On Tue, May 29, 2018 at 10:26 AM, Jakub Jelinek <ja...@redhat.com> wrote: > On Tue, May 29, 2018 at 10:16:49AM -0400, Jason Merrill wrote: >> On Tue, May 29, 2018, 4:31 AM Jakub Jelinek <ja...@redhat.com> wrote: >> > Initializing the decomposition temporary from an expression with array type >> > is a special aggregate initialization path in which we wouldn't mark the >> > expression as read for the purposes of -Wunused-but-set*. >> > >> > Fixed thusly, bootstrapped/regtested on x86_64-linux and i686-linux, ok for >> > trunk? >> > >> > 2018-05-29 Jakub Jelinek <ja...@redhat.com> >> > >> > PR c++/85952 >> > * init.c (build_aggr_init): For structured binding initialized from >> > array call mark_rvalue_use on the initializer. >> > >> > * g++.dg/warn/Wunused-var-33.C: New test. >> > >> > --- gcc/cp/init.c.jj 2018-05-25 14:34:41.000000000 +0200 >> > +++ gcc/cp/init.c 2018-05-28 19:04:10.504063972 +0200 >> > @@ -1678,6 +1678,7 @@ build_aggr_init (tree exp, tree init, in >> > if (VAR_P (exp) && DECL_DECOMPOSITION_P (exp)) >> > { >> > from_array = 1; >> > + init = mark_rvalue_use (init); >> >> This should be mark_lvalue_use, since the structured bindings refer to >> the elements of the array rather than copying them. OK with that >> change. > > I think they refer to the elements of the decomposition variable (i.e. exp). > > "If the assignment-expression in the initializer has array type A and no > ref-qualifier is present, e has type cv A and each element is > copy-initialized or direct-initialized from the > corresponding element of the assignment-expression as specified by the form > of the > initializer." > > is what applies in this case, and > > int a[2] = {1, 2}; > int D.2131[2] = a; > int x [value-expr: D.2131[0]]; > int y [value-expr: D.2131[1]]; > > <<cleanup_point int a[2] = {1, 2};>>; > int D.2131[2] = a; > return <retval> = x + y; > > is what original dump shows as implemented, so I don't see a being used here > as an lvalue, we copy the elements into the temporary and that is all where > it is referenced.
Ah, no, you're right for foo, where the structured binding declaration is not a reference. And it looks like we shouldn't hit this path for auto & [x,y] = a; but that should be added to the testcase. Jason