It seems that some versions of gcc still play bad with APR_RINGs when optimizations are on (O2). At least I can see warning when rings are created on the stack (eg. httpd's mpm_event, htcacheclean), and PR 57800 is possibly related (infinite loop with gcc-4.8 which I think is related to the assemby generated for APR_BRIGADE_PREPEND).
How about adding/forcing -fno-strict-aliasing to APR flags (and include it apr1-config and related so that it is also exported to the user)? And what would be the way to do that if that's acceptable? Thanks, Yann.
