------- Comment #3 from jv244 at cam dot ac dot uk 2009-09-19 11:37 ------- I've also run into this issue, it would be great if there would be a flag like -finit-undefined-intentout (?), so that intent(out) variable would first be undefined by the compiler (in a way that valgrind notices this with memcheck). This is easy to do on the source level (see below), so I would hope this is possible as well in the FE. The example would be:
MODULE test CONTAINS SUBROUTINE S1(a) INTEGER, DIMENSION(:,:), INTENT(OUT) :: a ! FE generated code INTEGER :: uninitialized_a_type_dummy a=RESHAPE((/(TRANSFER(uninitialized_a_type_dummy,uninitialized_a_type_dummy),i=1,SIZE(a))/),SHAPE(a)) ! rest of the procedure END SUBROUTINE END MODULE test USE test INTEGER :: a(3,3) a=0 CALL S1(a) write(6,*) a IF(ANY(a.NE.0)) WRITE(6,*) "No problem" END In this way, valgrind does produce a warning when a is being written after the call to S1. The above example works also for derived types with default initializers and similar. -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=31447
