[kaffe] Re: Darwin/x86 134 of 144 tests Failed

[Dalibor Topic]

Michael Franz wrote:
> Kiyo,
Hi Michael, konnichiwa Kiyo,

> On Thu, 9 Sep 2004 01:25:28 +0900 (JST), Kiyo Inaba
> <inaba at src.ricoh.co.jp> wrote:
> 
>>Michael,
>>
>>
>>>Is 300 seconds a resonable value?
>>
>>No.
>>
>>But, I don't agree to set this value shorter. There are several
>>resaons,
>>
>>1) It does not solve your problem at all.
>>2) Poperpc and Pentium are not the only CPU available.
>>etc, etc.
>>
>>Of course ideal solution is to set this value based on the cpu
>>performance, but I have no clear idea how to do that...
>>
> 
> What are your ideas for this?  

I see two options: do some magic calculation in the config, or let the 
user chose.

For the latter, I could make the $TIMEOUT variable more explicit, by 
allowing it to be set by the configure script. For 150 tests, each 
hanging for 5 minutes ... that's a lot of time.

On some platforms, in particular with the interpreter, it can take a 
while for tests to pass (or fail). Kiyo has witnessed regression test 
runs that went for a day, or longer. I think the current m68k-jit 
regression tests take about a day on his box. Riccardo can tell you 
amazing war stories of day-long builds on #kaffe, too ;)

So for the former, we could do a small calculation to estimate the 
maximum time a test should take by cpu and engine type. Interpreter is 
about 10 times slower than jits. Jit and Jit3 are different on different 
cpus. I think Riccardo could provide some estimates based on his 
benchmarks. I've cc:ed him.

A very crude (and factually wrong, anyway, but as we are guessing upper 
bounds here, the advantages of RISC systems do not do harm) estimation 
would be to look at the lowest CPU megahertz counts people are testing 
on: if it takes 5 seconds with jit on a 200 MHz ix86 then it may take 
5*10*10 secs on a 20 MHz m68k with intrp.

This would be of course error prone, and complex, and so on.

I'd perosnally prefer the configure script option, as that would allow 
you, the porters, to set it in a config.frag file, and forget about it ;)

>>And, at the very end, please consider 'Kaffe is not only for
>>your CPU architecture or OS' when you post :-)
> 
> I didn't mean to imply that PowerPC and x86 are the only CPUs, it just
> happens to be the only ones I use and currently the 300 seconds seem
> too long.

Chances are that you're being bitten by a threading issue that manifests 
itself heavily on darwin-x86. ;)

Do you have the same problems when you build kaffe with 
--with-threads=unix-jthreads ?

> What exactly is the timeout of 300 for?  
> 
>>Kiyo
>>P.S. If we have to change this value, I have to propose longer
>>     rather than shorter.
> 
> I guess if I know what it was really for, I might agree,

It's for the porters debugging interpreter failures on older CPUs, 
essentially.

cheers,
dalibor topic