On Tue, 26 Jul 2011 22:19 -0700, "jerry gay" <[email protected]> wrote: > On Tue, Jul 26, 2011 at 19:27, Moritz Lenz <[email protected]> wrote: > > On 07/27/2011 02:33 AM, James E Keenan wrote: > >> I object to this change for two reasons. > >> > >> First, on one platform I use, Parrot does not PASS 'make test' when I > >> configure with '--optimize'. > > > > So there's a bug, and it must be fixed. If the fix isn't easy, I'm fine > > with documenting the lack of '--optimize' as a workaround for that platform. > > I don't see why a platform-specific bug should mean we should recommend > > our users a slow parrot by default. I don't think slow-by-default does > > parrot any good. I'd go so far as to actually make --optimize the > > default for Configure.pl. > > > >> Second, I think a change in our README about how our users ought to > >> start out building Parrot really warrants more discussion than what > >> little I could find on #parrot today. > > > > I don't see how the need for more discussion should prevent a gradual > > improvement. > > > >> It appears to have been a > >> spur-of-the-moment decision. I think this should have been a Trac > >> ticket with type RFC. > >> > >> I would really like to see this reverted until we can discuss it more > >> thoroughly. > > > > If you feel strongly, feel free to revert that commit, but as I argued > > above, I can't see the reason behind either of your points. > > > if, as you argue, it should be the default, why must it be specified > as an option? > > parrot users deserve better; parrot developers should do work with > users in mind rather than asking them to do it. don't make users > deviate from the norm in order to get normal behavior. > > ~jerry
I strongly agree with the sentiment here. If an optimized Parrot is broken on a platform, we need to fix that bug. kid51++ mentioned the test in question is tt #1930. Unfortunately Darwin/PCC, where the test fails, is becoming a rare platform, but if we want to call it "supported", we need to fix it. I also want users to get the best possible Parrot without requiring them to know to pass --optimize to Configure.pl. It makes sense for a git clone not to be optimized by default because it's more likely to be broken and in need of debugging (which optimization makes harder). As an alternative, I propose that we make all release tarballs build with optimizations enabled by default. This would mean we'll need more emphasis on testing optimized Parrots in the week before a release, but I don't think this will pose a meaningful difficulty. Thanks, Christoph _______________________________________________ http://lists.parrot.org/mailman/listinfo/parrot-dev
