Re: perl ithreads support: why hardcode at configure time ?

[Stefano Lattarini]

On 01/11/2013 06:11 PM, Mike Frysinger wrote:
> On Friday 11 January 2013 04:08:26 Stefano Lattarini wrote:
>> On 01/11/2013 05:07 AM, Mike Frysinger wrote:
>>> i can't imagine this is a big runtime penalty, so why does configure
>>> check for the perl's thread settings and then hardcode that in the
>>> generated automake ?
>>
>> I don't know, I wasn't around when the change was introduced.  You'll have
>> to dig out the archives; if we don't find any compelling reason for having
>> the check at configure time rather than runtime, I agree your change is a
>> nice simplification ...
> 
> i couldn't find references to ithreads in the automake or automake-patches 
> archives
>
I only found an explanation of why "require ...; import ..." was used instead
of a simpler "use ...", but as for you, I was unable to find out a rationale
for having threading support checked at configure time.  So I say we apply
your proposed change, and reconsider it only if something breaks or someone
complains.

>>> --- a/configure.ac
>>> +++ b/configure.ac
>>> @@ -77,32 +77,6 @@ installed, select the one Automake should use using
>>>    ./configure PERL=/path/to/perl])
>>>  }
>>> -# We require ithreads support, and version 5.7.2 for CLONE.
>>
>> Here, the comments say (and the code agree) that we should support
>> ithreads only from perl 5.7.2 onwards ...
> 
> i thought that meant something else.  oh well.
> 
>>> +use Config;
>>> +our $perl_threads = $Config{useithreads};
>>
>> ... but there is no such check here.   And yes, so far we still support
>> perl 5.6 so far (albeit we might want to start requiring 5.8 in Automake
>> 1.14, but that's for another thread).
> 
> i'm not a perl hacker, so i don't know how to do that.  but a grep shows that 
> automake is already doing that sort of thing in the tests,
>
Oh, good point, I missed that.

> so i just copied that:
>
> +our $perl_threads = 0;
> +if (eval { require 5.007_002; }) # for CLONE support
> +  {
> +    use Config;
> +    $perl_threads = $Config{useithreads};
> +  }
>
Seems the sane thing to do, yes.

>> Also, might I ask you to format your patches with "git am"?  That will make
>> it super-easy for me to apply them.
> 
> i'm familiar with git.  i expected more push back on this, so i didn't write 
> a 
> proper commit as this was a RFC with a PoC patch inline.  no point in wasting 
> time doing a write up (or the god awful time waste known as the GNU ChangeLog)
> if the patch is going to get rejected.
>
Sensible approach.  But I'm not going to reject the patch ;-)  Unless
it causes some failure of course, but it appears by now that that will
be unlikely.

Thanks,
  Stefano