On 13 December 2016 at 16:52, Richard Biener <richard.guent...@gmail.com> wrote: > On Tue, Dec 13, 2016 at 3:45 PM, Ville Voutilainen > <ville.voutilai...@gmail.com> wrote: >> On 13 December 2016 at 16:42, Richard Biener <richard.guent...@gmail.com> >> wrote: >>> Also maybe N should be the number of exception objects rather >>> than bytes? Otherwise a target independent N is hard to specify >>> for say, a distribution that wants either a different default or a >>> static buffer. >> >> >> But how to specify the number of exception objects that may have >> different sizes? > > We have EMERGENCY_OBJ_SIZE for this (yeah, some guesstimate) > which can be easily target configured in libstdc++ itself. We possibly > should split it into a sizeof () of the exception header plus some > payload guesstimate.
Ah, yes, that's what the patch does with the default-size when there is nothing in the environment to customize the value. Makes sense. I don't know whether a separate byte-wise count would be useful, I must admit I have no field experience with such tunings.
