On Wed, Jan 4, 2012 at 12:24 AM, peter green <plugw...@p10link.net> wrote: > Luke Kenneth Casson Leighton wrote: >> >> On Tue, Jan 3, 2012 at 10:54 PM, Adam D. Barratt >> <a...@adam-barratt.org.uk> wrote: >> >> >>> >>> (fwiw, the not-yet-built list includes webkit and ruby1.9.1, each of >>> which have a number of other packages directly or indirectly stuck >>> behind them). >>> >> >> >> ahh... webkit. do you have a system anywhere that has 2gb of RAM? >> if not, i strongly strongly advise skipping the debug builds on any >> webkit library. >> >> as webkit is mostly c++ and as pretty much everything goes into one >> massive library with hundreds of object files that are very closely >> inter-dependent, linking the library requires a whopping 1.5gb of >> *resident* memory in order to complete within a reasonable amount of >> time. >> >> anything outside of that - even by a marginal amount - will result in >> the build machine absolutely thrashing its nuts off. >> >> 1gb is not enough (tried it...). 1.5gb will not be enough. 2gb is >> the bare minimum for the link stage > > Well anthiel built the armel webkit package (including debug packages) > successfully and according to the machines database it only has 1.5Gb of > ram. > OTOH the build DID take 33 hours.
:) 90% of that will have been absolute thrashing the nuts off the hard drive during the link phase. increase that RAM to a mere 2gb and it'll drop drastically to about 3 hours, with the link phase taking only about... *guesses* 20 mins instead of 30 hours. > Furthermore the failure on armhf didn't look like a timeout to me, the error > was "ERROR: can't resolve libraries to shared libraries: > javascriptcoregtk-3.0, webkitgtk-3.0". hmmm... those are internal libraries which are private to webkit. they get linked into libwebkitXYZgtk-N.N which is the public library but the symbols in those two libraries must *not* be made publicly accessible (nor the header files made available either). so if the build tries to do anything fancy/special by moving those out of the location they're supposed to be in to stop them being public... just speculation but something to watch out for. l. -- To UNSUBSCRIBE, email to debian-release-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org Archive: http://lists.debian.org/CAPweEDzo3M8rW-qfJT4Peuj_r0p=+M7W+eDc=_yqsh9q7ak...@mail.gmail.com
