Richard, I am looking at PR 28138, and I think the problem is due to
this patch. I verified that the problem started with this checkin and
then I looked at the prints from ggc_alloc_stat in ggc-page.c. Among
the output I get from the print at the end of ggc_alloc_stat is:
Allocating object, requested size=511, actual=1 at 0x2000000005450001 on 0x6000000000348f40
Obviously, requesting 511 bytes and getting 1 is a bad thing. I haven't
figured out where the code is going wrong yet but I thought I would see if
you had any ideas while I continued to look.