Re: Bootstraping compilers ?

I had the pleasure of working with some ex-SGI compiler people at
PathScale, and they had 2 major reasons for bootstrapping:

1) You don't want to hunt bugs in the other compiler that make it
compile your compiler wrong. We had a slowly increasing list of files
in our compiler that needed -O0 with gcc to compile correctly, or
obscure bugs would surface. Now maybe our compiler code wasn't
standard conforming, or maybe gcc had a bug. Either way, we don't want
to be in the business of fixing gcc.

2) You (should) have a great QA testsuite for your compiler, better
than the average customer has for their own programs. You can find and
fix a lot of obscure bugs by bootstrapping with lots of flag
combinations and looking for changes in the QA suite results.

On the general topic of QA, a good lesson we learned from gcc was that
building an entire Linux distro is a pretty useful exercise. You have
to be pretty gcc-compatible to do it, and sometimes the bugs are hard
to localize (binary search through huge lists of rpms..., and it'd
better be easily repeatable).

-- greg