Gentoo Archives: gentoo-lisp

From: Neil Funk <infinitedowntime@×××××.com>
To: Cyprien Nicolas <c.nicolas@×××××.com>
Cc: gentoo-lisp@l.g.o
Subject: Re: [gentoo-lisp] Ebuild for Racket 5.1
Date: Wed, 23 Feb 2011 21:42:55
Message-Id: AANLkTikdOy3NUJrtN=ZFQBDZBMdCbAAt1OD+WzX=rp21@mail.gmail.com
In Reply to: Re: [gentoo-lisp] Ebuild for Racket 5.1 by Cyprien Nicolas
I tested the new ebuild in the tree, with USE="X cairo foreign
(multilib) opengl -backtrace -cgc -doc futures places -plain -profile
-sgc -slatex -static" on ~amd64.
Worked (compiled) fine.

Next I enabled the "backtrace" flag - also worked fine.

But when I enabled the "profile" flag, the build hung as per my
previous post regarding 5.0.2. (Only it turns out that "profile" was
the culprit - the only one of those flags I didn't mention
originally.)  Last console message was "Compiling xform support...
Done making xform-collects."
lt-racketcgc taking up 100% CPU.



On Tue, Feb 22, 2011 at 2:51 PM, Cyprien Nicolas <c.nicolas@×××××.com> wrote:
> Hello lispers, > > I pushed an ebuild for racket-5.1 in the lisp overlay. I updated the > patch, it is needed for plain install to work (i.e. to skip the > compilation of rackets libraries, the long setup-racket phase). > > It has been lightly tested on my stable x86 laptop and stable > amd64-multilib desktop. At least it compiles without failure. > > Here are a few comments and thoughts I want to share: > > * In 5.0.2, I sped up the "raco setup" phase by adding the -j options to > the PLT_SETUP_OPTIONS. I added this to the install-cgc target in 5.1. > > * I renamed the desktop entry from DrScheme to DrRacket, as it's the new > name of the IDE. > > * I removed the build directory, in favor of a build directly in the > source directory. Such use of a build directory is not needed as the > source code is removed after the merge, it avoids unnecessary copying of > files. > > * 'slatex' nor 'static' USE flags have been enabled during my builds, so > it may just work, or not. > > > 1. Does enable cgc (or sgc) should disable the build for 3m? If not, we > could add a default enabled 3m USE for building the 3m-gc-based racket > (currently the default case). It is easy to add this possibility. > I'm fine by just using 3m, I don't know if there is specific users needs. > > 2. A bug has been opened about racket and PaX, but I don't know anything > about PaX, so I'll let somebody else to take care of the bug. > https://bugs.gentoo.org/show_bug.cgi?id=355285 > > 3. If someone has a hint on how to fix/hide the > "configure: WARNING: unrecognized options:"... > I didn't look into it too much yet. > > > Regards > ,Cyprien > > > PS: USE=cgc caused a high memory consumption on my desktop during the > raco setup, one build has even been killed because my system ran out of > memory, so use it with caution, or expect slow downs, if you don't > have/use the autogroup scheduler :) (I don't have a tmpfs PORT_TMPDIR). > > PS2: Thanks to Morgon and Neil for their feedback. and Dimitri for the > initial racket ebuild. > > > On 02/22/2011 08:05 PM, Morgon Kanter wrote: >> Not sure what's different between our systems, but I had both futures >> and places available (though not backtrace) and my build went just >> fine. Actually, my use flags were: X cairo doc foreign futures opengl >> places (everything else was unset). >> >> Some of these flags need documentation. >> >> -- Morgon >> >> On Tue, Feb 22, 2011 at 12:53 PM, Neil Funk <infinitedowntime@×××××.com> wrote: >>>> May I ask you which USE flags combination(s) you tried it with, and on >>>> which arch? >>> >>> I'm on ~amd64. >>> dev-scheme/racket-5.1 was built with the following: >>> USE="X cairo foreign (multilib) opengl -backtrace -cgc -doc -futures >>> -places -plain -profile -sgc -slatex -static" >>> >>> The only other suggestion I'd offer is regarding the backtrace, >>> futures, and places flags.  At least on 5.0.2, one or more of these >>> cause the build to hang forever at 100% CPU.  Only when I finally gave >>> up and killed the build did I notice the ewarn that these flags can >>> cause that.  I realize there's only so much we can protect people from >>> themselves, but it would have been nice if it made me acknowledge that >>> I intended to use a risky flag.  Ideally, we'd just fix whatever >>> causes it to hang, but that's an upstream issue - could we maybe just >>> mask those flags?  Although, I haven't tested them on 5.1 yet to see >>> if they're even still a problem... >>> >>> Cyprien, I'd be happy to help test with other configurations.  Thanks >>> for your work. >>> >>> -Neil >>> >>> On Tue, Feb 22, 2011 at 5:32 AM, Cyprien Nicolas <c.nicolas@×××××.com> wrote: >>>> On 21/02/11 18:07, Morgon Kanter wrote: >>>>> Hey folks, >>>>> >>>>> I noticed there wasn't yet an ebuild for the new version of Racket, so >>>>> I went ahead and made one. It's attached. >>>>> >>>>> A couple changes besides just bumping the version number: >>>>> * Doesn't really seem like that makefile patch was required anymore (I >>>>> didn't need it), so I commented that out. >>>>> * The new Racket GUI system doesn't appear to include any of those >>>>> external libraries that the ebuild used to issue an "rm" for, so I >>>>> removed those bits from the ebuild (as well as the formerly >>>>> commented-out rm commands that were replaced by something different). >>>>> >>>>> That's it for the changes, aside from a version bump. >>>>> >>>>> Thanks, >>>>> -- Morgon >>>> >>>> Actually, there is a work-in-progress ebuild, but I didn't had the time >>>> to test it intensively, especially with the cgc and sgc garbage >>>> collectors, and the plain USE flags (as the makefile patch is related to >>>> those USEs), and I just it an "emake failed" with USE=plain. >>>> >>>> Apart from some cosmetic changes (EAPI4, REAUIRED_USE="cgc? ( !sgc )", >>>> some spacing) my ebuild is the same :) >>>> >>>> May I ask you which USE flags combination(s) you tried it with, and on >>>> which arch? (applies for Neil too). >>>> Thanks for the feedback >>>> >>>> Regards >>>> >>>> ,Cyprien >>>> >>>> >>> > >