TODO (Prio A) ============= [.] ipkg with dependecies update: 20090614 mkl: ipkg are created with dependencies now, image creation still works. however online update is still untested, though. [ ] mol, rsc: improve user database concept. Especially for the system users we currently have the problem how to generate system users which are only necessary for certain services. Maybe some kind of /etc/passwd.d/* mechanism could be a solution? Discuss with team. Discuss UUID/GID of "rpcuser" and other important system users [ ] rsc: GDB__VERSION: shouldn't this be taken from the toolchain, in order to be consistent with the gdb which is provided there? Discuss with mkl. [ ] jbe: If someone adds a project local menu (*.in) file and uses 'project_specific' as its SECTION name ptxdist should automatically generate an additional submenu "Project Specific Configuration" and include this menu into it. This should avoid a local "Kconfig" copy in each ptxdist project for such a simple addition. [ ] lfu: some mountable devices are available after udev has started. hence we need an extra mount script to mount all entries in fstab instead out of rcS. [ ] jbe: all test scripts should check if their interaction with the filesystem where successfull. If a file copy to /tftpboot fails, it should *not* continue. TODO (Prio B) ============= [ ] jbe: mplayer.compile: It seems very PXA specific now? i586-unknown-linux-gnu-gcc [...] -c -o vo_pxa.o vo_pxa.c vo_pxa.c: In function 'draw_osd': vo_pxa.c:964: error: unknown register name 'r5' in 'asm' vo_pxa.c:964: error: unknown register name 'r4' in 'asm' [...] vo_pxa.c: In function 'config': vo_pxa.c:964: error: unknown register name 'r5' in 'asm' vo_pxa.c:964: error: unknown register name 'r4' in 'asm' [...] make[2]: *** [vo_pxa.o] Error 1 make[2]: Leaving directory `[...]/platform-i586/build-target/MPlayer-1.0rc2/libvo' --> menu contains pxa specific stuff. Maybe this should be modified? Even with disabled PXA specific options a build for x86 still fails: libvo/libvo.a(video_out.o):(.data+0x20): undefined reference to `video_out_x11' libvo/libvo.a(video_out.o):(.data+0x24): undefined reference to `video_out_xover' libvo/libvo.a(video_out.o):(.data+0x28): undefined reference to `video_out_pxa' [...] and much more Update 20090118 rsc: not reproducable, switched on again and we wait for sisyphus to explode. [ ] http://www.osadl.org/Single-View.111+M52e8d0c2b2f.0.html [ ] rsc: add "ptxdist export_bsp ", similar to "ptxdist export_src ". Structure: /README.txt /LICENSE.txt /Autostart.bat (optional) /index.html (optional) /install.sh /documentation/ /images/oselas.bsp-phytec-phycore-12-root-full.jffs2 (optional) /images/oselas.bsp-phytec-phycore-12-root-headless.jffs2 (optional) /keys/ (optional) /packages//foo_1.2.3_armel.ipk (optional) /sources/public/md5sums /sources/public/foo-1.2.3.tar.gz /sources/public/foo-1.2.3.tar.gz.md5 (optional) /sources/public/foo-1.2.3.tar.gz.asc (optional) /sources/private/bar-0.0.1.tar.gz /sources/unknown/baz-42.23.0.tar.bz2 /oselas/ptxdist-1.99.12.tar.gz /oselas/ptxdist-1.99.12-patches.tar.gz /oselas/oselas.bsp-phytec-phycore-12.tar.gz /oselas/oselas.toolchain-1.99.3.2.tar.gz /oselas/oselas.toolchain-1.99.3-arm-v5te-linux-gnueabi-gcc-4.3.2-java-glibc-2.8-binutils-2.18-kernel-2.6.27-sanitized_i386.tar.bz2 /oselas/oselas.toolchain-1.99.3-arm-v5te-linux-gnueabi-gcc-4.3.2-java-glibc-2.8-binutils-2.18-kernel-2.6.27-sanitized_i386.tar.deb - take care of all platforms: configs/*/platformconfig - images: images for all collections - for all optional packages: ask back if images, ipkgs etc. shall be bundled (with --force) [ ] jbe: Shown pathes are much too long in "ptxconfig" and "platformconfig" while running "ptxdist menu" [ ] jbe: We should provide a solution if someone selects "ptx" or "platform" while running "ptxdist menu" and these files are not new enough. (happens if someone wants to upgrade his ptxdist-1 based BSP). Currently there is only a error message, one should use "--force". Its correct, but is doesn't help! M$ help level! -> integrate --force into dialogs [ ] jbe: most of the time thttpd does not respond to the USR1 signal. - the start/stop script fails to stop this service [ ] bbu, lfu, rsc, sha: "bspfix feature" Usecase: * create BSP config with "ptxdist menuconfig" * "ptxdist bspfix [bspfix-name]" (creates 2nd config internally) * user makes changes to the software config (menuconfig) and saves * some kind of "diff" is made between "selected_platformconfig" and bspfix-config * if user has one or several "bspfixes" around, they are patched into his ptxconfig on the fly Aims: * developer: can create their own local modifications of the bsp, without destroying the official one * customer: can make his own changes ontop of our releases; changes are easy to move forward Before Release -------------- [ ] rsc: we need a "tested distros" strategy for 2.0.0, with pre-packaged toolchains. [ ] rsc: run dependency hunter and fix all remaining dependency problems [ ] update ChangeLog from git [ ] jbe, rsc: provide demo projects, including website documentation, maybe even elinux.org docs. Release tarballs in a way that they can be installed with ptxdist, which means that they have to be installed into the "projects/" directory in the toplevel dir, to be able to clone them. Provide debian packages of ptxdist, ptxdist-patches, toolchains and demo directories. [ ] bbu: check ptxdist first-time-usage scenario for chicken-egg problems e.g. if you don´t set sane defaults for PTXCONF_SETUP_PROJECTPATH, ptxdist won´t find its own projects. [ ] jbe: Would there be a solution to supress sysroot-* folders for target system (for example bare metall) that do not need a Unix filesystem? [ ] jbe: it must be possible to force some userland components from inside the platformconfig. For example a platform specific kernel driver module needs a special component of the UDEV. Currently it does not work. Duplicating symbols in both config files may work, but it will be "BROKEN BY DESIGN". Update 20091017 rsc: Would it be possible to add that platform specific package in the platform? [ ] rsc: configure continues, even if a test breaks. Example: checking lex library... sed: couldn't write 76 items to stdout: No space left on device grep: write error: No space left on device sed: couldn't write 46 items to stdout: No space left on device none needed checking whether yytext is a pointer... sed: couldn't write 29 items to stdout: No space left on device no [...anything proceeds...] [ ] jbe: working with hd.img is very delicate. The incomplete partition table confuses the linux kernel, so any rewrite of the image after done one time fails with "no space left on device". Also fdisk fails when trying to repair this partition table. Only Windows can rewrite the partition table in this case. After this, also Linux is able to write to the hard disk again! But the BIOS and Kernel on target side are successfull. Problem happend on SUSE-10.1 on host side and an USB to IDE adapter to initial write the hd.img to the harddisk. TODO somewhen ============= [ ] rsc: evaluate http://dolt.freedesktop.org [ ] rsc: license management: - generate license dependency graph (based on objdump info?) - bundle GPL sources automatically [ ] rsc: add sparse support [ ] jbe: review samba configure stage. For a non-root and cross-compile run it seems to omit many tests and switches of many features in the final application. Check if samba_cv_HAVE_IFACE_IFCONF=yes and samba_cv_HAVE_MMAP=yes can be used all over all architectures as defaults [ ] bbu: add minimal documentation for internal vars (useful for customer makefiles) -> see manual (not quickstart) [ ] jbe: samba fails to run with: sys_gethostbyname: Unknown host. eth0 can't determine netmask for eth0 WARNING: no network interfaces found Seems to be a failure with current glibc [ ] jbe: audit packages which are marked to be BROKEN [ ] lfu: often used config files for X like xorg.conf or xinitrd shall be integrated into ptxdist generic files. the installation shall follow the scheme of other rootfs configuration files [ ] jbe: audit all packages if they change the target CPU (with --march or --mtune); openssl does so [ ] jbe: uRamdisk should be built from ipkg packages instead of root/ content! [ ] jbe: why does fam-2.7.0 always try to run the autotools? It fails on most modern systems -> mark as broken. rsc: real fix: find out why and fix. [ ] feature request: add profiling support (kernel and userspace) best with host visualisation [ ] jbe: more help texts [ ] bbu: reimplement cuckoo-test - files w/ wrong build-arch - missing libs - statistics - stripped / non stripped files [ ] bbu: reimplement generic pre- and post-flight checks - dependency checks - cuckoo-test - user defined checks - package tests - in-system test -> ltp [ ] bbu: generic logging aproach w/ single and combined logs - combined log for normal use - build-logs per package, test-suite, etc would support features like - "quiet build", while still providing the full error output of the failing target - helpdesk notification fetaures - build statistics - better support for GUI driven builds [ ] bbu,rsc,mkl: brain storming session Discussion about the whole PTXdist build-, deployment- and testing-mechanics: - Support different target package formats Rationale: - Some standardization aproaches like carrier grade linux or the linux standard base make rpm or dpk packages mandatory. - Users asked for it Which package formats would be useful ? - ipkg - deb - rpm - openpkg - plain directory tree / tar[.gz] ! Depends: Check rpm and deb database footprint ! Depends: Discussion about the current installation scheme - Support binary package distribution Rationale: Most Customers don´t want to build the base System on their own. They want a well defined base system and then apply some changes, add files (probably on the fly) by installing binary packages. ! Problem: This might break the current PTXdist philosophy (executeable documentation) ! Depends: We would need some kind of install-filter or exclude-list - Deployment / Testing Frontend (maybe GUI driven) Rationale: needed in a binary package distribution scenario to customize the installation from a generic binary package. - Support precompiled customer environment Rationale: This would reduce compile times for the customer, since he would start directly w/ his own application w/o the need for a full ptxdist run. ! Problem: Build Hierarchy is not relocateable -> libtool problems ! Problem: Make depends on timestamps Possible solutions: - create intermediate build labels / paths - tweak things during installation (dirty, dirty ... how do we handle updates?) - Provide a "full blown" toolchain with all host-tools and chroot into it (dirty, dirty ... needs root privileges) - Provide a "full blown" toolchain with all host-tools inside a virtual machine, vmware image or CoLinux (dirty, slow, but runs even with Windows) (nasty, since it would be some kind of closed box solution) (problem: how do we handle alternative toolchains in this case?) - Support Partitioned Images ! Problem: We don´t want to use root privileges ! Problem: Generate valid Partition Tables ! Problem: ipkg list has to work inside the target system, so we would need two or more diffenrent installation "views": - The partition image generation view, which a) respects partition boundaries b) rewrites the resulting installation paths respectively. - The binary target package view (i.e. ipkg) which a) generates apropriate pre- and post- installation scripts to i.e. remount the partitions read/write b) creates binary packages, rooted in "/" on the target. - An optional "user-view", where something else can happen, i.e. transfering stuff to an ftp server or mounting an image with root privileges or as user via fuse. - TODO: - Create Use Cases: - What is covered by the current design - What would be nice, well designed and sane - What does $JON_DOE_HACKER want - What does $JON_DOE_HACKER _need_ - What does $CUSTOMER want - What does $CUSTOMER _need_ - Draw the big picture :-) - Create Roadmap [ ] mol: add possibility to 'ptxdist print' more than one variable in one make call bbu: maybe some kind of "ptxdist env" as in env(1) ? [ ] check if ${PTXCONF_SYSROOT_TARGET} can be written to and barf if not [ ] add sanity checks for all user configurable variables (especially paths) [ ] Problem: When doing things like "PKG_CONFIG_PATH=bla something", something does not inherit the environment variable PKG_CONFIG_PATH. To achieve this properly, we need to export these variables to the calling shell. The same applies to lots of variables from Rules.make. [ ] image build mechanism tries to access data and infos from already disabled targets! (if they where enabled before). It does so unless a manually delete of the remaining and (now) useless targets in "state/" [ ] Tree Structure rsc: Find sane place for generic files from projects/generic ero: Make subdirectories, e.g. for x.org stuff. Might use symlinks to avoid changing the scripts. sha: push make and in files into subdirs, add patches to that dirs as well. [ ] rsc: add some useful statistics at the end of build (root size, image size etc.). [ ] rsc: add checksums for download Interesting packages ==================== [ ] rsc: http://en.opensuse.org/Icecream [ ] rsc: http://cryptmount.sourceforge.net/ [ ] rsc: http://stapgui.sourceforge.net [ ] rsc: bashdiff: http://home.eol.ca/~parkw/index.html [ ] rsc: http://podgorny.cz/moin/UnionFsFuse [ ] rsc: http://clpbar.sourceforge.net/ [ ] rsc: http://xmlrpc-c.sourceforge.net/ [ ] rsc: http://www.vanheusden.com/httping/ [ ] rsc: http://urunlevel.sourceforge.net/ [ ] rsc: http://www.freshmeat.net/projects/bootutils [ ] rsc: http://www.reportlab.com/ftp/ [ ] rsc: http://rlib.sicompos.com/ [ ] rsc: MatrixSSL [ ] mkl: lilo, tinylogin targetinstall (rsc: find out what MKL means here) [ ] rsc: finish integration of quagga Test suite components to be added ================================= [ ] rsc: mutrace (mutex trace) [ ] rsc: http://ltp.sourceforge.net/tooltable.php [ ] rsc: Add more test suite stuff: http://gcc.gnu.org/install/test.html http://www.plumhall.com/stec.html [ ] rsc: lmbench [ ] rsc: audit for LSB compatiblity; are there any handable tests for checking compliance? [ ] rsc: have a look at http://asic-linux.com.mx/~izto/checkinstall/ DONE ==== [x] rsc: fix python 3.x update 20100428 rsc: done by mol [x] rsc: 'ptxdist images' for the i586 target sometimes does not work, the generated image is incorrect and grub stops booting. update 20100428 rsc: done [x] rsc: multi stage BSPs: platform team releases a BSP, users use it. May involve 'apt-get install OSELAS.BSP-phyCORE-12' update 20100428 rsc: done [x] jbe, rsc -> mgr: provide a "bare metal" project for Stellaris/Cortex-M3, to show how to use ptxdist-2 for non-Linux targets. update 20100428 rsc: done [x] rsc: verify 'ptxdist newpackage' mechanics. Check all use cases, especially when creating new packages in ptxdist itself (not in the project). Workspace check needs more sophistication. update 20100428 rsc: seems to be fixed in the meantime [x] rsc: port python 2.6 patches to 3.x -> mkl update 20100428 rsc: done by mol [x] Write macros $(call compile) and $(call configure). update 20100419 mol: done [x] libxslt crypto support is broken update 20100419 mol: done [x] jbe: org-server fails if no xkbfile is selected update 20100419 mol: not sure what the problem was, but it seems to be gone. [x] jbe: alsa-lib creates funny pathes at the install stage update 20100419 mol: done [x] rsc: xorg-app-xkbcomp needs xorg-lib-xkbfile, but dependency doesn't work update 20100419 mol: not sure what the problem was, but it seems to be gone. [x] rsc: xorg-data-xkbdata needs xkbcomp on the host - otherwhise we get a circular dependency ... update 20100419 mol: obsolete: xorg-data-xkbdata is gone. [x] bbu: add more real GNU coreutils update 20100419 mol: done [x] mol: bash configure --enable-job-control does not work when cross compiling posible fix in: https://iocaste.extern.pengutronix.de/svn/oselas/bsp/johndeere/OSELAS.BSP-JohnDeere-ComPC-trunk/rules/bash.make explained here: http://www.mail-archive.com/bug-bash%40gnu.org/msg03107.html update 20100419 mol: done [x] rsc: http://www.nongnu.org/failmalloc/ [x] rsc: sysstat: http://perso.wanadoo.fr/sebastien.godard/ [x] rsc: ethtool: http://sourceforge.net/project/showfiles.php?group_id=3242&package_id=19201 [x] rsc: ntpclient: http://doolittle.faludi.com/ntpclient/ [x] rsc: openntpd: http://www.openntpd.org/ [x] rsc: install e2fsck [x] rsc: LTT-ng [x] download strategy: oe can be configured to download from a central server first and only fallback to the internet if packet is not there Update 20091019 rsc: The community has found that we already have this feature :-) [x] ringlej. When the architecture is arm-ixp-big this message occurs: $ ptxdist kernelconfig /rules/ffmpeg.make:189: Please define the config options for this CPU type! 20090625: fixed with: git: 2724dc0cd74b716b1ef770d847a5509291ce4bef svn: 10028 [x] rsc@thebe:OSELAS.BSP-Phytec-phyCORE-trunk$ ./p compile sw-ek-lm3s3748.compile /home/rsc/git/ptxdist/scripts/libptxdist.sh: line 125: unset: `PTXCONF_SW_EK_LM3S3748.COMPILE': not a valid identifier ptxdist: error: 'sw-ek-lm3s3748.compile' is not selected in /home/rsc/svn/oselas/bsp/phytec/phyCORE/OSELAS.BSP-Phytec-phyCORE-trunk/selected_ptxconfig PTXdist: fatal error ... cannot start, sorry! update: 20090614 mkl: [frogger@hardanger:OSELAS.BSP-Pengutronix-AllYes-trunk]$ ./p compile sw-ek-lm3s3748.compile ptxdist: error: 'sw-ek-lm3s3748.compile' is not a valid package name PTXdist: fatal error ... cannot start, sorry! [x] some_autotool_fan: shebang-line of awk-scripts should use the $AWK value obtained from configure. 20090610 wsa: we hardcoded gawk where needed. configure checks now for gawk explicitly. [x] jbe: udev seems to load kernel modules by its own. How to disable it? 20090610 lfu: the 80-drivers rule causes such behaviour. We don't install it to target, since it's mostly useless for embedded targets. [x] bbu, lfu, rsc, sha: "collections" Observations / Ideas: * "modularity" doesn't fit our needs; we don't only want to build the base packages plus all modules; it is more something like: "build minimum set", or "build minimu set + feature x" * "collections" are lists of packages Use Cases: * "ptxdist --collectionconfig=base go" builds the package set "base" * "ptxdist --collectionconfig=headless_rt go" builds the package set "headless_rt" * "ptxdist --collectionconfig=all go" builds all selected packages * "ptxdist collection " creates "selected_collection" and makes it the default for subsequent commands Necessary: * separate "targetinstall" stage from "ipkg -> root/image" * How to call this? * API? "ptxdist --pkgset= foobar" installs root/ and image.blub from ipkgs [x] jbe: Would it be possible to omit the "ptxconfig" and "platformconfig" menu entries when run "ptxdist menu" when the "selected_ptxconfig" and "selected_platformconfig" are real files than links (or not exist yet)? If they are real files, the project is a single platform one. There is no need for this step! Update 2009610 mkl: done [x] rsc: fix 'ptxdist newpacket host'; find out what mkl intended with this "class" mechanism and make sure it produces sane output. Check with mkl. Example: ptxdist newpacket host-existing-target: - asks for version number, which should be derived from existing target version - asks for URL, but we want to use the download mechanics from target - should output warning that you have to change := into = for corresponding target packet mkl: this is not true! jbe: discuss the templeate rules and their content with mkl to shrink them or to adapt them to the current ptxdist behaviour [x] mkl/jbe: Host's environment variable PKG_CONFIG_PATH leaks in. We want a whitelist instead of the current blacklist. Update 20090610 mkl: done [x] jbe: fuse can be build without libiconv. The glibc iconv() works for it. But its configure is so stupid, it continues to detect a libiconv even if disabled and link against it. (for example if libiconv is already built by accident) Update 20090610 rsc: done [x] jbe: host-glib can be build against the iconv() in the host's glibc. There is no need for a host-libiconv! Update 20090610 rsc: fixed [X] jbe: Running "ptxdist menu" shows always a "u-boot-v2" entry that fails if called even if the platformconfig do not use u-boot-v2 [x] mkl/jbe: The 'ps' command from the procps package outputs "Unknown HZ value! (xx) Assume 100." on a recently started system. Doing it later again, this message does not occur again. Kernel detection failure? Update 20090609: (mkl) fixed [x] jbe: dropbear's init script seems to fail to do a simple "restart". For example if one changes the SSH portnumber in the script and try a restart, nothing changes! Update 20090609 rsc: fixed [x] jbe: the libiconv package fails to build, if a libiconv library is present on the host system (for example in /usr/lib). In this case it tries to link against this library, instead of its own. In one link step: ../srclib/libicrt.a /usr/lib/libiconv.so versus (if no libiconv is present on the host) ../srclib/libicrt.a ../lib/.libs/libiconv.so Update 20090609 mkl: fixed [x] bbu: default path to generic projects seems not to work Update 20090603 rsc: please provide test case, seems to work here. [-] jbe: if ipkg building crashes, PTXdist doesn't break. rsc: see pipeline error fixing in scripts/libptxdist.sh Update 20090603 rsc: we need a testcase, bug closed. [x] bbu: Discuss the whole definition of "package". At the moment, most package options are scattered across two or more *.in or *.make files, making "drop-in replacements" nearly impossible. Update 20090603 rsc: obsoleted by package model in ptxdist 2.0. [x] rsc: find out why iproute2 needs berkeley-db Update 20090603 rsc: fixed with recent iproute2 patches from jbe [x] rsc: iproute2 -> arpd needs db-44, but the select DB_44 does not work. Update 20090603 rsc: seems to be fixed in trunk, close. [-] jbe: classpath.compile fails [...] ../external/sax/org/xml/sax/helpers/NewInstance.java:70: warning: non-varargs call of varargs method with inexact argument type for last parameter; cast to java.lang.Object for a varargs call cast to java.lang.Object[] for a non-varargs call and to suppress this warning return (ClassLoader) m.invoke(Thread.currentThread(), null); ^ Note: * uses or overrides a deprecated API. Note: Recompile with -Xlint:deprecation for details. Note: Some input files use unchecked or unsafe operations. Note: Recompile with -Xlint:unchecked for details. 8 errors 2 warnings Update 20090603 rsc: not reproducable, tested on i586 and arm1136. Please add proper info how to reproduce in case this problem does still persist. [x] jbe: phyton24 states this but does not stop the build (in the install stage): cannot find -lncursesw collect2: ld returned 1 exit status cannot find -lpanelw collect2: ld returned 1 exit status [x] rsc: dbus-python doesn't compile, needs either python-config or asks python for -I and -L paths. While being there, we should update to python-2.6, because this is the stable series we want to have in ptxdist-2.0. update 20090512: should work now [x] jbe: INITMETHOD_BBINIT_ETC_INITD_LOGROTATE makes no sence with busybox's syslog. Doesn't work, because it does not create any *.log files, which logrotate tries to handle. 20090420: update mkl: handles now messages, *.log, *.err [x] mkl: remove inetd.conf leftovers [frogger@hardanger:rules]$ wcgrep inetd.conf|grep -v "inetd.conf.d"|grep -v postinst ./fam.in:52: the /etc/inetd.conf a line like this: ./dnsmasq.in:47: /etc/inetd.conf a line like this: ./dnsmasq.in:71: This string is added to inetd's configuration /etc/inetd.conf. ./portmap.make:132: # /etc/inetd.conf sniplet ./cvs.in:124: This string is added to inetd's configuration /etc/inetd.conf. @ROOT@ ./rsync.make:108: # /etc/inetd.conf sniplet ./rsync.in:69: the /etc/inetd.conf a line like this: ./rsync.in:95: This string is added to inetd's configuration /etc/inetd.conf. @CONFIG@ Update 20090603 rsc: done [x] mkl: fix startscript leftovers update: 20090512: mkl: done [x] mkl: 16:46 < ringlej> /opt/tools/lib/ptxdist-1.99.11/config/collection/Kconfig:10: can't open file "generated/ptx_collection.in" 16:47 < ringlej> That was it. I had none marked as "M" Update 20090603 rsc: fixed [x] jbe: if one starts ptxdist with the --ptxconfig=<...> option and the given path and filename is at least in the form "selected/ptxconfig" (was a typo here, should be "selected_ptxconfig") ptxdist states: /opt/lib/ptxdist-trunk/scripts/libptxdist.sh: line 415: cd: selected: No such file or directory Should be a more reasonable error message. Update 20090603 rsc: fixed [x] jbe: slang.compile fails while compiling png-module.c [...] pcre-module.c:6:18: error: pcre.h: No such file or directory [...] update 20090512: mkl: should work now [x] mkl: fix startscript leftovers update: 20090512: mkl: done [x] jbe: INITMETHOD_BBINIT_ETC_INITD_LOGROTATE makes no sence with busybox's syslog. Doesn't work, because it does not create any *.log files, which logrotate tries to handle. 20090420: update mkl: handles now messages, *.log, *.err [x] jbe: slang.compile fails while compiling png-module.c [...] pcre-module.c:6:18: error: pcre.h: No such file or directory [...] update 20090512: mkl: should work now [-] jbe: libmad tries to re-generate the configure script and fails if system's version is not the correct one [...] missing --run automake-1.10 --foreign aclocal.m4:16: warning: this file was generated for autoconf 2.61. You have another version of autoconf. It may work, but is not guaranteed to. If you have problems, you may need to regenerate the build system entirely. To do so, use the procedure documented by the package, typically `autoreconf'. configure.ac:29: version mismatch. This is Automake 1.10, configure.ac:29: but the definition used by this AM_INIT_AUTOMAKE configure.ac:29: comes from Automake 1.10.1. You should recreate configure.ac:29: aclocal.m4 with aclocal and run automake again. Update 20090420 rsc: not reproducable, neither @mkl nor @rsc [x] Dioptik: freetype 2.3.8 web link is dead From the freetype website "Don't use this version! This is a bugfix release for the 2.3 series. All users should upgrade. See this page for the release notes." Update: 20090417 mkl: done [x] jbe: tcpdump fails when IPv6 is enabled: checking getaddrinfo bug... buggy Fatal: You must get working getaddrinfo() function. or you can specify "--disable-ipv6". Update: 20090417 mkl: done [x] jbe: SDL menu: Unsure if SDL mixer and SDL image must depend on SDL? mkl: no they must select SDL [x] rsc: if 'ptxdist setup' points to the wrong JDK, jvisu claims that it needs a set JAVA_HOME. A better error message would involve 'ptxdist setup'. mkl: fixed by mol [x] jbe: Running 'ptxdist go' with an enabled but not yet configured kernel should give a reasonable error message. It should not just fail while downloading a nonexisting kernel. (happens when ptxconfig components forcing a kernel not yet configured in the platformconfig) Update 20090316 rsc: error message added [x] rsc: check jvisu, it has a FIXME that ant may be needed as a host tool. Remove, use the same mechanism as with Perl::XML Update 20090313: fixed by mol and mkl [x] jbe: It seems now be possible to install a /etc/rc.d/*telnetd link to /etc/init.d/telned without intalling a /etc/init.d/telnetd [x] Sisyphus: [...] ---------------------- target: glibmm.compile ---------------------- [...] make[6]: Entering directory `[...]/glibmm-2.14.1/glib/glibmm' [...] convert.cc: In constructor 'Glib::IConv::IConv(const std::string&, const std::string&)': convert.cc:52: error: 'g_assert' was not declared in this scope make[6]: *** [convert.lo] Error 1 make[6]: *** Waiting for unfinished jobs.... date.cc: In member function 'Glib::ustring Glib::Date::format_string(const Glib::ustring&) const': date.cc:311: error: 'g_assert' was not declared in this scope make[6]: *** [date.lo] Error 1 rsc: glib has recently been updated to a newer 2.12 version; find out which glibmm version fits the new one and adapt it. Update 20090303 rsc: updated all gtk stack components to latest stable. [x] jbe: if one enters "ptxdist platformconfig" one can earn this message: error: The config file 'selected_platformconfig' is missing the symbol: 'PTXCONF_PLATFORMCONFIG_VERSION', which means it is broken. But also the "--force" option does not help. Only "--sledgehammer" helps to create this symbol. Intended behaviour? mkl: 20090302: "--force" works now as expected [x] rsc: add 'quiet' support to start scripts (probably centralized in rcS) [x] mkl: reactiveate new install_copy feature (no explicid destination) rsc: ask mkl what kind of problems have been observed with the old implementation that has been thrown out again update: 2009-01-25: mkl: rsc has upgraded the install_copy macro to handle this [x] jbe: e2fsprogs: HTREE can be disabled, but at least mkfs.ext2 creates (silently) a HTREE based filesystem! e2fsck will fail on such a filesystem unless HTREE support will be activated in the menu! mkl: 2009-2-27: htree enable i.e. hardcode htree supprt btw: disabling affects only the fsck tool, not mk2efs [x] jbe: avoid the "make: Nothing to be done for `world'." message when running a "ptxdist images". [x] jbe: xorg-lib-fontenc.compile: What does it mean? Intervention needed? [...]/arm-v5te-linux-gnueabi/bin/ld: warning: creating a DT_TEXTREL in object. [x] jbe: same in cairo.compile [x] jbe: xorg-lib-fontenc.compile (but this time seen on PowerPC) mkl: 2009-02-27: this means that we link a static lib into a shared one, and this should be avoided. [x] jbe: it seems to extract a kernel always touches the local kernelconfig.target (or similar). This confuses svn, as this file always changes without calling "ptxdist kernelconfig" rsc: mol has a patch for this, integrate mkl: 20090626 no problem with with ptxdist-1.99.11, probably in the whole 1.99.x series of ptxdist [x] rsc: network handling with qemu does not work reliably; it works when being run with sudo (see start/stop/connect scripts in i586) and /dev/net/tun must be made rwrwrw. [x] rsc: in x86 simulation: ipkg list doesn't work? mkl: 20090226: works root@phyCORE:~ ipkg list alsa-lib - 1.0.18 - alsa-utils - 1.0.18 - atk - 1.20.0 - atop - 1.23 - [...] [x] jbe: avoid the convenience stage question, when running "pxtdist images". jbe: avoid the convenience stage question again and again and again.... [x] jbe: host-fontconfig.compile if gcc [...] -c -o fc-case.o fc-case.c; \ then mv -f ".deps/fc-case.Tpo" ".deps/fc-case.Po"; else rm -f ".deps/fc-case.Tpo"; exit 1; fi In file included from fc-case.c:25: ../src/fcint.h:118:21: error: fcalias.h: No such file or directory make[3]: *** [fc-case.o] Error 1 mkl: 20090226 works on himalia [x] jbe: host-glade.prepare [x] Bo Hansen: util-linux fails to build more.c when no curses library is present. Different development platforms are behaving differently. jbe's host detects the lack of curses and omits more.c to build in this case. So the build does not fail. Bo's host decide to build more.c even if curses is not present and fails. [x] jbe: if one enable the "Local kernel tree" by accident but hasn't been setup the "Prefix for kernel trees", ptxdist should output a more useful error message than just: ---------------------- target: kernel.extract ---------------------- Error: empty parameter to 'extract()' mkl: 20090226: done: [frogger@hardanger:OSELAS.BSP-phyCORE-trunk]$ ./p go /home/frogger/pengutronix/ptxdist/ptxdist-trunk/rules/kernel.make:43: *** /home/frogger/pengutronix/ptxdist/ptxdist-trunk/rules/kernel.make:44: *** PTXCONF_KERNEL_LOCAL_FLAG feature activated, but /home/frogger/pengutronix/ptxdist/ptxdist-trunk/rules/kernel.make:45: *** PTXCONF_SETUP_KERNELDIR_PREFIX is unset! /home/frogger/pengutronix/ptxdist/ptxdist-trunk/rules/kernel.make:46: *** /home/frogger/pengutronix/ptxdist/ptxdist-trunk/rules/kernel.make:47: *** This feature is mainly for developers, who to want have their kernel sources /home/frogger/pengutronix/ptxdist/ptxdist-trunk/rules/kernel.make:48: *** outside of ptxdist. You can turn it off by deselecting "Local kernel tree": /home/frogger/pengutronix/ptxdist/ptxdist-trunk/rules/kernel.make:49: *** "ptxdist platformconfig" -> "Linux kernel" -> "Local kernel tree" /home/frogger/pengutronix/ptxdist/ptxdist-trunk/rules/kernel.make:50: *** /home/frogger/pengutronix/ptxdist/ptxdist-trunk/rules/kernel.make:51: *** If you want to use the feature, please enter a propper prefix /home/frogger/pengutronix/ptxdist/ptxdist-trunk/rules/kernel.make:52: *** to your kernel tree /home/frogger/pengutronix/ptxdist/ptxdist-trunk/rules/kernel.make:53: *** "ptxdist setup" -> "Source Directories" /home/frogger/pengutronix/ptxdist/ptxdist-trunk/rules/kernel.make:54: *** -> "Prefix for kernel trees" /home/frogger/pengutronix/ptxdist/ptxdist-trunk/rules/kernel.make:55: *** and specify where to look for your kernel tree /home/frogger/pengutronix/ptxdist/ptxdist-trunk/rules/kernel.make:56: *** /home/frogger/pengutronix/ptxdist/ptxdist-trunk/rules/kernel.make:57: *** . Stop. [x] jbe: host-intltool.prepare checking for XML::Parser... configure: error: XML::Parser perl module is required for intltool Update 20090118 rsc: add a function to 'ptxdist go' that can run stuff depending on ptxconfig; I don't want to add a ptxdist dependency to XML::Parser and perl, just for glade. Something that runs on 'ptxdist go' or 'ptxdist ' (FIXME which one?) and can output something like "error: need perl and XML::Parser to build glade". Update 2009-01-20: jbe: add a test for perl XML::Parser into these makefiles and fail gracefully with a nice error message. So this test runs on demand only. [x] jbe: "ptxdist platformconfig" moans: error: The configfile version and ptxdist version do not match: configfile version: ptxdist version: 1.99.svn Users current focus is on "selected_platformconfig". But ptxdist moans about an empty 'selected_ptxconfig' here! Instead it should: - output the correct filename which is causing the problem here - silently does its job, because the platformconfig is all right here - should output this message, when running a "ptxdist menuconfig" [x] Whenever a new packet dependency is added to the menus and one does not update the used ptxconfig file, the next build will only detect the ".install" dependency and fail due to the fact the ".[get|extract|prepare|compile]" are still missing. When ptxdist is able to detect the dependency to ".install", it should also add the other targets (get|extract|prepare|compile) from inside the silently. The current error is confusing, because it occures even the user didn't change anything in his/her project. [x] jbe: Would it be possible to delete a directory defined by PTXCONF_SYSROOT_TARGET in the form "${PTXDIST_WORKSPACE}/local/${PTXCONF_ARCH_STRING}" Currently the "local" part still exists when running "ptxdist clean". This is for ptxdist-1 backward compatibility. Update 20090126: the correct is path is: "${PTXDIST_WORKSPACE}/local/${PTXCONF_GNU_TARGET}" and it is cleaned now [x] jbe: host-dbus-glib.compile ../../dbus/dbus-binding-tool: error while loading shared libraries: libdbus-glib-1.so.2: cannot open shared object file: No such file or directory [x] jbe: Adding "ant" to the required host tools. Emit a warning if its not present, but not fail (required only to build jvisu) Update 2007-01-24: rsc: rejected. Ant is only needed by tools which use Java, and we don't want a prerequisite to java in PTXdist. Maybe add some loud warning if packets are built which use java, such as jvisu. Update 2009-01-20: jbe: add a test for 'ant' into jvisu's makefile and fail gracefully with a nice error message. So this test runs on demand only. Update 2009-01-26: done, no tests in the configure script for now. looks this way now, everybody happy??? Update 2009-02-05: jbe is happy now. make: *** [/home/frogger/pengutronix/bsp/OSELAS.BSP-Pengutronix-AllYes-trunk/platform-phyCORE-MPC5200B-tiny/state/jvisu.compile] Error 127 [frogger@hardanger:OSELAS.BSP-Pengutronix-AllYes-trunk]$ ./p compile jvisu /home/frogger/pengutronix/ptxdist/ptxdist-trunk/rules/jvisu.make:18: *** ant is mandatory to build JVisu /home/frogger/pengutronix/ptxdist/ptxdist-trunk/rules/jvisu.make:19: *** please install ant /home/frogger/pengutronix/ptxdist/ptxdist-trunk/rules/jvisu.make:20: *** . Stop. [x] jbe: /usr/share/apache2/libexec/mod_python.so Misses libpython2.4.so.1.0 at runtime Update: 2009-01-29: mkl broke it mol fixed it [x] jbe: These libs/executeables are including host library search paths (RPATH) Check selected_toolchain/powerpc-603e-linux-gnu-readelf -d for long host paths: [x] mod_python from the Apache2 package [x] inotifywait from the inotify-tools package [x] jbe: These libs/executeables are including host library search paths (RPATH) Check selected_toolchain/powerpc-603e-linux-gnu-readelf -d for long host paths: [x] libcpufreq from the cpufrequtils package [x] mc from the Midnight Commander package [x] jed from the JED package [x] jbe: correct the ouput of "ptxdist menuconfig" when configuration is finished. The output of *** End of configuration. *** Execute 'make' to build the project or try 'make help'. is outdated... rsc: while doing this, update to a recent kconfig version, with saner colours instead of this unreadable yellow. [x] JBE: http://gcc.gnu.org/bugzilla/show_bug.cgi?id=27363 [x] jbe: timezone files installation doesn't work as expected [x] jbe, rsc: add sane minimal in-system documentation: README [x] jbe: review all READMEs, COMPILE-TESTs, SPECIFICATIONs and so on files in the ptxdist install folder. Most of them are trash only [x] jbe: use tmpfs everywhere instead of the ramfs [x] jbe: Current implementation of "screen" is useless due to lack of generic configuration [x] jbe: replacing an *.make file for project use is easy. But how to replace a *.in file? rsc: no 0.10.6 issue, defer [...] same as below: [x] bbu: Discuss the future kconfig include scheme. At the moment, overriding single *.in files is a real pain. - There should be a way to autogenerate the needed files before starting kconfig or kconfig should just ignore double definitions (which might be a bad idea) Features needed: - override *.make as well as *.in - create a sane menu structure [x] rsc: 'ptxdist print ' is pretty slow now; shall work with retval=0 if variable exists and retval!=0 if not [x] RSC: http://htop.sourceforge.net/ now part of ptxdist [x] jbe: Working with local sources seems broken, when a _SOURCE is defined to the regular $(SRCDIR)/. The get stage fails in this case. Our template will always create and fill this variable. Don't know if this is intended and the failure is elsewhere. Update 20090120 jbe: *_SOURCE is now defined to an empty variable. [p] jbe: fltk.compile In file included from ../FL/Fl_Single_Window.H:29, from doublebuffer.cxx:37: ../FL/Fl_Window.H: In member function 'void Fl_Window::size_range(int, int, int, int, int, int, int)': ../FL/Fl_Window.H:91: warning: conversion to 'uchar' from 'short int' may alter its value ../FL/Fl_Window.H:91: warning: conversion to 'uchar' from 'short int' may alter its value ../FL/Fl_Window.H:91: warning: conversion to 'uchar' from 'short int' may alter its value ../fluid/fluid: ../fluid/fluid: cannot execute binary file make[2]: *** [fast_slow.cxx] Error 126 make[2]: *** Waiting for unfinished jobs.... finished: fltk.compile ---> but it does not fail Update 20090120 rsc: nobody is using fltk right now @ ptx, so mark it broken [p] jbe: fltk.install cp: cannot create regular file `/usr/lib/libfltk_jpeg.a': Permission denied make[2]: *** [install] Error 1 cp: cannot create regular file `/usr/lib/libfltk_jpeg.a': Permission denied make[2]: *** [install] Error 1 ---> but it does not fail Update 20090120 rsc: nobody is using fltk right now @ ptx, so mark it broken [x] jbe: grep.compile: Host leakage [...]/arm-v5te-linux-gnueabi/bin/ld: skipping incompatible /usr/lib/libpcre.so when searching for -lpcre [...]/arm-v5te-linux-gnueabi/bin/ld: skipping incompatible /usr/lib/libpcre.a when searching for -lpcre [...]/arm-v5te-linux-gnueabi/bin/ld: cannot find -lpcre collect2: ld returned 1 exit status Update 20090119 rsc: added a well defined switch to the grep packet; we are not sure if this is the whole fix; until somebody shows us something different, we assume it to be fixed. [x] jbe: proftpd.compile i586-unknown-linux-gnu-gcc [...] -I.. -I../include -O2 -Wall -c pr_fnmatch.c In file included from [...]/sysroot-i586-unknown-linux-gnu/usr/include/asm/sigcontext.h:5, from [...]/sysroot-i586-unknown-linux-gnu/usr/include/bits/sigcontext.h:28, from [...]/sysroot-i586-unknown-linux-gnu/usr/include/signal.h:333, from [...]/sysroot-i586-unknown-linux-gnu/usr/include/sys/wait.h:31, from ../include/conf.h:95, from pr_fnmatch.c:38: [...]/sysroot-i586-unknown-linux-gnu/usr/include/asm/types.h:8: error: conflicting types for 'mode_t' [...]/sysroot-i586-unknown-linux-gnu/usr/include/sys/types.h:72: error: previous declaration of 'mode_t' was here Update 20090118 rsc: later toolchain uncover this bug; we upgrade from 1.3.0a to 1.3.2rc3 which fixes the issue. [x] jbe: madwifi.compile: Kernel version dependend on x86: Emits a huge amount of errors about wrong pointers and parameter counts. Update 20090118 rsc: madwifi doesn't build with current kernels any more, so we mark it BROKEN. [x] jbe: valgrind.prepare checking the libc version... unsupported version configure: error: Valgrind requires glibc version 2.2 - 2.7 Update 20090118 rsc: fixed, we use a newer valgrind version now [x] jbe: libx86.compile i586-unknown-linux-gnu-gcc [...] -c -o lrmi.o lrmi.c lrmi.c: In function 'set_regs': lrmi.c:191: error: 'IF_MASK' undeclared (first use in this function) lrmi.c:191: error: (Each undeclared identifier is reported only once lrmi.c:191: error: for each function it appears in.) lrmi.c:191: error: 'IOPL_MASK' undeclared (first use in this function) lrmi.c: In function 'run_vm86': lrmi.c:613: error: 'VIF_MASK' undeclared (first use in this function) lrmi.c:613: error: 'TF_MASK' undeclared (first use in this function) lrmi.c: In function 'LRMI_int': lrmi.c:840: error: 'IF_MASK' undeclared (first use in this function) lrmi.c:840: error: 'IOPL_MASK' undeclared (first use in this function) make[2]: *** [lrmi.o] Error 1 Update 20090118 rsc: fixed [x] jbe: Busybox: (on PowerPC and X86 only) [...] cc1: warnings being treated as errors In file included from util-linux/fdisk.c:622: util-linux/fdisk_osf.c: In function 'xbsd_writelabel': util-linux/fdisk_osf.c:961: error: unused parameter 'p' [x] jbe: chrony.compile powerpc-603e-linux-gnu-gcc -O2 -g -DFEAT_RTC=1 [...] -c rtc_linux.c In file included from rtc_linux.c:61: io_linux.h:39:2: error: #error "I don't know the values of the _IOC_* constants for your architecture" rtc_linux.c: In function 'switch_interrupts': rtc_linux.c:680: error: 'CHRONY_IOC_NONE' undeclared (first use in this function) rtc_linux.c:680: error: (Each undeclared identifier is reported only once rtc_linux.c:680: error: for each function it appears in.) rtc_linux.c:680: error: 'CHRONY_IOC_NRBITS' undeclared (first use in this function) rtc_linux.c:680: error: 'CHRONY_IOC_TYPEBITS' undeclared (first use in this function) rtc_linux.c:680: error: 'CHRONY_IOC_SIZEBITS' undeclared (first use in this function) [...] Update 20090118 rsc: fixed [-] jbe: busybox.compile (on PowerPC) [...] CC util-linux/volume_id/ext.o cc1: warnings being treated as errors In file included from util-linux/fdisk.c:622: util-linux/fdisk_osf.c: In function 'xbsd_writelabel': util-linux/fdisk_osf.c:961: warning: unused parameter 'p' make[2]: *** [util-linux/fdisk.o] Error 1 make[1]: *** [util-linux] Error 2 Update 20090118 rsc: this is a duplicate, also reported above. [x] jbe: gwt.extract (at least a PowerPC) This stage seems to run the autotools. Update 20090118 rsc: it runs autotools because it is instructed to do so without any deeper sense, in the .make file. Fixed. [x] jbe: Segmentation fault? ----------------------- target: locales.install ----------------------- generating "en_US" - this can take some time /work/ptxdist-trunk/scripts/make_locale.sh: line 29: 14673 Segmentation fault \ I18NPATH=${SYSROOT_USR}/share/i18n ${LOCALEDEF} ${ENDIAN} -i $LOCALE_DEF \ -f ${CHARMAP} --prefix=${PREF} $LOCALE_NAME calling localedef binary failed [x] jbe: directfb.compile powerpc-603e-linux-gnu-gcc -DHAVE_CONFIG_H [...] -c ppcasm_memcpy_cachable.S -fPIC -DPIC -o .libs/ppcasm_memcpy_cachable.o ppcasm_memcpy_cachable.S:37:26: error: linux/config.h: No such file or directory [x] jbe: libtheora.prepare "[x] sdl support" results into a: [...] checking for sdl... yes checking for sdl >= fix.me.now... configure: error: *** sdl >= fix.me.now not found by pkg-config on your system [x] jbe: libcgicc: Runs host's autotools when the compile stage starts on some systems (*not* on himalia). On system where it does not run the autotools the compile stage fails with: make[3]: Entering directory `[....]/platform-i586/build-target/cgicc-3.2.7/doc' if test ! -d html; then \ mkdir html; \ fi; DATE=`date '+%-d %b %Y'` VERSION=3.2.7 /bin/echo Doxyfile Doxyfile cp cgicc.png gnu-head-sm.jpg cgicc-doc.css html mv html/index.html html/index.html.bak mv: cannot stat `html/index.html': No such file or directory [x] migrate libc from ptxconfig to platform [x] rsc: config dependency dropper doesn't work yet; it assumes that all labels correspond to their package names, but that's not true for busybox, so the dropping doesn't work for bb. The right solution would be to change PTXCONF_BB_ to PTXCONF_BUSYBOX__, but that would break all migrations of old projects. Can only be done with a proper migration script. [x] Subject: Re: [ptxdist-commit] ptxdist r9228 - in trunks/ptxdist-trunk/patches: libffi-3.0.7 libffi-3.0.7/generic I am not sure this is the correct fix. The problem seems that ptxdist uses ac_cv_sizeof_long_double with a value of 8, and that is where the problem is, because that should be 12 on x86 and 16 on x86_64. That error results in double and long double to have the same size and configure to conclude there is no long double. And since that is never true on x86 with gcc, there was no need for the ifdef with HAVE_LONGDOUBLE in the x86 case. - Erwin rsc: add proper definitions according to platform definition (PTXCONF_SIZEOF_LONG_DOUBLE) [x] rsc: fix install_copy without source+destination; it worked before with destination-only, but there was some problem. Ask mkl. Update 20081228: rsc: seems like it works in the current trunk. [x] rsc: autogenerated menues with overwriting. 20081109 mkl: [x] directfb [x] xorg [x] gtk [x] bluez [x] crosstools [x] debug [x] gdb [x] gtk [x] host-gtk [x] hosttools [x] libc [x] ltp [x] net-sec [x] networking [x] ntp [x] sdl [x] tests [x] wireless [x] xorg still have to be migrated [x] jbe: Getting the value of an internal ptxdist variable returns always 1 as its exit code ("ptxdist print ROOTDIR" returns with the correct value on stdout but with exit code = 1) Update 20081227: rsc: fixed, but now 'ptxdist print' is pretty slow. [x] when the drop command is used without a stage, complain about that and suggest to use clean instead [x] jbe: When someone do not need a Linux kernel one earns: rules/kernel.make:109: warning: ignoring old commands for target `' [x] rsc: fallback server for udev [x] rsc: toolchain selection doesn't work if gcc revision is empty=nocheck [x] rsc: platformconfig: check what about PTXCONF_IMAGE_IPKG_ARCH; it should go to the platformconfig code. arm -> oabi armel -> eabi little endian armeb -> eabi big endian This would make our packages compatible with maemo + debian [x] 20081107 mkl: we use not PTXDIST_IPKG_ARCH_STRING which is generated in scripts/lib/ptxd_make_install.sh [x] rsc: we don't seem to have a target option for the endianess! [x] rsc: SuSE has an old stat version which doesn't support --printf; this breaks building of hd.img. Add test to configure. [x] 20081107 mkl: we use "stat -c" now [x] jbe: When someone do not need u-boot-v2 one earns: /rules/u-boot-v2.make:80: warning: overriding commands for target `' [x] jbe: Fix the too many --without-pam and the always --without-sendmail given in the sudo.make in the prepare step rsc: fixed by cleaning up package and disabling all untested things [x] jbe: if the kernel.extract stage fails, ptxdist returns with exit code 0 Can be replicated by deleting the configures series file for the patch stack in use. [x] BBU: Implement scheme for alternative download-URLs as fallback (e.g. mirrors) [x] add menu entry for user local config files in /etc [x] mol: ptxdist commands behave differently for arguments which exceed the "current" command; some continue parsing and some exit. [-] Start a "Global Patch Repository Unification Project". Potential members: * PTXdist * T2 * OpenEmbedded http://www.openembedded.org/repo/org.openembedded.dev/packages/ * Gentoo * Debian [x] bbu: audit *.in and *.make files, broken install definitions everywhere [tm] (mostly c&p errors, probably) [x] repair make makro 'install'; there are sometimes three """. See procps.install for reference. [x] jbe: reduce directory count on top level, combine build-* and local-* into a shared subdir rsc: we've agreed on this variant: build-host/ build-target/ build-cross/ local/ local-target/ local-cross/ [x] jbe: cleanup rules/other/Toplevel.make [x] jbe: remove ptxdist's documentation from the installation. It's now useless and outdated as a better alternative is present with the application notes [x] jbe: hd images must be built from ipkgs to support the ipkg tool on the target [x] if a packet is not selected, it's dependencies are not generated. If you kick a stage then, the dependend stages are not executed. Solution: generate the stages as well, so we get an error in that case. [x] RSC:is "select GLIBC_DEBUG" in gdb.in necessary?? [x] jbe: Switching from generating hd image to jffs2 image _rebuilds_ the the whole project! Known Problem: the virtual-cross-tools depends on host-mtd-utils, which is selected by JFFS2. this must depend on images somewhere. mkl knows how to fix it... [x] jbe: current trunk complains glib.make:40: warning: overriding commands for target `/home/all_src/glib-2.8.6.tar.bz2' host-glib.make:38: warning: ignoring old commands for target `/home/all_src/glib-2.8.6.tar.bz2' [x] jbe: if the "ptxdist toolchain" step was omited, ptxdist complains about toolchain vendor instead of missing toolchain [x] jbe: if ptxdist was called with wrong parameters (and complains about it) it should return also an error return code (so something like "ptxdist clen kernel && ptxdist go" fails) [x] jbe: ipkg does not work on the target anymore. Cannot install anything. [x] jbe: suppress cross toolchain and toolchain vendor check while --native build [x] rsc: put export LANG=C somewhere into PTXdist, to make compiler output more readable [x] jbe: libxcb depends on host's xslt installation! Current fix is to be checked. [x] jbe: xcb-proto depends on host's xmllint! [x] jbe: ptxdist: platformconfig... platforms/images.in:14:warning: 'select' used by config symbol 'IMAGE_HD' refers to undefined symbol 'HOST_GENPART' platforms/images.in:193:warning: 'select' used by config symbol 'IMAGE_JFFS2' refers to undefined symbol 'HOST_MTD_UTILS' platforms/images.in:226:warning: 'select' used by config symbol 'IMAGE_UIMAGE' refers to undefined symbol 'HOST_UMKIMAGE' platforms/images.in:271:warning: 'select' used by config symbol 'IMAGE_EXT2' refers to undefined symbol 'HOST_GENEXT2FS' [x] jbe: add help what symbol "PTXCONF_PLATFORM" should contain (or what it means if this entry is still empty) [x] jbe: "ptxdist help" does not output the new option "platformconfig" [x] jbe: if the ptx dialog will be called from within the menu dialog the loop should stop, if calling the ptx dialog resturns with an error. In one case one of the *.in files was missed and the error message can't be read because the menu dialog overwrite it immediately [x] jbe: udev. Generic startscript complains about a missing udev-adm [x] jbe: fakeroot seems to have problems to run in a 64 bit environment. Not at the image creation stage, only at the "install" stage Update: - fakeroot fails on all glibc 2.5 based systems [x] jbe: rules file for zlib is broken since ptxdist-0.10.6! What the hell means libz.so.0.0.0???? In ptxdist-0.10.5 it was libz.so.1.2.3! [x] jbe: transfer ipkg's runtime configuration to the ipkg packet itself -> its not rootfs [x] jbe: uboot-1.2.0's build system is not multi processor save -> supress -j option [x] pureftp's configure runs twice and needs host's autotools! [x] audit the logfile of AllYesArm for autotool calls Update 2007-01-24: rsc: the solution is simple: just add AM_MAINTAINER_MODE to the configure.in/ac if the autotools are re-run. This avoids that the maintainer dependencies are put into the packet. [x] util-linux: re-calls automake during compile; [x] libxslt: re-calls automake during compile [x] pure-ftpd: re-calls automake during compile [x] jbe: all local sources (file://) are still do a touch into the package directory. (into "Source Directory"). If this file gets deleted, ptxdist builds the whole package again (and recreate the file in the Source Directory). Update 2007-01-24: fixed; order of PACKET-empty check in extract was changed. [x] jbe: Chrony fails to compile Update 2007-01-25: jbe: better fix; force chrony to use the wide ncurses lib if present, else the non wide one Update 2007-01-19: rsc: fixed; ncurses didn't install the non-widechar variants of the libraries. [x] OSELAS.BSP-Pengutronix-TechPool-trunk$ p select ptxconfig.dotgnu /home/rsc/svn/ptxdist-trunk/scripts/libptxdist.sh: line 26: /media/rscusb1_plain/svn/oselas/bsp/pengutronix/huettinger/OSELAS.BSP-Pengutronix-TechPool-trunk/ptxconfig: No such file or directory Update 2007-01-19: rsc: fixed [x] jbe: starting with a new and empty project should provide the user with a generic (and default) ptxconfig instead of an error message. But it should not do it automatically! It should ask before doing it! Otherwise we will end up with many ptxconfig files in many directories... Update 2007-01-19: rsc: it's not so easy as "touch ptxconfig && ptxdist oldconfig". Error message is improved the preferred way to start a new project is cloning. [x] jbe: host-mknbi fails to build Update 2007-01-19: rsc: doesn't work, hasn't for ages, deleted [x] jbe: liboldX fails to build the ipkg but the build continues! Update 2007-01-19: rsc: fixed [x] unbreak host-zlib; it is necessary for host-umkimage if there is no zlib devel packet installed on the host. [x] omitting glibc installation should be possible, first hack attempt done Update 2007-01-12: No idea what the reason was; we don't have a concept for anything else than shared libs and glibc. [x] rsc: fakeroot-the-hosttool seems not to be found - reported by wrap. It looks to me as if $(PTXDIST_WORKSPACE)/local/bin has to be included into the path somehow. Ask mkl. [x] Our current host tools use their target counterparts for getting the sources; if the target counterpart is not selected, the dependencies are not created correctly! Test: switch off ipkg but activate host-ipkg. [x] audit 'get' stage [x] Menu Structure JBE: Use menuconfig instead of menu + configs whereever possible. It looks nicer in "menuconfig" and its much better supported when using "xconfig" [x] ptxdist should reject working as ROOT [x] add prerequisite checks to configure script; find out which prerequisites we really need. [x] Toolchain selection JBE: add an identifier to a (central) toolchain (in "make setup"). It should be possible to use more than one toolchain in a local ptxdist folder (for different targets/projects). So the idetifier selects the right toolchain for this project. This identifier simplifies the toolchain selection. There could be a central folder in the system with all informations (and identifiers) about any available toolchain installed on this system. --> done: "ptxdist toolchain path" per project [x] discuss UID of user "nobody" and GID of "nogroup" [x] jbe: ptxdist images fails for jffs2 fileimage creation it seems local-host/sbin is not in the PATH [x] jbe: procps.install stage fails with: make: Ungültige Option -- y make: Ungültige Option -- y make: Ungültige Option -- L make: Ungültige Option -- / make: Ungültige Option -- L make: Ungültige Option -- / Seems the install macro emits: make [...] CFLAGS=""" LDFLAGS=""" [...] [x] jbe: udev needs rules to control permissions on device nodes; jbe currently writes docs for this. [x] jbe/rsc: /etc/resolv.conf and /etc/nsswitch.conf: Find a solution to generate DNS configuration files from other sources like boardsetup (I have no clue what this means). "Then either have no resolv.conf at all (use /etc/hosts only), or generate resolv.conf from 'ptxdist boardsetup' information and generate a sane nsswitch.conf." [x] rsc: check if we can update mtd-utils from 1.0.0 to 1.0.1; mkl has already prepared the patches; find out why the revision number was not bumped. mkl: 1.0.1 needs acl, so we stay with 1.0.0 [x] ssh: server start scripts are missing [x] ssh: key gen mechanics needed [x] dbus needs user messagebus -> /etc/passwd; jbe: needs other user, non-root; add this user to /etc/passwd. Debian has a user called "messagebus", we should do it the same way. Define a sane UID (ask jbe) [x] rsc: find out why mkl added this new dependency mechanism which searches for ${${LABEL}_DIR}/.ptx-extract. [x] rsc: merge mkl's host/cross/target-pkg-config fixes. We need this to have a proper x.org and host-dbus. [x] jbe: make DHCP client useful on the target. Currently only the binary will be built and copied, but does not work on the target [x] jbe: update DHCP tp 3.0.5 or correct path to 3.0.4 [x] jbe: creating links into non existent directories fails, but does not break the build! Update 2007-01-12: rsc/mkl: the install_link macro has mkdir -p calls in all relevant places; should not happen -> jbe: retest. TODO.rsc -------- [ ] connman: [ ] add configure switch for glib detection [ ] if connecting dbus doesn't work, we get no output; seems like daemon() doesn't connect to syslog first