Re: Solaris bootstrap failure

["Kaveh R. Ghazi" <ghazi at caip dot rutgers dot edu>]

 > From: Mark Mitchell <mark@codesourcery.com> 
 > 
 > I am having a devil of a time trying to debug a bootstrap failure on
 > the 3.0 branch.  On sparc-sun-solaris2.8, I'm getting a comparison
 > failure for cp/tree.o.  However, rerunning by hand the commands to
 > build cp/tree.o generates the same file in both cases!
 > 
 > Building only 1-stage and running `make check-gcc' gives no additional
 > test failures, relative to historical norms.
 > 
 > I'm going to try the binary search thing, but it's going to take a
 > while given the speed of the machine I have access to.  Does anyone
 > have any ideas here?  (I vaguely remember somebody mentioning that
 > a --disable-checking bootstrap failed on Solaris recently, but I can't
 > find the message.)

I can confirm I see the same thing on solaris2.7:

 > Bootstrap comparison failure!
 > cp/tree.o differs

When I try rerunning stage 1 & 2 by hand, the files are indeed the
same.  However they are both different from what appeared in the
bootstrap last night!  In fact, the file sizes are different:

 136 -rw-r--r--   1 ghazi      129388 Mar  3 01:08 stage2/cp/tree.o
 136 -rw-r--r--   1 ghazi      129396 Mar  3 01:27 cp/tree.o.orig
 136 -rw-r--r--   1 ghazi      129376 Mar  3 08:43 cp/tree.o.s2
 136 -rw-r--r--   1 ghazi      129376 Mar  3 08:49 cp/tree.o

This suggests to me that its not a stageN miscompilation.  Perhaps its
an uninitialized variable that triggered differing behaviors during
the bootstrap, but not now when run close together.

Another oddity, turning on --enable-checking=misc,tree,rtl,gc makes it
go away.

		--Kaveh
--
Kaveh R. Ghazi			Engagement Manager / Project Services
ghazi@caip.rutgers.edu		Qwest Internet Solutions