Re: egcs build report

[Mark Mitchell <mark at markmitchell dot com>]

>>>>> "Joe" == Joe Buck <jbuck@synopsys.com> writes:

    >> > Yes, we want -g included in the flags by default.
    >> 
    >> Why?
    >> 
    >> IMHO, this makes egcs more ``expensive'' in terms of disk space
    >> and build time, and I'd say that 99% of all _users_ will never
    >> use it.

    Joe> But those users will install binary distributions, which
    Joe> generally will be built without -g.

This is clearly a debate with many reasonable points of view.  I'll
just offer mine, without intending criticism of anyone else's.

In my experience, working for a commercial entity that used gcc, but
not on Linux, few binary distributions were available.  None were
available for many of the platforms we needed, like NCR MP-RAS, and
those available even for mainstream Solaris were not of recent
vintage.

I think many commercial UNIX users install and build GCC themselves.
Because of fixincludes, one must in fact do this for minor revisions
of the OS.  We used to have separate Solaris installations for 2.4,
2.5, and 2.5.1 since we built on each of these different machines.

Thus, the extra disk space *was* noticeable.  We stripped binaries
before installing them.

    >> Could it make sense not to include -g for releases?

    Joe> This has not traditionally done to help convey the sense that
    Joe> free software/open source is not simply a freebie, it is an
    Joe> effort that people can contribute to.

That's true.  On the other hand, we have had such success as GNU
developers (and we are to be congratulated for it!) that people who
are just users, and not hackers, are using our products.  We should be
proud of this; it indicates that our products are useful and usable.
However, there are now many people dowloading, building, and
installing software who have no desire to debug it.

Users should not have to understand anything about compilers in order
to download and build software; I'd rather they didn't even have to
understand the notion of `debugging information'.

    Joe> GNUware has always been done this way.  Considering that disk
    Joe> prices are now about 1/100th of what they were when this
    Joe> policy began, I don't think it's a problem.

Our NFS servers got full, so I do think it's a problem.

    Joe> A compromise might be to include in the installation
    Joe> instructions "If you're really short on disk space and not
    Joe> competent to help us debug, do xyz".  Proper instruction
    Joe> should be given!  As g++ FAQ maintainer, I don't want to

At a minimum we should do this.

I suggest that the autoconf defaults should be changed from `-g -O2'
to `-O2'.  Then, there should be an --enable-debugging (or similar)
flag to turn on -g, and, for ease of use by all of us full-time
hackers, an environment variable we could set to tell configure we
want the old behavior.  I'm willing to implement the autoconf hacks,
if we agree this is thing to do.

Note that I'm suggesting we do this for *all* packages, not just gcc.

-- 
Mark Mitchell 			mark@markmitchell.com
Mark Mitchell Consulting	http://www.markmitchell.com