platforms problems

[James E Keenan]

On 8/1/12 8:19 PM, Andrew Whitworth wrote:

>
> I've also tried looking for an old PPC mac to purchase for myself, but
> I've been unable to find one in workable condition for a price I can
> pay.
>

Well, perhaps when I buy a new Mac, I can give you mine :-)

> [snip]

>
> We can declare Darwin/PPC to be a platform we no longer support. That
> is my preferred solution for that platform at this point, considering
> it's age and relatively small (and shrinking) user base.

The significance of Darwin/PPC is not that it is a platform we need to 
support.  Its significance is that it is a platform on which Parrot 
master has built and tested successfully at least since 2006 and 
probably since 2001.  Today, more than 11 years since we first started 
building and testing on it, it still has a better test record than, say, 
Windows.

It continues to build on Parrot master and, as of July 24, the 
whiteknight/io_cleanup1 branch was PASSing 'make test' here as well. (I 
am re-running that branch now to double check.)

So the locus of the problem is the threads branch, where the fact that 
tests *hang* -- which is a different problem from tests *failing* -- has 
been confirmed by dukeleto.  And Darwin/PPC is not the only platform on 
which the threads branch is FAILing 'make test.  That suggests to me 
that the problem is in the branch, not in the platform.

This further suggests to me that we, as a group, do not understand what 
is happening in the threads branch or, more fundamentally, what the 
purpose of the threads branch is and how merging it into master would 
improve Parrot.  Only two contributors have committed to the branch in 
the past month, and only four contributors in the past two months.  (And 
since I was one of those four when I made changes to pmc2c, that's 
really only three. ;-) )  Quick!  If anyone other than whiteknight can 
explain the purpose of the threads branch without pausing for more than 
thirty seconds, raise your hand now.  Ah, I thought so.

I don't claim to understand threads in any computer technology.  All I 
know is that (a) whenever the topic of threads comes up, everyone's 
first reaction is: "Threads are hard"; (b) even in Perl 5, perl with 
threads is a very different beast from perl without threads.  In twelve 
years of doing Perl 5, I've never had a reason to compile or use a perl 
with threads.  I don't think I'm alone in that.  Hence, I suspect that 
only a literal handful of people in the Parrot world have a good 
understanding of threads, know what's in the threads branch and can 
explain why we need that functionality in Parrot.

So the real question is not whether Parrot "supports" Darwin/PPC.  The 
real question is:  Do the benefits from merging the threads branch into 
master outweigh the risks we incur when we don't know why tests are 
failing on platforms where they have long passed?

Thank you very much.
Jim Keenan