fostel (04/15/83)
The inclusion of printf into the system might be quite a good idea.
Shared memory is more complex then seems appropriate for a simple problem.
I am a bit puzzled by comments that this sort of inclusion would take
us towards the ungainly monstrosities of the past, e.g. OS/360.
From a users point of view, I do not see how the complexity is ANY
different if printf is in the system or in the libraries. The difference
in complexity is apparent to those who might like to change printf;
they deserve much worse than a bit of logistic confusion.
Such an inclusion is emminantly easy to acheive, causes no real confusion,
and would likely save a bit in memory space and page faults. SO WHY NOT?
The Pipe system call is a good example. It is in the system largely so the
transfers can be done in memory, rather than via a shared file. That is an
inclusion in the Kernal which is NOT needed, except for performance gain.
That seems just the situation with printf. The interesting question is
whether the amount of performance gained is worth the trouble. Or whether
there are other library entries that should be included in the same catagory (a 1% gain on each of 10 different library routines ...). Or whether the
extra proc call overhead to the system wipes out the gains. Or whether the
librs would finally be recognized as integral to the notion of "UNIX".
It might also disrupt the "ethnic purity" of the system call set. Ethnic
purity and 50 cents will get you a cup of coffee in my establishment.
----GaryFostel----