RE: honoring gcc test stack size

[Dave Korn]

On 13 February 2008 16:36, Hans-Peter Nilsson wrote:

>> Date: Wed, 13 Feb 2008 10:12:45 -0600
>> From: Joel Sherrill <address@hidden>
> 
>> Would it make sense in the future as a (beginner?)
>> GCC project to let me generate a set of GCC testresults
>> and let someone see which tests have issues honoring
>> the stack size.
>> 
>> (Run 1) Run with a small 2-8K stack.
>> (Run 2) Run with 256K or 512K stack and print stack
>>   usage via an atexit() function.
> 
> I'd suggest something like 64M (large enough to PASS most hungry
> tests, but doesn't oom your machine), 512K is still a bit small.

  As a datapoint, the standard per-thread stack allocation on win32 platforms
is 2MB.  That might make a more reasonable cutoff point, because in what's
probably a bimodal distribution, it's the one and only datum that's stuck in
the middle of the range, and we'd want it to fall in the big-stack side of any
divide rather than the tiny-embedded-machine-stack side.


>> Would that be useful?
> 
> Yes, I think so.

  :)  Concur.  AFAIR it's an entirely ad-hoc arrangement as to whether
testcases regard or disregard the #define and gets overlooked sometimes.

    cheers,
      DaveK
-- 
Can't think of a witty .sigline today....