On Fri, Mar 04, 2016 at 11:42:48AM -0500, Fritz Reese wrote: > > Here I propose a patch to gfortran which allows initializing > components of derived type variables with a new compile flag. > > Currently the options -finit-integer=, -finit-real=, -finit-logical=, > and -finit-character= are avaialable to initialize variables of each > respective type; however, there is no way (other than explicitly in > source) to initialize derived type variables. > > In this patch I add the flag -finit-derived. This flag allows > components of derived type variables to be initialized as with local > variables. This flag obeys the values given to the other -finit-*= > flags; with -finit-local-zero derived type variables are initialized > to zero. > > The brunt of the work happens in expr.c (gfc_default_initializer). I > add a boolean parameter to gfc_default_initializer which indicates > whether or not to generate initializers for components that do not > already have them. If true, an expression is generated for each > component which does not already have an explicit initializer. > > This parameter is passed potentially as true in two places in > resolve.c (apply_default_init and resolve_fl_variable_derived). > Generation is guarded by can_create_initializer, which ensures similar > conditions as in resolve.c (build_default_init_expr). > > To do the generation, gfc_default_initializer calls upon a new > function component_init in expr.c. In order to use the -finit-* flags > this calls upon the behavior formerly implemented in resolve.c > (build_default_init_expr); I moved this behavior to the public > function gfc_build_default_init_expr in expr.c; > build_default_init_expr in resolve.c now simply wraps this function to > protect it from being called on invalid symbols. > > I wrestled with whether to change the interface for > gfc_default_initializer or create an entirely new function (like > gfc_generate_derived_initializer). I decided to change the old > function because their behaviors would be almost identical, and there > are only a few calls to the former. > > The patch is based on trunk. It builds and passes all regression tests > on x86-64-gnu-linux.
Fritz, I'm trying to catch up in my inbox. Has anyone reviewed this patch? -- steve
