Subj : Re: deadlock avoidance
To   : comp.programming.threads
From : robertwessel2
Date : Thu Jul 07 2005 01:45 am

Satendra wrote:
> Hi Jose,
> I guess my statement was misinterpreted.  What i meant by atomic here
> is that if we have lets say an integer i with initial value 0 .Thread
> T1 unconditionally modified to 2 (i = 2)  then T2 at any time will read
> it as either 0 or 2 but nothing in between, I agree that with c++
> objects this might not be true. I still don't understand why a function
> who is just checking/reading the value of a global variable can't be
> reentrant ?


Consider an implementation of C on an eight-bit microprocessor where
ints are 16 bits.  And the assignment is x=513 (with an original value
of zero).  The generated code will clearly have to temporarily leave x
in a state where if it were read, would be 512 or 1.  And on many
architectures, non-aligned accesses have little, if any, guarantee as
to atomicity, particularly in the presence of multiple processors.

If you use sig_atomic_t, that will generally* be safe to access as a
unit.  On an eight-bit processor, sig_atomic_t will likely be a char.


*The C standard doesn't actually define sig_atomic_t to have any
particular performance with respect to threads, but most thread-aware C
implementations do what you'd expect.

.