Re: [Boost-users] Coroutine jump_fcontext crash (Boost 1.61)

Subject: Re: [Boost-users] Coroutine jump_fcontext crash (Boost 1.61)
From: Jason Mancini (jayrusman_at_[hidden])
Date: 2016-06-21 21:16:00

> You should probably have an ABI compatibility namespace if defining this
> differently on each end causes ABI breakage.

For instance, we have 45 boost installations (boost version * gcc versions * 32/64 bit * options), and 30 valgrind installations.  That would require 1350 boost+valgrind installations to generate all possible debug cases, hahaha.

One ABI solution would be to leave the valgrind_stack_id declared in the struct at all times (or a field of equivalent size).  At least the ABI wouldn't change.

#if defined(BOOST_USE_VALGRIND)

  unsigned valgrind_stack_id;

#else

  unsigned unused_valgrind_stack_id;

#endif

My solution is to force the define in the 7 header files that use it, such that it is always enabled, for the library/installation build, and all end-user code.  Because ABI class layout mismatches are horrible to debug, we've been through that before.

It would be nice if the installed *.so was not dependent on valgrind.h when BOOST_USE_VALGRIND is in play.

Jason