Using absolute paths breaks on systems that place the binaries in
another directory and it defeats local overlays with PATH.
Signed-off-by: Henning Schild <henning.schild@siemens.com>
This patch adds a command line argument to the shell plugin, that allows to
enter the shell without changing the bitbake configuration.
Signed-off-by: Claudius Heine <ch@denx.de>
Before this patch, starting the shell in a clean environment, does not
checkout all repository or configures all layers correctly.
This patch adds the missing checkout and configuration steps to the
shell plugin.
Signed-off-by: Claudius Heine <ch@denx.de>
Having more than one them is a sign of serious problems in the
configuration. Better catch that early and fail loudly.
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Do not swallow errors of the init script. Rather leave our wrapper
immediately and let run_cmd report the splash and terminate kas.
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Instead of flooding the log with misleading error messages, always print
a note which path is used as root for repositories that have no URLs.
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
If the repo, where the configuration file resides in (in-tree
configuration) is not a git repository, take the directory where the
configuration file is located at as the path of this repo instead of failing.
This patch does not need to raise the configuration file version,
because it does not depend on the configuration file.
Signed-off-by: Claudius Heine <ch@denx.de>
This allows to define GIT_PROXY_COMMAND="oe-git-proxy" outside of kas
and pass this to the build. Now we can pull from native git repos when
there is only a http proxy available.
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Sort the header sections by their keys and the layers by their names.
This will re-enable writing deterministic config files, and bitbake can
use its cache again, accelerating rebuild startup times a lot.
While at it, indent the layers in bblayers.conf nicely.
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Use a separately incremented integer to track the configuration file
format version. We start with 2 due to the change that 23c3a951f6
introduced. 1 is declared to be equivalent to the original '0.10'.
The separate versioning has the advantage of being able to increment it
already during the development cycle, and using it with config files
that test/exploit the new format.
Using an integer has the advantages of a) differentiating it clearly
from the now independent kas version and b) simplifying the version
parsing. We can now also remove the string type restriction.
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
When a configuration file depends on a include from a repo, which is
defined in another included file, a 'KeyError' occures.
This fixes this, so that only known repos are tried to be fetched.
Signed-off-by: Claudius Heine <ch@denx.de>
This exception is always thrown with both a message and a file name.
Combine them to a single string, instead of throwing tuples.
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
ConfigStatic does not respect the --target overwrite so far. Fix this
by pulling the default definition into ConfigPython (ConfigStatic
already does that via the Config base class) and set the target provided
via the command line in the static config after parsing all files.
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
As suggested by Claudius Heine, reorder the blocks so that we can add
the pylint exception to the if half.
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
The `load_config` function in the includehandler already checks for the
file extension, so checking it before that, is redundant.
Signed-off-by: Claudius Heine <ch@denx.de>
The application should not be stopped from this code, removed `sys.exit`
and added exceptions to improve error handling.
Also added some more checks that were missing.
Signed-off-by: Claudius Heine <ch@denx.de>
In case of ConfigStatic, proxy settings only come from the host
environment. We need them early, prior to fetching the first missing
repo that is pulled via an include.
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
For me 'distro.id()' returned 'ubuntu' and thats not equal to 'Ubuntu'.
This patch converts the distro id to lowercase before comparing it.
Signed-off-by: Claudius Heine <ch@denx.de>
This seems to be the cleanest and simplest way of properly terminating
kas while one or more sub-processes are running: just ignore the signal.
The sub-processes will get it as well, and they will terminate
themselves properly. kas will get an error then and shut down. Problem
solved.
Forwarding the signal didn't work reliably anyway because we couldn't
count on kas being a process group leader in all cases.
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
This behavior got lost through 0bdd7a8d52. Specifically, errors during
git clone no longer shut down kas. Fix this by reporting back an error
code from _repo_fetch_async and terminate in repos_fetch if it's
non-zero.
Drop the return value of repos_fetch - no caller expects any.
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Make the header sections in local.conf and bblayers.conf easier
identifiable by naming them and writing them in chunks, separated by
blank lines. This can help with debugging the resulting configuration.
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Another, hopefully better attempt: Ignore the TERM signal in the sending
process so that we can properly clean up all the futures before actually
closing. This also avoids a "Terminated" message even on regular kas
endings.
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
That only creates exceptions and complaints if more routines are running
or we are already in the exiting process. Rather leave the coroutine and
catch the error on the waiting end.
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Don't add a newline to stderr dumps on failing commands, the logger does
that already. And don't print success messages if the fetch command
actually failed.
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Documentation for the 'build' and 'shell' command line were missing.
Also fixed kas command line documentation mentioning 'ebs-yocto'.
Signed-off-by: Claudius Heine <ch@denx.de>
With Sphinx it is possible to create the command line documentation
automatically if there is a function that returns just the command line
parser.
Currently the creation of the argument parser is rather entangled with
the rest of kas. This patches seperates this.
Signed-off-by: Claudius Heine <ch@denx.de>
When updating the previously used version of pylint 1.6.5 to 1.7.2
some new issues where found.
This patch fixes these issues:
************* Module kas.config
E:402, 0: Bad option value 'redefined-variable-type' (bad-option-value)
************* Module kas.includehandler
E: 33, 0: No name 'version' in module 'distutils' (no-name-in-module)
E: 33, 0: Unable to import 'distutils.version' (import-error)
R:239,12: Unnecessary "else" after "return" (no-else-return)
************* Module kas.libkas
C:214, 7: Do not use `len(SEQUENCE)` as condition value
(len-as-condition)
************* Module kas.repos
R: 54,12: Unnecessary "else" after "return" (no-else-return)
Signed-off-by: Claudius Heine <ch@denx.de>
Changing even simple settings like target or machine requires the
creation of additional configuration files. This does not scale well.
This patch allows the target, machine and distro to be overwritten by
environment variables (called KAS_TARGET, KAS_MACHINE and KAS_DISTRO
respectively).
It also fixes how the environment variables for the proxy settings are
handled. Currently the settings in the config files overwrite the proxy
settings from the environment, but since that would be inconsistent this
patch switches that around. With this patch the environment overwrites
the proxy settings in the configuration file.
Signed-off-by: Claudius Heine <ch@denx.de>
Building on top of run_cmd_async, this reworks repo_fetch to a
repository list fetcher repos_fetch that runs those operations in
parallel.
The two users, ReposFetch and ConfigStatic, are converted to exploit
this parallelization.
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Make it more straightforward by removing complete variable and
terminating the loop on the condition - no more missing repos -
directly. Rename variables to clarify what they reference.
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Enhance _stream_subprocess to run_cmd_async, a co-routine variant of
run_cmd that the caller can use to parallelize command execution.
run_cmd becomes a simple wrapper that waits for the async variant to
complete.
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Instead of writing "commit", issue a more telling log message if some
repo already contains the required hash.
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
The two code paths are unfortunate. Factor out the difference and use a
single run_cmd invocation. This also adds the forgotten -q when using
a reference directory.
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
The current implementation did some more hackish solution to work
around some of kas infrastructure, without changing to much.
This patch cleans this up and therefor remove the more obsure part
of the include mechansim.
Signed-off-by: Claudius Heine <ch@denx.de>
Isar allows to build Debian-based images using bitbake and the usual
layer structures. It's very similar to OE, we just need to account for
a different init script name.
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
If the configuration file structure changes over time, it might
introduce incompatibilities.
This patch introduces versioning of configuration file.
Configuration files needs to add a 'version' entry into their 'header'
like this:
header:
version: '0.9'
After loading this file, the version is checked agains kas version.
Every version 'M.m.p' of kas is allowed to load configuration files with
version 'M.m' and all versions backwards up to and including
the '__compatible_version__' from kas/__version__.py
Signed-off-by: Claudius Heine <ch@denx.de>
In the current configuration file we mix between entries that are file
specific (like includes and later version) and entries that are used by
the rest of kas, that are not file specific (machine, target, repos,
etc.) It might not be obvious to differenciate between those entries.
This patch introduces a header section into the configuration file, to
contain every setting that is specific to the file and not the complete kas
configuration. The 'includes' statement is moved there, because its file
specific.
Signed-off-by: Claudius Heine <ch@denx.de>
Splitting configuration files into multiple files and implementing an
include mechanism allows to handle configurations more flexible and
allowing following scenarios:
- including kas configuration file from external meta layer into
own project
- splitting many different similar configurations into multiple files
with only a couple of common files
To include file its necessary to add a 'includes' entry into the kas
configuration. To include files from the same repo, do it like this:
-----
includes:
- ../base/base_include.yml
-----
The path is relative to the current configuration file. If they start
with a path seperator ("/") they are absolute.
To include files from a different repository, do it like this:
-----
includes:
- file: bsps/my_machine_include.yml
repo: collected_machines
repos:
collected_machines:
url: https://url.to.git.repo/
revspec: master
-----
You have to make sure that the repository definition is available
in your current file, or in any include that is available at this
point.
Yaml ("*.yml") and Json ("*.json") files are supported.
Included in this patch are also some changes to the configuration
file structure. Here is an overview:
The value of the 'repos' key is a dictionary that maps a repo identifier
to a dictionary that contains 5 entries:
- url: The url to the repository. If that is missing, no git operations
are used
- refspec: The git refspec of the repository that is used. If that is
missing the latest commit of the default branch is used.
- name: The is defines under which name the repository is stored. If
that is missing the repository identifier is used
- path: The path where the repository is stored. If no url is given and
a path is missing, the repository is referencing the repository under
which the configuration file is stored. If no url is given and a path is
specified, the repository is referencing a directory where layers might
be stored. If an url is specified, path can be used to overwrite the
default (kas_work_dir + repo.name) checkout directory.
- layers: This contains a dictionary of layers, that should be included
into the bblayers.conf. The keys are paths relative to the repository
and the values can be used to exclude layers if they are one of
"excluded", "disabled", "false", "0", "n", or "no". Also boolean values
are accepted. Any other value, including "None" means that this layer is
included into the bblayers.conf. If "layers" is missing or empty, the
repository itself is included into the bblayers. If this is specified, the
repository itself is not included into the bblayers.conf.
Signed-off-by: Claudius Heine <ch@denx.de>
It was decided that 'sublayers' is to confusing, because this term is
not used in the bitbake/openembedded context.
Signed-off-by: Claudius Heine <ch@denx.de>
Config.has_changed didn't work reliably so far, so the state cache was
only allowing to associate a work directory with configuration file.
However this was rather unintuitive, specifically when moving the work
dir to a different home - nothing told the user about the persistent
association.
We therefore agreed to remove the state cache for now. It can be
reintroduced at any time once a consistent usage model exists.
To establish a relationship to a work directory inside a shell session,
kas now evaluates the KAS_WORK_DIR environment variable.
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
This shortens the include from "from .__version__ import __version__"
to "from . import __version__" within the module.
Signed-off-by: Claudius Heine <ch@denx.de>
Currently the shell provided via the 'shell' command is always '/bin/sh'
and does not have a TERM variable set. This is a rather spartan shell
experience.
With this patch we add the TERM variable to the whitelist, so that
pagers work better and made the shell binary dependent from the SHELL
environment variable. If no such variable is set, we fall back to using
'/bin/sh'.
Signed-off-by: Claudius Heine <ch@denx.de>
The ssh-agent environment variables of an ssh-agent that is running
outside the kas build system are not transfered to the bitbake
environment. This causes fetching of private git repositories to
fail.
This patch adds those variables to the environment whitelist.
Signed-off-by: Claudius Heine <ch@denx.de>
