This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: Solaris bootstrap failure
- To: mark at codesourcery dot com, mrs at windriver dot com
- Subject: Re: Solaris bootstrap failure
- From: "Kaveh R. Ghazi" <ghazi at caip dot rutgers dot edu>
- Date: Thu, 8 Mar 2001 11:59:39 -0500 (EST)
- Cc: gcc at gcc dot gnu dot org
> From: Mike Stump <mrs@windriver.com>
>
> > Date: Tue, 6 Mar 2001 23:27:23 -0500 (EST)
> > From: "Kaveh R. Ghazi" <ghazi@caip.rutgers.edu>
>
> > Any other ideas?
>
> The standard technique of binary searching with cvs and full
> bootstraps will tell you exactly what triggered it. Though, as we
> might guess now, that probably isn't the real bug.
>
> If you modify the compiler used to be one that squirrels away the
> complete environment, and the entire command line, and does -E on the
> input file, and squirrels that away too, and a staging process that
> stages those intermediate files as well, you might just be able to
> observe it in action.
I tried something similar via adding --save-temps to BOOT_CFLAGS, but
then the failure went away. I think using the integrated preprocessor
(which --save-temps turns off?) might be necessary.
> Last time I saw one of these, someone had introduced a bit of non
> initialized stack. :-(
In that case, would testing with BPs or some kind of memory/malloc
checker help? If someone can help try those, that would be great.
> Another way that might allow one to observe this type of problem is to
> build on tons of different hosts, and compare the output between the
> hosts for the same target, on the same code. Doing this for a full
> range of inputs is best, as often, it will only be a smaller
> percentage of files that trip it up.
Well we know cp/tree.o triggers it. Perhaps if someone builds a
Bounded Pointer x86 cross to solaris2.7 I can supply the .i file from
a snapshot known to fail and we'll see if that helps trigger
something.
> Bear in mind, binutils can also cause these types of problems, even
> with identical .s files. I've seen that too. To narrow is done, one
> can try a different binutils with a known to fail botstrap.
I'm using native tools. They could contains bugs, I don't know.
as: WorkShop Compilers 5.0 Alpha 03/27/98 Build
ld: Software Generation Utilities - Solaris/ELF (3.0)
and for stage1:
cc: WorkShop Compilers 5.0 98/12/15 C 5.0
> And my last thought, with some luck, the binary searching for the
> first instance of bootstrap failure, might lead to the actual code
> that is broken.
The problem is, we can't *binary* search, cause the problem seems to
come and go on a day-by-day basis. And since it only appeared when we
turned off --enable-checking on the branch we don't know how long its
been lurking in the sources hidden by the checking switch.
--Kaveh
--
Kaveh R. Ghazi Engagement Manager / Project Services
ghazi@caip.rutgers.edu Qwest Internet Solutions