IRC log for #crosstool-ng on 20110428

06:35.10*** join/#crosstool-ng kos_tom (~thomas@col31-4-88-188-83-94.fbx.proxad.net)
06:37.06*** join/#crosstool-ng y_morin (c1317c6b@gateway/web/freenode/ip.193.49.124.107)
06:37.06*** mode/#crosstool-ng [+o y_morin] by ChanServ
14:18.00*** join/#crosstool-ng jero (~jero@2607:fad8:4:0:62eb:69ff:feca:c4f7)
16:39.49*** join/#crosstool-ng y_morin_ (~ymorin@ARennes-256-1-36-148.w90-32.abo.wanadoo.fr)
16:39.49*** mode/#crosstool-ng [+o y_morin_] by ChanServ
18:20.59*** join/#crosstool-ng kos_tom (~thomas@humanoidz.org)
18:55.15*** join/#crosstool-ng linuxjacques (6217fec1@nslu2-linux/jacques)
18:56.34linuxjacquesI'm trying to build the powerpc-e500v2-linux-gnuspe sample from the latest HEAD and it's failing on PPL with "configure: error: Cannot find GMP version 4.1.3 or higher."
18:56.50linuxjacquesyet it already built GMP 5.0.1
19:28.45y_morinlinuxjacques: revert to using GMP 4.3.2.
19:30.11linuxjacquesy_morin|away: yeah I tried that - same error
19:30.39linuxjacquesnow trying without GRAPHITE which removes need for PPL
19:33.54linuxjacqueshmm, it looks like the PPL configure sub-executable (conftest built by configure to check GMP version) was segfaulting
19:34.23linuxjacquesI have these in my messages log: conftest[2258] trap divide error ip:400518 sp:7fffe99b26d8 error:0 in conftest[400000+1000]
20:03.43y_morin|awaylinuxjacues: that's perfectly normal. In the process of testing stuff, ./configure generates some executables and runs them. If some do segfault, it can be expected behavior. It happens here as well.
20:20.31linuxjacquesy_morin|away: yeah it _could_ be normal, or it could be the conftest which is supposed to return a value based on the version of GMP - hard to tell. Either way, I can't get PPL to pass its configure test.
20:20.56linuxjacquesthe toolchains built without GRAPHITE / PPL
20:21.00linuxjacquestoolchain
21:20.34y_morin|awaylinuxjacques: What's your host? Gentoo with PAX?
21:21.04linuxjacquesy_morin: this time it's Fedora14 X86_64
21:21.10linuxjacques(I do use gentoo at home)
21:21.41linuxjacquescould it be picking up the host (Fedora) GMP libs?
21:21.52y_morinCan you look at the config.log from PPL and see what it was trying to do at the time of breakage?
21:22.04y_morinlinuxjacques: yes it can do tht as well.
21:22.09y_morin*that
21:22.47linuxjacquesy_morin: yeah I can find it again - it was building a little conftest which just linked a GMP lib (IIRC) and ensured the version was >= 4.1.3
21:23.22linuxjacquesthe Fedora GMP would have been built with gcc 4.5.1 - not sure if that has a different ABI than 4.6.0 (which crosstool-ng is building)
21:23.34y_morinYes, I know. But in the config.log, you'll see exactly what was going on: the program, the compile command line.
21:23.50y_morinThen you could redo the same build, but with the -Wl,-verbose flag
21:24.00y_morinAnd then you'd see where it finds the library.
21:24.17y_morinTo rule out it picking the system library instead of our own version.
21:24.42linuxjacquesy_morin: Ok looking
21:25.08linuxjacquesbut how would I be sure the env was the same between what I do on command line and crosstool-ng build?
21:25.23linuxjacquessometimes there are lib paths etc.
21:25.28y_morinThe env has nothing to do when running gcc and ld.
21:25.36y_morinAll that matters is the arguments on the command line.
21:26.17y_morinYou'll see something like: 'failed program was' just after the last failure (other program failures are OK).
21:28.11linuxjacquesconfigure files aren't exactly ... readable - looks like this check starts on line 11163 of the PPL configure
21:28.30y_morinlinuxjacques: no, I'm speaking about the config.log, not the ./configure...
21:29.17y_morinjust search for the error message.
21:29.52linuxjacquesy_morin: aah, OK
21:31.35linuxjacquesI used the same build dir to build without PPL, so it looks like it wiped out the ppl build dir - let me re-config and try it again (should only take a couple of minutes)
21:31.50y_morinOK, no problem...
21:33.22linuxjacquesit's building GMP now - then maybe one other package, then PPL
21:33.36y_morinGMP -> MPFR -> PPL
21:33.37linuxjacquesyeah it's GMP, MPFR, PPL
21:34.20linuxjacquesOK, it failed on PPL, and there's a fresh segfault in my messages log - looking for PPL config.log now
21:36.43linuxjacquesI believe this is the relevant part of the PPL config.log: http://pastebin.com/fQqBS7aC
21:37.46linuxjacquesthis is interesting: GMP header (gmp.h) and library (ligmp.*) version mismatch: header gives 4.3.1; library gives 5.0.1.
21:38.02linuxjacquesthe Fedora version of GMP is 4.3.1
21:38.17linuxjacquesso it's finding the correct lib but wrong headers?
21:38.23y_morinAnd you configured 5.0.1 in crosstool-NG, right?
21:38.45linuxjacquesy_morin: int this case, yes, though I have tried 4.3.2 as well
21:39.14linuxjacquesmaybe I should remove the Fedora gmp-devel package
21:39.56y_morinIn fact, if you look at the conftest compile line (6 in the pastebin), you'll see it does not pass any -I flag.
21:40.26y_morinNo, companion libraries are becoming mainstream in current distors, and we have to support that as well.
21:40.35y_morin*distros
21:40.56linuxjacquesyeah I didn't feel right suggesting it :-)
21:41.19y_morinAnd because it does not pass any -I flag, it includes gmpxx.h from your host include path, not our.
21:41.34y_morin(line 98 in the pastebin)
22:01.09linuxjacquesy_morin: as a test, I removed the Fedora gmp-devel package, and now the PPL build fails with: conftest.cpp:84:19: fatal error: gmpxx.h: No such file or directory
22:01.22linuxjacqueswhich I guess would be expected with no -I
22:27.12linuxjacquesre-installing the Fedora gmp-devel package and choosing GMP 4.3.1 in crosstool-ng allows PPl to build
22:38.05bhundveny_morin: have you tested building x86_64 on i[x]86 with static toolchain?
22:39.30y_morinbhundven: probably not, I can't remember. I mostly do real crosses here (eg. ARM and MIPS).
22:39.38bhundvenok
22:40.19y_morinSo if you tell me it is broken, I wouldn't be too surpised. Disapointed, maybe; surprised, no. :-/
22:40.26bhundvenheh
22:40.57bhundvenwell, I'm still trying to figure out the actual problem, so I'm not raising any alarms yet..
22:41.26y_morinbhundven: but if you have a fix, tonight is your last chance at it before the release: I am rebuilding all samples before the release on Saturday evening.
22:41.39bhundvennot yet... just getting into it.
22:41.48bhundvennext release ;)
22:42.10y_morinAnd building all of them takes roughly 16h on my machine, so I'm not ready to handle a third or second build...
22:42.16y_morinYep, probably! ;-)
22:43.52y_morinhates Java: the Java frontend takes 20min to build...
22:44.43y_morinlinuxjacques: I'm not too sure what to expect at this step (I mean no -I). It works for me here.
22:45.19y_morinlinuxjacques: I guess that, although you removed the -devel package, you still have the runtime (eg. the libgmp.so.x.y.z is still here)
22:45.51y_morinIn this case, it can still link with it. It's just the headers it does not find... :-(
22:45.57y_morinWhich is weird.
22:46.00linuxjacquesy_morin: yeah removing the runtime would probably want to remove the compiler
22:46.10y_morindefinitely, yes. Sad...
22:46.32y_morinOTOH, I am working on making the complibs in crosstool-NG optional, in case the host system already have them.
22:47.03y_morinBut that's not ready, my host still does not have them (I'm still on Debian Lenny, and not ready to upgrade).
22:47.21linuxjacqueslol, doing "yum erase gmp" wants to remove just about every package on the system
22:47.31y_morinHe...
22:48.31linuxjacquesFedora rawhide has 4.3.2, maybe I'll build / install that
22:53.26y_morinlinuxjacques: probably, yes. It is built with the host compiler, so it should be the ABI.
23:00.59*** join/#crosstool-ng linuxjacques_ (6217fec1@nslu2-linux/jacques)
23:01.08linuxjacques_got disconnected :-\
23:02.07linuxjacques_I built/installed 4.3.2 from the rawhide srpm and now PPL is happily building against that version
23:02.29y_morin\o/
23:14.35*** join/#crosstool-ng mnt_real (~mnt_real@bas1-montreal43-1177755902.dsl.bell.ca)

Generated by irclog2html.pl Modified by Tim Riker to work with infobot.