This conception was originally introduced to implement constant network senders (and New - their implementation). Later it was also reused to implement struct selectors (and, respectfully, Field).
The thing is Fieldcan (with some performance overhead) be implemented without config message (see #683) while New is not. New really needs something else because it doesn't have inports.
So, if it's only case for New isn't cfgMsg too big abstraction? Too generic, too universal? How about doing the same as we do for Panic? We could e.g. inject "ConstantMessage" into ctx. On the level of IR we can keep cfgMsg abstraction as is though.
RN only 2 components needs this
FieldNewThis conception was originally introduced to implement constant network senders (and
New- their implementation). Later it was also reused to implement struct selectors (and, respectfully,Field).The thing is
Fieldcan (with some performance overhead) be implemented without config message (see #683) whileNewis not.Newreally needs something else because it doesn't have inports.So, if it's only case for
Newisn'tcfgMsgtoo big abstraction? Too generic, too universal? How about doing the same as we do forPanic? We could e.g. inject "ConstantMessage" intoctx. On the level of IR we can keep cfgMsg abstraction as is though.Related to #683 and #684