Re: which build is the fastest?

[Bruno H. <br...@cl...>]

Hi Sam,

> > If you give a man a toy, he will play with it.
> > Now that we have all those multibuild targets in Makefile.devel, we can
> > find out which build is the fastest (e.g., if I make small
> > architecture-agnostic change, which build do I test it in?)

This is something you can easily find out, by putting a 'time' command
in Makefile.devel and then running a 'multibuild-<platform>' target with '-k'.

In my experience, the following factors are important (in decreasing order
of importance):
  1) Put the GCC binaries on a local disk, preferably on a SSD.
     Do not put them on a CIFS mount - this gives terrible compilation times.
  2) Use an older GCC version. GCC 3.x is way faster than 7.x.
  3) Revisit the optimization options: -O is a good compromise, -O2 increases
     the compilation times significantly, and -O2 -fexpensive-optimizations is
     just excessive.

> > Bruno, do you mind if I add "time" to every multibuild target?
> 
> Also, does the configure script have to support platforms without
> /usr/bin/time?

I do mind, yes. And you guessed the reason: We haven't tested portability.
The POSIX 'time' command produced ugly output. 'time' as a bash built-in is OK.

> I noticed that you use sed instead of tail

Yes: Solaris 10 'tail' does not understand the '-n' option.
See also the chapter "Portable Shell Programming" [1].

> so I wonder what the assumptions are.

The assumptions are those listed by the GNU coding standards [2].

Bruno

[1] https://www.gnu.org/savannah-checkouts/gnu/autoconf/manual/autoconf-2.69/html_node/Portable-Shell.html
[2] https://www.gnu.org/prep/standards/html_node/Utilities-in-Makefiles.html