On Oct 6, 2012, at 21:56, Ryan Schmidt wrote:
> On Oct 6, 2012, at 15:42, Joshua Root wrote:
>> On 2012-10-7 07:35 , Joshua Root wrote:
>>> On 2012-10-7 06:58 , Blair Zajac wrote:
>>>> Looks like there's a problem with the build bots:
>>>> 
>>>> http://build.macports.org/builders/buildbase-snowleopard-x86_64/builds/544/steps/configure/logs/stdio
>>> 
>>> Apparently only the Snow Leopard one, actually.
>> 
>> It seems to have broken somewhere in or just before ports build 11369.
> 
> We've had the problem before of ports deleting /dev/null, with qt4-mac:
> 
> https://trac.macports.org/ticket/22701
> 
> And git-core:
> 
> https://trac.macports.org/ticket/33127
> 
> The problem occurs when a port compiles a test with "-o /dev/null", because 
> although that will discard the output file, which is what's wanted, it will 
> also *delete* the output file (i.e. delete /dev/null) if the test fails. 
> Since we only saw this problem on Snow Leopard I guess only gcc does that, 
> and clang doesn't.
> 
> Build 11369 was mosesdecoder, and I did find the "-o /dev/null" construct in 
> the file jam-files/sanity.jam. I've just verified on my Snow Leopard test 
> machine that building mosesdecoder as-is does delete /dev/null. Now I'll 
> restart and patch the port to use "-o dummy" instead and I'll try again.

Yes, that worked. Committed in r98496.

Bill, could you please restart the Snow Leopard buildbot computer (or manually 
recreate /dev/null if you know how), then re-run port build 11369 and all 
subsequent port and base builds?

Hiroshi, could you let the developers of mosesdecoder know about this problem 
and the fix?

Thanks.

_______________________________________________
macports-dev mailing list
[email protected]
http://lists.macosforge.org/mailman/listinfo/macports-dev

Reply via email to