Re: how to instruct gmake to read included fragmented makefiles paralally..

[David Boyce]

At 03:44 AM 6/21/2007, Philip Guenther wrote:
> On 6/21/07, Divakar Venkata (divvenka) <address@hidden> wrote:
> > We use non-recursive make system with over 1000+ fragmented
> > makefiles all over source code. We have seen make spending
> > lot of time in reading these files and calculating dependency graph
> > before starting build.

I also agree with Philip, with a couple of additional points:

1. I measure the parsing phase by adding a phony target called 
"nothing". Then, running

        time make nothing

should, in theory, do a good job approximating the cost of the first 
(parse) phase.

2. I have a similar non-recursive, include-based build model. The 
number of our makefile fragments is smaller (~250) but each is large 
(lots of lines) and dense (long lines). Here is the report issued by 
wc (characters, words, bytes) for my makefile set:

        69530  351573 6729699 total

GNU make takes only a few seconds to parse this. My experience argues 
that the parsing time is almost always negligible, because if your 
build is of a scale that parsing time is even noticeable the build 
time will be so large as to dwarf it. Of course there's always the 
case of the "null build" (where make simply reports everything being 
up to date) but really, if you have (say) an 8-hour build and it 
takes (say) 60 seconds to determine that nothing needs to be done, 
you should spend those 60 seconds saying "Thank you Stuart Feldman" 
(original author of make).

David Boyce