CLFS and Optimizations and MAKEFLAGS

William Harrington kb0iic at berzerkula.org
Thu Sep 26 08:01:25 PDT 2013


On Sep 26, 2013, at 3:34 AM, Pierre Labastie wrote:

> Will commit your patch (since I have a limited ability to test CLFS,  
> and
> that patch only affects CLFS, I trust you). The question is whether we
> should have a similar mechanism for LFS (possiblity to build chapter 5
> with makeflags only while building chapter 6 with makeflags and
> optimization, which is not possible presently AFAICT).
>
> Another related and more general question is whether "make install"
> shouldn't always be run with -j1.
> Since the bottleneck is mainly I/O during installation, I would expect
> that the performance would not be significantly degraded. OTH, in some
> cases, packages which build  OK with make -jN do not always install  
> (see
> http://lists.cross-lfs.org/pipermail/clfs-dev-cross-lfs.org/2013-September/001720.html) 
> .
> It would be easy to change that in command generation.
>
> What do the others think ?

Hello Pierre,

A slight change to the OPT_3 bool

It should only be availble if BOOK_CLFS.

                 config  OPT_3
                         bool    "Cross tools (MAKEFLAGS only), temp  
tools and final system" if BOOK_CLFS

Notice I added if BOOK_CLFS at the end of the bool line under config  
OPT_3.


For the perl issue, it may be cause of not using the make upstream  
fixes patch in chapter 5 make. I'm going to verify. By adding the  
patch into ch5 and see if the problem persists. Also, MAKEFLAGS is  
exported, and the build system should use it properly when generating  
the Makefiles. I'm not sure why any dev would be using job control  
during install, but they do!  I also do understand that sometimes when  
running make install, that some packages are still compiling code and  
linking libraries!

Frankly, I do not understand why devs wouldn't force job control to 1  
or turn it off altogether while doing make install. Same for some  
testsuites.

Sincerely,

William Harrington



More information about the alfs-discuss mailing list