Subject: Testing the new toolchain
To: None <current-users@netbsd.org>
From: Geoff Wing <mason@primenet.com.au>
List: current-users
Date: 08/22/2001 06:37:49
Todd Vierling <tv@wasabisystems.com> typed:
: (Note that the new build structure doesn't work as committed just yet;
: there's a pending issue with inheriting variables in make's environment that
: I have to tackle after I get some sleep.  I'll post again when this is
: fixed.)

Just noting my experiences at a manual bootstrap here (on i386).
 
: You may build the tools yourself, manually, at any time (good for diddling
: inside the tree) by going to src/tools and running "make build". 

Except due to the way ``make'' works with the dependencies used, if you
have old binaries of those tools lying around in the source locations,
i.e. from a previous build, then it won't build it properly and will die
on the tool install part.  May not have applied to all tools - but I can't
remember which tools it did apply to.  Of course, I ended up not using them
(since they're not ready yet)

I manually rebuilt (with USE_NEW_TOOLCHAIN set):
gnu/lib - dies in libgcc and libstc++ because I hadn't built the new gcc yet
        - only really needed libbfd and libiberty, I think, at this stage
gnu/usr.bin

Then install gnu/lib/libbfd and gnu/lib/libiberty and then gnu/usr.bin
Then I could rebuild everything.

gcc-2.95.3 (with -O2 and -Os flags) keeps producing larger and seemingly
less optimised binaries as compared with egcs-1.1.2  (from inspection of
size and via objdump -S)

sys/arch/i386/stand/biosboot (i.e. the resultant /usr/mdec/biosboot.sym)
didn't work properly for me (on P-III) with  -march=i386 -mcpu=pentiumpro
but did work without the pentiumpro scheduling.  Maybe that's already
known?

Other than that, everything's going fine.  Good stuff.

Regards,
-- 
Geoff Wing : <gcw@pobox.com>
Rxvt Stuff : <gcw@rxvt.org>
Zsh Stuff  : <gcw@zsh.org>