Vincent Trouilliez <address@hidden> wrote:
So no need to regresss feature wise by removing this macro, all the
user need is a concise but clear information, at the right place in
the document.
Don't worry. My personal guess is there are probably a dozen or so
legitimate users of INTERRUPT(), yet these people can reasonably be
expected to understand what they are doing, so they can live with
explicitly declaring their handler as __attribute__((interrupt)) as
well. So eventually removing INTERRUPT() is not really a regression.
I just deprecated it for the next release of avr-libc, and removed it
from all the introductionary documentation. (The macro itself is
still documented though.)
Historical note: all that introduction has once been taken from Rich
Neswold's independent documentation of avr-libc which he wrote by a
time when there had been no official documentation for it. We simply
converted it into doxygen-style, without really looking at its
contents. That's why INTERRUPT and SIGNAL appeared with too little
distinction there. Over time, it turned out users ran into a lot of
confusion about it, and I think there's general agreement among
everybody now that the original naming was rather confusing. What is
now named SIGNAL() should have been named INTERRUPT() right from the
beginning as that's what the users would expect. But changing this is
too late now.