[Bug c/66447] New: -Werror=maybe-uninitialized indicates trashing of variable across longjmp

["jbowler at acm dot org" <gcc-bugzilla at gcc dot gnu dot org>]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66447

            Bug ID: 66447
           Summary: -Werror=maybe-uninitialized indicates trashing of
                    variable across longjmp
           Product: gcc
           Version: 5.1.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c
          Assignee: unassigned at gcc dot gnu.org
          Reporter: jbowler at acm dot org
  Target Milestone: ---

Download libpng-1.6.17.tar (this version, not the next one which will have
work-round, you may have to look in the archives for this version), e.g.:

ftp ftp.simplesystems.org
<log in as anonymous>
cd pub/png/src/libpng16
get libpng-1.6.17.tar.xz
quit

Extract, e.g:

tar xf libpng-1.6.17.tar.xz
cd libpng-1.6.17

Then:

mkdir test
cd test
CFLAGS="-pedantic -ansi -O2 -Wall" cC=gcc-5.1.0 ../configure --enable-werror
make pngvalid

Expect: successful compilation of pngvalid.c
Observe:
../contrib/libtests/pngvalid.c:4038:30: error: 'exception_prev_1' may be used
uninitialized in this function

The second error will be fixed in later versions.  This one seems to be bogus;
the variable (actually exception__prev, #defined to eliminate warnings about
shadows) is not modified between the setjmp and the longjmp, therefore this is
valid, totally conformant ANSI-C90.  The "-pendantic -ansi" is unnecessary to
show the problem, I just put it in because:

a) That is what libpng is.
b) ANSI X3.159-1989 4.6.2.1 lines 11-14; I believe the variable must be
preserved across setjmp/longjmp.

John Bowler