Thanks Eric for your fast answer! Unfortunately, this doesn't solve my problem. To make a very very long discussion long, here a small example showing the problem from another side: I have an expression wrapper called mp_terminal and I define terminals, but not in the documented way (which I suspect is the problem) but instead deriving from them (for convenience purposes), for example: struct True : mp_terminal::type> {...} These terminals are used with a simple grammar: struct BuildGuards : proto::or_< proto::when< proto::logical_or, GuardOR() >, proto::when< proto::logical_and, GuardAND() >, proto::when< proto::logical_not<BuildGuards >, GuardNOT() >, proto::when < proto::terminal, proto::_()

{}; Then, expressions like: True()|| False() give the expected result (struct GuardOR). Even True()&& (Dummy1() || !(False() || True())) give the expected result: struct GuardAND > > > But a "!" alone (meaning followed by just a terminal, not the other parts of the BuildGuards grammar) breaks all (as do all unary operators I tried), like !False() or True()|| !False(). !False() => struct GuardNOT > > Notice that !(False() || True()) works! Now, if I do not derive from mp_terminal but instead proceed as documented, all works as expected: proto::terminal<True>::type True_; ... !False_ => correct result. Sadly, this documented usage is unpractical in my use case. So, supposing that I managed this easy grammar, I imagine that the usage I do of terminals (deriving from them) is not allowed. Then, why do most use cases work like a charm? Only unary operators seem to fail. Attached are 2 test files. V1 doesn't work in all cases while V2 does. I use 1.38. Thanks a lot! Christophe _________________________________________________________________ Windows Live™: Keep your life in sync. Check it out! http://windowslive.com/explore?ocid=TXT_TAGLM_WL_t1_allup_explore_012009