Re: [PATCH] Fix PTA with lots of SFTs

[David Daney <ddaney at avtrex dot com>]

Richard Guenther wrote:
> On Nov 27, 2007 9:38 PM, David Daney <ddaney@avtrex.com> wrote:
> > Richard Guenther wrote:
> > > On Nov 27, 2007 8:03 PM, Daniel Berlin <dberlin@dberlin.org> wrote:
> > > > On 11/27/07, David Daney <ddaney@avtrex.com> wrote:
> > > > > /home/daney/gccsvn/native-clean/./prev-gcc/xgcc
> > > > > -B/home/daney/gccsvn/native-clean/./prev-gcc/
> > > > > -B/usr/local/test/x86_64-unknown-linux-gnu/bin/ -c -g -O2 -DIN_GCC -W
> > > > > -Wall -Wwrite-strings -Wstrict-prototypes -Wmissing-prototypes
> > > > > -Wold-style-definition -Wmissing-format-attribute -pedantic
> > > > > -Wno-long-long -Wno-variadic-macros -Wno-overlength-strings -Werror
> > > > > -DHAVE_CONFIG_H -I. -I. -I../../clean/gcc -I../../clean/gcc/.
> > > > > -I../../clean/gcc/../include -I../../clean/gcc/../libcpp/include
> > > > > -I../../clean/gcc/../libdecnumber -I../../clean/gcc/../libdecnumber/bid
> > > > > -I../libdecnumber ../../clean/gcc/tree-ssa-structalias.c -o
> > > > > tree-ssa-structalias.o
> > > > > cc1: warnings being treated as errors
> > > > > ../../clean/gcc/tree-ssa-structalias.c: In function 'find_what_p_points_to':
> > > > > ../../clean/gcc/tree-ssa-structalias.c:4764: error: 'j' may be used
> > > > > uninitialized in this function
> > > > > ../../clean/gcc/tree-ssa-structalias.c:4764: note: 'j' was declared here
> > > > > make[3]: *** [tree-ssa-structalias.o] Error 1
> > > >
> > > > Hmmm, this is actually theoretically true, since we don't set j when
> > > > low > high in get_first_overlapping_subvar.
> > > But then we ICE due to the assert.  Which is of course disabled with
> > > --disable-checking.  /me thinks that we should just remove the capability
> > > of disabling gcc_assert()s.
> > Or just initialize j in this case.  For x86_64-pc-linux-gnu it appears
> > to be the *only* place where -Werror kills the bootstrap.
>
> But if j is not set by get_first_overlapping_subvar we might silently generate
> wrong code.  So maybe we want a different gcc_do_not_disable_assert ()
> which doesn't guard sanity checks but required otherwise fatal invariants.  But
> practially all gcc_asserts should be of such nature.

Well I don't care one way or the other.  I was just seeing if I could 
reproduce http://gcc.gnu.org/bugzilla/show_bug.cgi?id=34144 . I hit this 
first.

The install documentation mentions that you can use --disable-checking, 
so something should probably change.

David Daney