not for the 
> last one. Note that it is not equivalent to placing .WAIT between 
> each and every dependency.

Yes, I realized that. However, I think that what I implemented is much 
simpler and gives already an interesting feature.

If one wants to have:
foo: a1 a2 .WAIT a3 handled in the BSD way (that is, a3 waits for both 
a1 and a2), one can rewrite the rule as:

a-parallel: a1 a2
foo: a-parallel .WAIT a3

> 3. BSD make associates the .WAIT semantics with a target, not with a 
> dependency. While being a slight difference, it could be a nuisance 
> to those with BSD make "background".
>
> In the message mentioned above I described how .WAIT is implemented in 
> BSD make; that seems to be the "right" approach. However, it is a 
> large change - that I haven't had the time to implement, 
> unfortunately.
As I have no such BSD background, this is not a problem to me ;-)

Maybe we could accept a different behaviour that BSD's. (Well, I hope so ;-)

Christophe.