gcc3 audit

Matthew Dillon dillon at apollo.backplane.com
Fri Aug 27 10:57:25 PDT 2004

:we could need a list of problems creeping up with gcc3 so that we can 
:track the bugs down and slowly move to gcc3.
:The list could consist of
:a) possible sources of errors
:b) definite errors
:c) proposals how these could be fixed
:starting with point b: loader has problems with gcc3. joerg once 
:tracked this to the forth interpreter. i also had problems with btx in 
:virtual pc. so if somebody feels bored she could compare assembler 
:output for ficl :) or come up with a way to debug this (bochs?)
:   simon

    I thought I had fixed the issues with gcc-3 compiled bootstrap and
    loader code.  Try compiling it up under gcc3 (e.g. /usr/src/sys/boot;
    setenv CCVER gcc34; make clean; make; make install) and report on any
    issues.  If you break your bootstrap you will have to boot from CD
    (or another slice) and then restore working bootblocks with disklabel -B
    and possibly also copying /boot/loader from the CD to the HD.  I find
    it easier to have a second DragonFly slice that I can boot from so I
    can fix the bootsrap code in the first slice when I break it.

    Generally speaking we should do a pass on the FreeBSD-5 cvs tree
    looking for gcc3isms they have fixed, but it isn't going to be easy.
    The quality of the commit message descriptions in the FreeBSD-5 tree
    has gone down considerably.


More information about the Kernel mailing list