Re: honoring gcc test stack size

[Joel Sherrill]

Hans-Peter Nilsson wrote:
> > Date: Tue, 12 Feb 2008 15:47:14 -0600
> > From: Joel Sherrill <address@hidden>
> >     
>
>   
> > I investigated one failure: pr20621-1.x3
> >     
>
> You mean gcc/testsuite/gcc.c-torture/execute/pr20621-1.c at -O3?
> From the looks of it, it would take about 2*0x4000*4 bytes stack.
>
>   
grrr... that means it needs even more than 128K. 
> > and discovered that it is using a LOT
> >     
>
> #define LOT ?
>   
Well... it is a relative term but when it comes to task stacks
in RTEMS, I tend to think of 2-8K.  A full executable with
termios console can be <32K code statically linked with
all drivers

Code size for a full application with web server, telnetd,
and ftpd with our ported BSD stack is usually 250-400K
code space.
>   
> > of stack.  I don't know what the stack
> > size limit parameter you gave me is supposed
> > to do.
> >     
>
> Have a look at the source and the test-cases. ;)
> Hint: gcc/testsuite/lib/gcc.exp:gcc_target_compile.
> Apparently gcc/testsuite/gcc.c-torture/execute/pr20621-1.c
> doesn't use it.  I guess it should.
>
>   
This is beyond my comfort zone right now.  I am trying
to digest what we have done so far so I can move on to the
other targets. :)
> > I increased the init task's stack to 128K
> > and that one passed.
> >     
>
> Aha, so a LOT < 128K! :)
>
>   
:)  Well it is a configurable number so if making it bigger runs
more tests, then so be it.

I remember griping about glibc's printf using 64K stack over
10 years ago. 

I'm still counting bytes and cycles unfortunately.  It keeps me
honest.
> (I know, I've been there too, some time ago.
> Though these days a MB is "meh".)
>
>   
ahhh MB...
> brgds, H-P
--joel