In the most cases it was enough to just add lock. In case of worker
trigger, since there is atomic operation on timer, it was also required
to add queue (coz python doesn't have atomics)
Some parsers are shared between different subcommands. It causes errors
when wwe add new arguments to one of them. This commit adds some tests
to cover those cases (except for well-known differencies)
Add support of changes generation. Changes will be generated (unless explicitly asked not to) automatically during check process (i.e. `repo-update --dry-run` and aliases) and uploaded to the remote server. Changes can be reviewed either by web interface or by special subcommands.
Changes will be automatically cleared during next successful build
Old solution causes amount of thread to be growing as well as stack is
increased during each iteration. Instead of cycle-free implementation,
this commit just uses while cycle
It has been a while since all pages have moved to json instead of form
data, except for login page. This commit changes login to json data
instead of form one
Instead of automatic package addition now it is required to add package
manually after clone. Less magic, plus would allow to use caches for
multi-repo setup (see #109)
* Allow to use single web instance for any repository
* some improvements
* drop includes from user home directory, introduce new variables to docker
The old solution didn't actually work as expected, because devtools
configuration belongs to filesystem (as well as sudo one), so it was
still required to run setup command.
In order to handle additional repositories, the POSTSETUP and PRESETUP
commands variables have been introduced. FAQ has been updated as well
* raise 404 in case if repository is unknown
Earlier applied fix bc9682373d introduced
errors with interaction, because (in docker container) HOME variable was
passed also to subprocesses. This fix limits variables to be passed to
the whitelisted ones
* allow to use one application for multiple repositories
* update tests
* handle None append argument everywhere
* rewrite repository definition logic
* drop optional flags from docs
* support of new schema in systemd units
* add migration docs and ability to migrate tree automatically
* use repostory id instead
* verbose multiarchitectureerror
* object path support for s3 sync
* fix tests after rebase
In some cases for better readability of logs, exceptions are now raised
without parent exception stacktrace. Also updated docs and contributing
guidelines
* add support of remote task tracking
* add remote call trigger implementation
* docs update
* add cross-service upload
* add notes about user
* add more ability to control upload
* multipart upload with signatures as well as safe file save
* configuration reference update
* rename watcher methods
* erase logs based on current package version
Old implementation has used process id instead, but it leads to log
removal in case of remote process trigger
* add --server flag for setup command
* restore behavior of the httploghandler
This field is required in order to pass config validation in case if
section name differs from default one. Also by default keyring_generator
and mirrorlist_generator have been renamed to keyring-generator and
mirrorlist-generator respectively for consistence
It makes sense to read some values from environment. In particular this
feature is useful in case of running application in containers in ci/cd
See #108 for more details
Used implementation of the hasher includes salt itself, thus additional
salt is optional and can be safely (in terms of security) treat as empty
string
Since llast upgrade build is broken. Lets fully migrate to
pyproject.toml. Note for maintaners: because data_files option is
deprectated (see https://github.com/pypa/setuptools/discussions/2648)
you will have to install files manually inside your packaging process
The new --(no-)increment flag has been added to add, update and rebuild
subcommands. In case if it is true and package version is the same as in
repository, it will automatically bump pkgrel appending (increasing)
minor part of it (e.g. 1.0.0-1 -> 1.0.0-1.1).
Inn order to implement this, the shadow (e.g. it will not store it in
database) patch for pkgrel will be created
* The issue appears when repository contains PKGBUILD in root. In this
case it will copy tree with loosing package information, because
the repository will be cloned to temporary path with random generated
name
* The issue appears when branch which is different from master is used
for any reposittory with git files (e.g. single-pkgbuild repo or repo
with submodules)
The main reason for having shell handler is to be able to fix if
something (e.g. migrations) goes wrong. In this way we need to reduce
actions inside this wrapper
This feature sometimes causes the main process termination. Since the
child process has been already terminated (or going to) there is no need
to terminate it manually
This migration includes
* removal of community repository as it is no more
* fixed paths for devtools configurations
* migration of archlinux packaging git url
Original implementation sends requests to httpbin which sometimes might
not be available. With proposed changes we are blocking redirects and
just check request itself
In previous revisions server was terminated by itself, thus no lock or
socket was removed. In new version, graceful termination of the queue
has been added as well as server now handles singals
Old implementation has used add step in order to fetch dependencies,
which could lead to build errors in case if dependency list was updated.
New solution uses dependencies which are declared at current version and
fetch them (if required and if enabled) before update process.
Closes#90
The newest mypy produces the following warning:
src/ahriman/application/handlers/search.py:43: error: Non-overlapping identity check (left operand type: "Union[_DefaultFactory[Any], Literal[_MISSING_TYPE.MISSING]]", right operand type: "Type[List[Any]]") [comparison-overlap]
which is more likely caused by updated dataclass models to protoocol (however decorators are still calllable). This commit masks problematic line from checking
This change requires srcinfo at least 0.1.2 version. Unfortunatelly aur
api don't support architecture specific arrays for now, so we just leave
it as is
Closes#82
In some cases (e.g. during addition of the package to build queue) we don't have
full information about package inself; in these cases we produce lines
with empty architecture, which duplicates normal ones.
This commit changes architecture column type to required and also
filters packages which don't have architecture set yet.
Closes#83
This commit also extends configuration of the multilib option, adding
the ability to exlcude multilib repository from repositories list
Note, that in order to support repository list and mirror correctly,
alpm configuration section is now architectture specific
Some commands have been moved to another group and thus having another
default name (old subcommands are still available...for now):
* daemon -> repo-daemon
* key-import -> service-key-import
* repo-clean -> service-clean
* repo-config -> service-config
* repo-config-validate -> service-config-validate
* repo-setup -> service-setup
* repo-shell -> service-shell
* version -> help-version
Note that this commit contains the following breaking changes:
* remote pull and remote push triggers are now enabled by default (with
empty target list)
* remote pull and remote push triggers now require target option to be
set (old behaviour had fallback on `gitremote`)
* validation is now considered to be stable, so it is enabled by default
in docker image (can be disabled however)
Some commands were made unsafe in old versions, but nowadays they can be
run without having special privileges.
There was also a bug in which status commands were not available if you
are not ahriman user and unix socket is used. It has been fixed by
switching to manual socket creation (see also
https://github.com/aio-libs/aiohttp/issues/4155)
This reverts commit 11732a8609.
Original solution has introduced special workaround (strict flag) which
contradicts the concept of immutable context. Moreover, it introduces
possible side-effects, because child process will use the one set by
parent instead of having own one.
The correct solution is to re-create context in process entry point
Sorry, it was Jan 1 and I was drunk :(
