Re: stack overflow at compile time

glen herrmannsfeldt <gah@ugcs.caltech.edu> wrote:
>> The maximum may occur on a path that's actually impossible, but at least
>> you do get a bound.
>
>But it is difficult to tell at compile time which routines are actually
>called, even though you may know there is no recursion and only fixed
>sized allocation. You could, then, way overestimate the stack usage.


As I said, the "worst-case" path may in fact be an impossible one. More
sophisticated static analysis may be able to prune some of the impossible
paths, but it's much more expensive, and the improvement may be limited.


Yes, it's possible for the simple analysis to badly overestimate usage.
How serious that is, depends on how frequently it happens and how much an
overestimate hurts.


>With recursion you could easily underestimate it, not knowing the
>maximum recursion level.


If you want trustworthy results, the analysis has to assume that recursion
depth is potentially unlimited (consider Ackermann's Function) unless it
is bounded by user annotations or more sophisticated static analysis.
--
spsystems.net is temporarily off the air; | Henry Spencer
mail to henry at zoo.utoronto.ca instead. | henry@spsystems.net

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