| Age | Commit message (Collapse) | Author | Files |
|---|
|
Previously only the top-level index was created, which did not
work if PACKAGE_FEED_ARCHS whitelisting (or explicitly listing
architectures in dnf repo files by hand) was in use:
https://lists.yoctoproject.org/pipermail/yocto/2018-March/040327.html
https://bugzilla.yoctoproject.org/show_bug.cgi?id=12419
[YOCTO #12419]
Signed-off-by: Alexander Kanavin <alexander.kanavin@linux.intel.com>
Signed-off-by: Ross Burton <ross.burton@intel.com>
|
|
* --no-signature saves unnecessary .patch modifications when executed on
host with different git version
* --no-numbered saves unnecessary .patch modifications when number of the
applied patches is changed (the number is still in the filename so the
order how they should be applied is still preserved)
* both options exist for very long time, I've tested them with git 1.9.1
from Ubuntu 14.04 and I'm quite sure they were available even in much
older releases, so there shouldn't be any issue on relatively new sanity
tested distros
Signed-off-by: Martin Jansa <Martin.Jansa@gmail.com>
Signed-off-by: Ross Burton <ross.burton@intel.com>
|
|
Signed-off-by: Ross Burton <ross.burton@intel.com>
|
|
Ignoring patch context increases the chances of patches being
applied incorrectly. Depending on what code is being patched, this can go
completely unnoticed and create subtle bugs, sometimes with security implications.
Please see here for a specific example:
https://bugzilla.yoctoproject.org/show_bug.cgi?id=10450
On the other hand, we cannot simply force all patch context to match exactly:
doing this would break a lot of recipes suddenly, across all layers.
So let's try a softer approach: issue a warning, and gently update
patches over a longer span of time. When most of the warnings are eliminated,
we can start enforcing a strict patch application policy.
I do understand that this patch creates a lot of warnings all of a sudden, however
I believe the problem does need to be addressed. All of oe-core recipes have their
context already fixed.
Sample warning:
WARNING: vulkan-1.0.61.1-r0 do_patch:
Some of the context lines in patches were ignored. This can lead to incorrectly applied patches.
The context lines in the patches can be updated with devtool:
devtool modify <recipe>
devtool finish --force-patch-refresh <recipe> <layer_path>
Then the updated patches and the source tree (in devtool's workspace)
should be reviewed to make sure the patches apply in the correct place
and don't introduce duplicate lines (which can, and does happen
when some of the context is ignored).
Details:
Applying patch demos-Don-t-build-tri-or-cube.patch
patching file demos/CMakeLists.txt
Hunk #1 succeeded at 63 (offset 2 lines).
Hunk #2 succeeded at 76 with fuzz 1 (offset 2 lines).
[YOCTO #10450]
Signed-off-by: Alexander Kanavin <alexander.kanavin@linux.intel.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
package types
Previously this was done only for rpm packages; now also ipk/deb scriptlet
failures are reported.
In the future this will become a hard error, but it can't yet happen
due to the legacy 'exit 1' way of deferring scriptlet execution to first boot which
needs a deprecation period.
Signed-off-by: Alexander Kanavin <alexander.kanavin@linux.intel.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
matching
Particularly, this was causing 'devtool modify' to erroneously add those
.orig files into commits. This was getting in the way, if the goal
was to amend/update those existing patches.
Signed-off-by: Alexander Kanavin <alexander.kanavin@linux.intel.com>
Signed-off-by: Ross Burton <ross.burton@intel.com>
|
|
When running bitbake -c populate_sdk <image_name>, it is expected that
packages matching SDKIMAGE_INSTALL_COMPLEMENTARY name mask (unless
declared in PACKAGE_EXCLUDE_COMPLEMENTARY) are installed to resulting
SDK. Underlying mechanism issues a package manager install call for set
of complementary packages. However the mechanism doesn't seem to inform
the user all too obviously in case the package manager command behind
install_complementary() method fails -- and since it is combined with
attempt_only=True option, user might end up wondering why several *-dev,
*-dbg packages are missing from resulting SDK.
Improve associated install() method behaviour in affected OpkgPM and
DpkgPM classes so that a problematic state of affairs becomes directly
obvious for bitbake user, resulting in shell output like:
WARNING: someimage-1.0-r0 do_populate_sdk: Unable to install packages.
Command '...' returned 1:
Collected errors:
* Solver encountered 1 problem(s):
* Problem 1/1:
* - package somepkg-dev-1.0-r0.x86 requires somepkg = 1.0-r0, but
none of the providers can be installed
*
* Solution 1:
* - allow deinstallation of someotherpkg-1.1-r1.x86
* - do not ask to install a package providing somepkg-dev
* Solution 2:
* - do not ask to install a package providing somepkg-dev
Signed-off-by: Niko Mauno <niko.mauno@vaisala.com>
Signed-off-by: Ross Burton <ross.burton@intel.com>
|
|
In cases when a package like qemu might have files with same names
in multiple directories, the rename logic might go wrong and create
multiple rename pair for a single directory.
Make sure that we process each rename pair once. Also, don't print
FILELIST as part of PKGSIZE to ensure that it gets printed only once
when reporting package changes.
Fixes [YOCTO #12559]
Signed-off-by: Anuj Mittal <anuj.mittal@intel.com>
Signed-off-by: Ross Burton <ross.burton@intel.com>
|
|
If you have a package that does not generate a manifest due to using a
noexec rule, the package name should be printed so the problem can be
tracked down. With out the patch you get an error that makes it look
more like the package_manager is broken as shown below.
oe-core/meta/lib/oe/package_manager.py', lineno: 534, function: create_packages_dir
0530:
0531: for dep in rpmdeps:
0532: c = taskdepdata[dep][0]
0533: manifest, d2 = oe.sstatesig.find_sstate_manifest(c, taskdepdata[dep][2], taskname, d, multilibs)
*** 0534: if not os.path.exists(manifest):
0535: continue
0536: with open(manifest, "r") as f:
0537: for l in f:
0538: l = l.strip()
File: '/usr/lib/python3.5/genericpath.py', lineno: 19, function: exists
0015:# This is false for dangling symbolic links on systems that support them.
0016:def exists(path):
0017: """Test whether a path exists. Returns False for broken symbolic links"""
0018: try:
*** 0019: os.stat(path)
0020: except OSError:
0021: return False
0022: return True
0023:
Exception: TypeError: stat: can't specify None for path argument
Signed-off-by: Jason Wessel <jason.wessel@windriver.com>
Signed-off-by: Ross Burton <ross.burton@intel.com>
|
|
Using glibc-locale to install locales only makes sense if we're using glibc.
Signed-off-by: Ross Burton <ross.burton@intel.com>
|
|
Signed-off-by: Ross Burton <ross.burton@intel.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
Signed-off-by: Ross Burton <ross.burton@intel.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
Signed-off-by: Ross Burton <ross.burton@intel.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
- No need to use bb.utils.which() as subprocess will search $PATH
- Clarity flow by moving the install inside the try/except
Signed-off-by: Ross Burton <ross.burton@intel.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
Prompted by bitbake commit 2ba8a6b25ccc12e7b543e8450121e5311c7a701d,
add .txz to the list of archives used within get_recipe_local_files()
here as well.
Signed-off-by: Paul Eggleton <paul.eggleton@linux.intel.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
In a few places we have checks to see path B is the parent of path A, by
adding / to the end of the path B and then seeing if path A starts with
the suffixed path B. Unfortunately there are two potential flaws:
(1) path A needs to be suffixed with / as well or the directory itself
won't match (semantics perhaps, but in a lot of scenarios returning True
is correct); (2) you need to run os.path.abspath() on both paths first
or you will wrongly return False for some relative paths where you
should return True. Let's solve this once and for all by writing a
function that takes care of these and put it in oe.path.
Signed-off-by: Paul Eggleton <paul.eggleton@linux.intel.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
Now that SIGGEN_EXCLUDE_SAFE_RECIPE_DEPS supports the wildcard syntax,
these exclusions no longer need to be hardcoded
Signed-off-by: Joshua Watt <JPEWhacker@gmail.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
The special string "*" on the left hand side of the dependency
specification matches all recipes except the one on the right hand side.
Signed-off-by: Joshua Watt <JPEWhacker@gmail.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
Changes to the icecc-create-env recipe should not cause all recipes to
rebuild just because the have inherited icecc.
Signed-off-by: Joshua Watt <JPEWhacker@gmail.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
Create a common function for locating task manifest files rather than
several implementations with missing pieces.
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
Currently do_rootfs gets to see all rpms in the deploy directory. This filters
that view to only rpms which the image recipe has actual depends upon which
potentially removes some sources of confusion in the image construction.
This makes builds more reproducibile and also fixes contamination issues
where dnf picks up packages it shouldn't be able to 'see'.
[YOCTO #12039]
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
parallel_make_argument() was incorrectly taking the maximum of the limit
and the calculated value instead of the minimum.
Signed-off-by: Joshua Watt <JPEWhacker@gmail.com>
Signed-off-by: Ross Burton <ross.burton@intel.com>
|
|
The code to extract the integer number of parallel build threads and
construct a new argument from them has started to be copied in multiple
locations, so create two new helper utilities to aid recipes.
The first helper (parallel_make()) extracts the integer number of
parallel build threads from PARALLEL_MAKE. The second
(parallel_make_argument()) does the same and then puts the result back
into a format string, optionally clamping it to some maximum value.
Additionally, rework the oe-core recipes that were manually doing this
to use the new helper utilities.
Signed-off-by: Joshua Watt <JPEWhacker@gmail.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
warning to a note
Now that the first boot deferral needs to be requested explicitly,
it's not really something to be concerned about.
Signed-off-by: Alexander Kanavin <alexander.kanavin@linux.intel.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
first boot
'exit 1' is not optimal for two reasons:
1) Code is hard to read; it is not obvious that it means 'defer what follows to first boot'.
2) Worse, this hides actual errors in the scriptlets; there is no difference between scriptlet
failing because it's intended to be run on target and scriptlet failing because there's a bug or
a regression somewhere.
The new, supported way is to place the code that has to run on target into pkg_postinst_ontarget(),
or, if a more fine-tuned control is required, call 'postinst-intercepts defer_to_first_boot' from
pkg_postinst() to explicitly request deferral to first boot.
Signed-off-by: Alexander Kanavin <alexander.kanavin@linux.intel.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
postinst_intercept hooks
That hook is empty, and doesn't need to be executed; it merely indicates that packages
that have used it are requesting to defer their postinst scripts to first boot
unconditionally.
Signed-off-by: Alexander Kanavin <alexander.kanavin@linux.intel.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
function
Signed-off-by: Alexander Kanavin <alexander.kanavin@linux.intel.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
Signed-off-by: Ross Burton <ross.burton@intel.com>
|
|
repos don't conflict
The repository indexes updated during do_populate_sdk_ext and do_populate_sdk
can conflcit. Add the missing lockfile calls for deb/ipk and in the rpm case,
ensure different directories are used for the index for the two sdk cases.
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
Colour value removes in red and additions in green, making it easier to scan the
output for relevant changes.
This adds a --colour option to specify whether colouring should be on, off, or
detected. The default is detected, and depends on whether stdout is a TTY (same
behaviour as git).
Signed-off-by: Ross Burton <ross.burton@intel.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
Using host gpg has been problematic, and particularly this removes
the need to serialize package creation, as long as --auto-expand-secmem
is passed to gpg-agent, and gnupg >= 2.2.4 is in use
(https://dev.gnupg.org/T3530).
Sadly, gpg-agent itself is single-threaded, so in the longer run
we might want to seek alternatives:
https://lwn.net/Articles/742542/
(a smaller issue is that rpm itself runs the gpg fronted in a serial
fashion, which slows down the build in cases of recipes with very
large amount of packages, e.g. glibc-locale)
Note that sstate signing and verification continues to use host
gpg, as depending on native gpg would create circular dependencies.
[YOCTO #12022]
Signed-off-by: Alexander Kanavin <alexander.kanavin@linux.intel.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
The entries of the created manifest file are always in a
different order. To ensure a deterministic build output
the entries are ordered alphabetically.
Signed-off-by: Michael Blättler <michael.blaettler@siemens.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
When we migrated rpm v5 -> v4, we lost the ability to drop "per file"
dependencies from the rpm backend for things like "/bin/bash" and
"/usr/bin/env" which meant the sdks were becomming 'bloated'.
This restores the functionality using a dummy package, similarly to
the way the buildtools perl issue was addressed. It also removes
the non-functional old code so as not to confuse people in future.
I ran into this problem trying to filter dependencies to only rpms
a build directly depends upon and it turns out we have some determinism issues
in this area so this is something key to fix.
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
It had been removed since 2011:
commit b774bf44ef004276da12a83ebd69715c00b596ac
Author: Lianhao Lu <lianhao.lu@intel.com>
Date: Tue Aug 16 16:26:49 2011 +0800
package(_ipk).bbclass: opkg using ALL_MULTILIB_PACKAGE_ARCHS
Signed-off-by: Robert Yang <liezhi.yang@windriver.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
This can make debug easier, for example, makes it easy to run the
command mannually.
Signed-off-by: Robert Yang <liezhi.yang@windriver.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
It had a problem when nested layer before, e.g.:
layer_a/layer_b/
And when layer_b is handled before layer_a, then layer_a dir existed, so
it would be treated as already handled, which was wrong, check
conf/layer.conf can fix the problem.
Signed-off-by: Robert Yang <liezhi.yang@windriver.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
Currently we can't access this function from lib/oe as its a class function.
Move it to allow such access.
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
This module refers to oe.types, so it needs to actually import oe.types.
Fixes errors when parsing certain OE-Core recipes within the layer
index update script.
Signed-off-by: Paul Eggleton <paul.eggleton@linux.intel.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
This function is not used by any classes or recipes that I can find, so lets
delete it.
Signed-off-by: Ross Burton <ross.burton@intel.com>
|
|
DNF has a time-based cache policy (and a great sense of humour) so it's possible
that 'dnf makecache' won't actually refresh any caches. Force the cache updates
by passing --refresh.
Signed-off-by: Ross Burton <ross.burton@intel.com>
|
|
Specifically:
1) remove +git${SRCPV} stuff from comparison and output; it's just
unnecessary clutter;
2) write the commit id of the latest version tag into the output;
this saves quite a bit of trouble of manually checking what that
commit id is when doing version updates;
3) when UPSTREAM_CHECK_COMMITS is set, ignore the tags altogether;
instead check if the latest commit is different to the one we use,
and if so, report that the recipe can be updated to said commit
(which is also written into the output, as in 2). Multiple
recipes are failing the upstream check because they never
issue tags, now we can fix them.
Signed-off-by: Alexander Kanavin <alexander.kanavin@linux.intel.com>
Signed-off-by: Ross Burton <ross.burton@intel.com>
|
|
If you're not sure what changes devtool finish is going to make, or
you're not sure you're finished with your modifications, it is useful to
be able to see what devtool finish is going to do beforehand, so add
a -N/--dry-run option to make that possible.
(It's also very useful for debugging devtool finish itself.)
Signed-off-by: Paul Eggleton <paul.eggleton@linux.intel.com>
Signed-off-by: Ross Burton <ross.burton@intel.com>
|
|
devtool upgrade did not properly handle setting SRC_URI checksums for
recipes that use named SRC_URI entries and also use those names in the
SRC_URI checksums. A further complication was where the name contained
an expression that changed with the version e.g. ${PV} (probably quite
rare, but the dnsmasq recipe in meta-networking is currently one such
recipe.) All of these are now handled properly.
Additionally, drop the _get_checksums() function that wasn't being
called from anywhere in the code.
Note that this now turns nowrap_vars in recipeutils.py to be a list of
regexes, hence things such as [ and ] need to be appropriately escaped.
Signed-off-by: Paul Eggleton <paul.eggleton@linux.intel.com>
Signed-off-by: Ross Burton <ross.burton@intel.com>
|
|
If a value was split over multiple lines (e.g. as SRC_URI usually is)
then we were inserting the value as one item in the lines list with
newlines between each line. There's nothing wrong with this if you're
writing the list out to a file, but if you want to generate a patch (as
patch_recipe_file() will do if the patch parameter is set to True) then
the diff output looks a bit odd. Split the value before adding it to the
lines list to resolve this.
Signed-off-by: Paul Eggleton <paul.eggleton@linux.intel.com>
Signed-off-by: Ross Burton <ross.burton@intel.com>
|
|
find_layerdir() should really return absolute paths, so make it do so.
This fixes devtool finish not deleting files it should do after devtool
upgrade if the specified path is relative, since the devtool finish code
was assuming that find_layerdir() was returning an absolute path.
Fixes [YOCTO #12318].
Signed-off-by: Paul Eggleton <paul.eggleton@linux.intel.com>
Signed-off-by: Ross Burton <ross.burton@intel.com>
|
|
A flaw was found on my Ubuntu 14.04.5 LTS, on which that gnome-terminal is
the default terminal, when I run any of the tasks:
bitbake busybox -c menuconfig/devshell/devpyshell
bitbake virtual/kernel -c menuconfig/devshell/devpyshell
I got a error as follows:
"Failed to execute child process "oe-gnome-terminal-phonehome" (No such file or directory)"
Seems the environment of the process calling Popen is not passed to the
child process, this behaviour is a known issue in Python bug tracker:
http://bugs.python.org/issue8557
It could be fixed by using an absolute path instead per test.
Signed-off-by: Ming Liu <liu.ming50@gmail.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
With a '*' as a wildcard for the signature here we can also match a
portion of the task name with the result that we may match a sigdata
file for the wrong task. Luckily the signature is always the same
length - 32 characters - so we can simply use 32 '?' characters instead.
(A regex would have been another alternative, but the wildcard should be
effective and I felt like a regex would complicate the code more than
this solution).
Fixes [YOCTO #11763].
Signed-off-by: Paul Eggleton <paul.eggleton@linux.intel.com>
Signed-off-by: Ross Burton <ross.burton@intel.com>
|
|
When we wish to use the package feed for local development, it does
not uses GPG signed feeds by default but dnf uses package signature
check. We need to configure the GPG signature check out so it works
out of box.
With this patch, installing non-signed packages works:
$: dnf install <package>
Signed-off-by: Otavio Salvador <otavio@ossystems.com.br>
Signed-off-by: Ross Burton <ross.burton@intel.com>
|
|
When faced with multiple sets of licenses combined with | (OR), it was
possible for oe.license.is_included() to choose a set of licenses with
a blacklisted license and then report failure, even if choosing
another set of licenses would have resulted in a successful
result. This happened when the chosen set still contained more
whitelisted licenses than the other set.
This change makes sure a set with any blacklisted license is always
considered with a lower weight than a set with only whitelisted
licenses.
Example: Faced with the license string "GPL-3.0 & GPL-2.0 & LGPL-2.1 |
Proprietary" and with "GPL-3.0" being blacklisted, the old code would
report a failure since "GPL-3.0 & GPL-2.0 & LGPL-2.1" still contains
more whitelisted licenses than "Proprietary" does.
This change also adds a unit test for oe.license.is_included().
Signed-off-by: Peter Kjellerstedt <peter.kjellerstedt@axis.com>
Signed-off-by: Ross Burton <ross.burton@intel.com>
|
|
To be able to better debug remove() behaviour, add more logging
to rpm calls via bb.note(). The change also makes remove() logging
more consistent with other package managers' remove() (e.g., opkg).
Signed-off-by: Mikko Ylinen <mikko.ylinen@linux.intel.com>
Signed-off-by: Ross Burton <ross.burton@intel.com>
|
