I notice that direct_machine.cpp has a workaround for llvm.39241:
// In this file, &&foo is used in a static variable initializer, and it's not
// entirely clear whether this should prevent inlining of the function or not.
// In practice, though, clang 7 can end up inlining the function with ThinLTO,
// which breaks at least is_return. https://bugs.llvm.org/show_bug.cgi?id=39241
// So all in all, we need at least the noinline attribute. noclone
// is not supported by clang.
attribute((noinline))
const void * direct_run(const bool get_table_mode,
const instr * program,
const byte * data,
Machine::stack_t * stack,
slotref * & __map,
uint8 _dir,
Machine::status_t & status,
SlotMap * __smap=0)
This compiler bug is already marked as fixed: https://bugs.llvm.org/show_bug.cgi?id=39241
Shall the above workaround be removed?
I notice that direct_machine.cpp has a workaround for llvm.39241:
// In this file, &&foo is used in a static variable initializer, and it's not
// entirely clear whether this should prevent inlining of the function or not.
// In practice, though, clang 7 can end up inlining the function with ThinLTO,
// which breaks at least is_return. https://bugs.llvm.org/show_bug.cgi?id=39241
// So all in all, we need at least the noinline attribute. noclone
// is not supported by clang.
attribute((noinline))
const void * direct_run(const bool get_table_mode,
const instr * program,
const byte * data,
Machine::stack_t * stack,
slotref * & __map,
uint8 _dir,
Machine::status_t & status,
SlotMap * __smap=0)
This compiler bug is already marked as fixed: https://bugs.llvm.org/show_bug.cgi?id=39241
Shall the above workaround be removed?