On 06/22/2017 08:20 PM, David Malcolm wrote:
PR c++/81167 reports a case where a NULL is passed to one of these %qH, and it turns out that we now ICE for this case (with a gcc_assert) whereas previously we printed a '' for the type > This patch slightly reworks the %H and %I-handling code so that it gracefully handles NULL, fixing the ICE in that PR. Whilst I was at it, I also fixed things so that if only one of the %H/%I codes is present, we do the right thing (i.e. fall back to the %T behavior).
I dislike this kind of silently accepting bogus use of an interface. I much prefer the compiler to explode.
At least make it barf with a checking build, even if a production build permits the abuse.
nathan -- Nathan Sidwell