**** BEGIN LOGGING AT Sun Mar 29 03:01:48 2020 Mar 29 10:29:22 Finally I took a sledgehammer to the problem, and just modified lib/bb/checksum.py to ignore .git subdirectories whenever checksumming anything Mar 29 10:30:03 A non elegant assumption that .git/ directories are never meaningful input to a build, I suspect this will fix the needless rebuilds Mar 29 13:57:53 Hi there, I'm still trying to build scikit-learn for my device (cortexa9). scikit needs 2 dependencies, numpy and scipy. Building numpy package (normal and native) works great, but building scipy is much painful. I managed to build it though, but when trying to build scikit-learn again, I got an error: "scipy is not installed" during do_compile. I guess I also need to build native package for scipy ? But I have no idea how can I do Mar 29 13:57:53 that. I tried to add 'BBCLASSEXTEND = "native"' in my scipy recipe (also add it to lapack and libgfortran), but then it complains on unbuildable dependency chain (Missing or unbuildable dependency chain was: ['python3-scikit-learn', 'python3-scipy-native', 'lapack-native', 'libgfortran-native', 'gcc-runtime-native']). Do you have any idea how to fix it please ? Mar 29 14:01:34 not sure my flow is correct too Mar 29 14:02:36 If I add the BBCLASSEXTEND to gcc-runtime-native, then I got this dependency error: (Missing or unbuildable dependency chain was: ['python3-scikit-learn', 'python3-scipy-native', 'lapack-native', 'libgfortran-native', 'gcc-native-x86_64']) Mar 29 14:05:11 Or this ... ['python3-scikit-learn', 'python3-scipy-native', 'lapack-native', 'libgfortran-native', 'virtual/gcc'] Mar 29 14:10:37 more yocto at foss-north: https://youtu.be/FZOdbAJqDNY Mar 29 15:14:51 * gtristan not sure if it's entirely upstream vector making life horribly difficult or what it is Mar 29 15:16:08 here I have a hand full of CMake modules which do export(PACKAGE ${LIBRARY_NAME}), essentially what this is doing is storing some state in ${HOME}/.cmake/packages Mar 29 15:17:01 This state records the *current* build directory (like as if that is going to exist when reverse dependencies build), into the HOME dir so that reverse dependencies can find their include files Mar 29 15:17:43 So of course, when those dependencies are not built this time, but instead are stored in sstate cache, the build fails because those ephemeral build directories dont exist that run Mar 29 15:17:47 * gtristan palmface Mar 29 15:18:16 * gtristan looks for another sledgehammer fix Mar 29 15:18:44 maybe I can set those to the adjacent git directories where the sources are guaranteed to be, using some obscure cmake env var incantation Mar 29 15:21:54 probably I have to make a run through a dozen or so CMake modules and downstream patch them to correctly install their header files into ${prefix} Mar 29 16:15:19 gtristan: things definitely shouldn't be storing state in $(HOME) :( Mar 29 17:26:19 RP: hey, if I have a metalayer which changes e.g. mesa version to a newer one, can that still somehow pass yocto-check-layer ? Mar 29 17:45:22 marex: I guess it would depend how and when it does that but it could be tricky. It kind of defeats some of the aims of compatibility :/ Mar 29 17:58:43 Marex: maybe if the new recipe is tagged to not default by setting DEFAULT_PREFERENCE = "-1"? Then PREFERRED_VERSION would need to be explictly set to get it Mar 29 18:11:52 RP: well, I need to update mesa in zeus, because etnaviv patches Mar 29 18:12:06 smurray: that could work Mar 29 18:27:53 Hi guys, I am learning about update mechanisms on Yocto and I was fascinated by the Clear Linux approach. I now see that meta-swupd isn't actively maintained, what happened? Mar 29 18:39:30 kriive: Intel cut its resourcing on OE/Yocto stuff substantially Mar 29 18:51:22 smurray: too bad Mar 29 18:51:40 kriive: yep Mar 29 18:53:17 So basically, which software update should I investigate on, if I wanted small delta updates and relatively easy setup? OS-Tree? Mender? Mar 29 18:54:47 I am alone in developing and implementing all the stuff, it would be easier for me to have less components to take care of Mar 29 21:16:53 kriive: OS-Tree perhaps is a good option for what you want meta-swupdate is another option Mar 29 22:44:03 Thank you khem, I'll look into it **** ENDING LOGGING AT Mon Mar 30 03:00:00 2020