Re: Fw: failed sparc build of pandoc 18.104.22.168-1
+++ John MacFarlane [Jul 01 12 18:08 ]:
> +++ Joachim Breitner [Jul 01 12 22:18 ]:
> > Hi John,
> > I’m wondering: Why did you add -threaded then? I’d expect the threaded
> > RTS to actually be slower unless you make (good!) use of multiple
> > threads...
> I'm afraid I didn't have a good reason. I figured it would be good
> to have the flexibility -- and perhaps future changes to pandoc could
> make better use of it. I did benchmark a bit and found that specifying
> multiple cores with RTS options actually made pandoc run slower.
> I did not try benchmarking the version linked with -threaded but run
> with one core against a version linked without -threaded.
I have tried this, and there seems to be a slight but measurable
performance penalty. Accordingly, I have removed -threaded from
the default ghc-options upstream. It's probably not worth updating
the debian packages, except perhaps in the case of sparc, since
the change might allow pandoc to compile there.