Re: What's wrong with alloca() ?

In article 91-12-089
wws@renaissance.cray.com (Walter Spector) writes:
>Note that the routines 'array' and 'another1' are local to routine 'junk'.
>They are allocated upon entry (doesn't matter if it is the stack or heap)
>and deallocated upon exit.


It does matter where they are allocated: setjmp and longjmp are complicated
if they have to maintain a list of objects to be freed. If variable size
arrays are allocated on the heap, there must be a list of such objects where
longjmp can read it (the simplest implementation appears to be to keep the
objects as a linked list, with a word in the stack frame pointing to the head
of the list). longjmp on the IBM RT, which uses the stack for alloca, is 4
instructions long (2 instructions to restore registers, 1 to copy the return
value, and 1 to return).


Another problem with variable size objects on the stack is that out of memory
is typically signaled by a segmentation violation (or similar fatal error)
instead of by a NULL return from malloc. Constant size objects are subject
to the same problem in theory, but in practice very few C programs that do
not call alloca() reach the stack memory limit. (This turned out to be the
cause of a mysterious X server crash: X uses alloca for temporary space, and
it is possible to send a single request to the X server that makes it try to
allocate more space than the default stack limit on our BSD workstations).




--
        John Carr (jfc@athena.mit.edu)
--

Return to the comp.compilers page.
Search the comp.compilers archives again.