On 7/31/2019 12:59 PM, Corinna Vinschen wrote: > On Jul 31 12:35, Michael Haubenwallner wrote: >> Hi Corinna, >> >> On 7/30/19 6:07 PM, Corinna Vinschen wrote: >>> Hi Michael, >>> >>> On Jul 30 17:22, Michael Haubenwallner wrote: >>>> Hi, >>>> >>>> following up >>>> https://cygwin.com/ml/cygwin-patches/2019-q2/msg00155.html >>>> >>>> It turns out that fixup_shms_after_fork does require the child pinfo to >>>> be "remember"ed, while the fork retry to be silent on failure requires >>>> the child to not be "attach"ed yet. >>>> >>>> As current pinfo.remember performs both "remember" and "attach" at once, >>>> the first patch does introduce pinfo.remember_without_attach, to not >>>> change current behaviour of pinfo.remember and keep patches small. >>>> >>>> However, my first thought was to clean up pinfo API a little and have >>>> remember not do both "remember+attach" at once, but introduce some new >>>> remember_and_attach method instead. But then, when 'bool detach' is >>>> true, the "_and_attach" does feel wrong. >>> >>> I'd prefer to drop the reattach call from remember, calling both of them >>> where appropriate. >>> >> >> Fine with me, even if that looks a little more complicated for spawn. > > Pushed, with just a small formatting tweak.
I can confirm that this fixes the problem I reported in https://cygwin.com/ml/cygwin-patches/2019-q2/msg00155.html. Ken