| HAWKEY(3) | Hawkey | HAWKEY(3) |
hawkey - Hawkey Documentation
Contents:
This document describes the API changes the library users should be aware of before upgrading to each respective version. It is our plan to have the amount of changes requiring changing the client code go to a minimum after the library hits the 1.0.0 version.
Depracated API items (classes, methods, etc.) are designated as such in this document. The first release where support for such items can be dropped entirely must be issued at least five months after the issue of the release that announced the deprecation and at the same time have, relatively to the deprecating release, either:
These criteria are likely to tighten in the future as hawkey matures.
Actual changes in the API are then announced in this document as well. ABI changes including changes in functions' parameter counts or types or removal of public symbols from libhawkey imply an increase in the library's SONAME version.
Query.filter() now returns a new instance of Query, the same as the original with the new filtering applied. This allows for greater flexibility handling the Query objects and resembles the way QuerySets behave in Django.
In practice the following code will stop working as expected:
q = hawkey.Query(self.sack) q.filter(name__eq="flying") # processing the query ...
It needs to be changed to:
q = hawkey.Query(self.sack) q = q.filter(name__eq="flying") # processing the query ...
The original semantics is now available via the Query.filterm() method, so the following will also work:
q = hawkey.Query(self.sack) q.filterm(name__eq="flying") # processing the query ...
In Python's Package instances accessors for string attributes now return None instead of the empty string if the attribute is missing (for instance a pkg.sourcerpm now returns None if pkg is a source rpm package already).
This change is towards a more conventional Python practice. Also, this leaves the empty string return value free to be used when it is actually the case.
The Query key value used for filtering by the repo name is HY_PKG_REPONAME now (was HY_PKG_REPO). The old value was misleading.
hy_repo_create() for Repo object initialization now needs to be passed a name of the repository.
All Goal methods accepting Query as the means of selecting packages, such as hy_goal_install_query() have been replaced with their Selector counterparts. Selector structures have been introduced for the particular purpose of specifying a package that best matches the given criteria and at the same time is suitable for installation. For a discussion of this decision see rationale_selectors.
Similar change happened in Python, the following constructs:
q = q.filter(repo="updates")
need to be changed to:
q = q.filter(reponame="updates")
The old version of this didn't allow using the same string to both construct the query and dynamically get the reponame attribute from the returned packages (used e.g. in DNF to search by user-specified criteria).
The following will no longer work:
if pkg.evr_eq(some_other_pkg):
...
Instead use the result of pkg.evr_cmp, for instance:
if pkg.evr_cmp(some_other_pkg) == 0:
...
This function compares only the EVR part of a package, not the name. Since it rarely make sense to compare versions of packages of different names, the following is suggested:
if pkg == some_other_pkg:
...
All instantiations of hawkey.Repo now must be given the name of the Repo. The following will now fail:
r = hawkey.Repo() r.name = "fedora"
Use this instead:
r = hawkey.Repo("fedora")
See Query installs obsoleted in the C section. In Python Queries will no longer work as goal target specifiers, the following will fail:
q = hawkey.Query(sack) q.filter(name="gimp") goal.install(query=q)
Instead use:
sltr = hawkey.Selector(sack) sltr.set(name="gimp") goal.install(select=sltr)
Or a convenience notation:
goal.install(name="gimp")
keyname parameter was added to the function signature. The new parameter allows filtering by a specific relation to the resulting packages, for instance:
hy_query_filter_package_in(q, HY_PKG_OBSOLETES, HY_EQ, pset)
only leaves the packages obsoleting a package in pset a part of the result.
The new version of hy_query_filter_package_in() handles this now, see above.
In Python, the following is no longer supported:
q = query.filter(obsoleting=1)
The equivalent new syntax is:
installed = hawkey.Query(sack).filter(reponame=SYSTEM_REPO_NAME) q = query.filter(obsoletes=installed)
The function was not systematic. Same result is achieved by obtaining obsoleting reldeps from a package and then trying to find the installed packages that provide it. In Python:
q = hawkey.Query(sack).filter(reponame=SYSTEM_REPO_NAME, provides=pkg.obsoletes)
The old name was by error, the functionality has not changed: this function has always returned the full NEVRA, skipping the epoch part when it's 0.
Previously, repr(pkg) would yield for instance <_hawkey.Package object, id: 5>. Now more complete information is present, including the package's NEVRA and repository: <hawkey.Package object id 5, foo-2-9\.noarch, @System>.
Also notice that the representation now mentions the final hawkey.Package type, not _hawkey.Package. Note that these are currently the same.
hy_sack_create() now accepts third argument, rootdir. This can be used to tell Hawkey that we are intending to do transactions in a changeroot, not in the current root. It effectively makes use of the RPM database found under rootdir. To make your code compile in 0.3.8 without changing functionality, change:
HySack sack = hy_sack_create(cachedir, arch);
to:
HySack sack = hy_sack_create(cachedir, arch, NULL);
It became necessary to differentiate between the default forms used by subject.nevra_possibilities() and subject.nevra_possibilities_real(). Therefore there is little sense in setting the default form for an entire Subject instance. The following code:
subj = hawkey.Subject("input", form=hawkey.FORM_NEVRA)
result = list(subj.nevra_possibilities())
is thus replaced by:
subj = hawkey.Subject("input")
result = list(subj.nevra_possibilities(form=hawkey.FORM_NEVRA))
hy_sack_create() now accepts fourth argument, flags, introduced to modify the sack behavior with boolean flags. Currently only one flag is supported, HY_MAKE_CACHE_DIR, which causes the cache directory to be created if it doesn't exist yet. To preserve the previous behavior, change the following:
HySack sack = hy_sack_create(cachedir, arch, rootdir);
into:
HySack sack = hy_sack_create(cachedir, arch, rootdir, HY_MAKE_CACHE_DIR);
Update your code by mechanically replacing the name.
A new sack by default no longer automatically creates the cache directory. To get the old behavior, append make_cache_dir=True to the Sack.__init__() arguments, that is change the following:
sack = hawkey.Sack(...)
to:
sack = hawkey.Sack(..., make_cache_dir=True)
Reflects the similar change in C API.
hy_goal_package_obsoletes() was flawed in that it only returned a single obsoleted package (in general, package can obsolete arbitrary number of packages and upgrade a package of the same name which is also reported as an obsolete). Use hy_goal_list_obsoleted_by_package() instead, to see the complete set of packages that inclusion of the given package in an RPM transaction will cause to be removed.
In other words, hy_goal_list_erasures() and hy_goal_list_obsoleted() return disjoint sets.
Directly reflecting the core changes. In particular, instead of:
obsoleted_pkg = goal.package_obsoletes(pkg)
use:
obsoleted = goal.obsoleted_by_package(pkg) # list obsoleted_pkg = obsoleted[0]
For old function behavior use new function hy_query_filter_latest_per_arch()
In Python's Query option latest in Query.filter() now filter only the latest packages ignoring architecture. The original semantics for filtering latest packages for each arch is now available via latest_per_arch option.
For example there are these packages in sack:
glibc-2.17-4.fc19.x86_64 glibc-2.16-24.fc18.x86_64 glibc-2.16-24.fc18.i686 >>> q = hawkey.Query(self.sack).filter(name="glibc") >>> map(str, q.filter(latest=True)) ['glibc-2.17-4.fc19.x86_64'] >>> map(str, q.filter(latest_per_arch=True)) ['glibc-2.17-4.fc19.x86_64', 'glibc-2.16-24.fc18.i686']
The functions were deprecated because there can be multiple advisories referring to a single package. Please use the new function hy_package_get_advisories() which returns all these advisories. New functions hy_advisory_get_* provide the data retrieved by the deprecated functions.
The only exception is the hy_package_get_update_severity() which will be dropped without any replacement. However advisory types and severity levels are distinguished from now and the type is accessible via hy_advisory_get_type(). Thus enum HyUpdateSeverity was also deprecated. A new HyAdvisoryType should be used instead.
The old functions will be dropped after 2014-07-07.
hy_goal_write_debugdata() has a new const char *dir argument to communicate the target directory for the debugging data. The old call:
hy_goal_write_debugdata(goal);
should be changed to achieve the same behavior to:
hy_goal_write_debugdata(goal, "./debugdata");
Analogous to core changes.
Attributes baseurl, location, sourcerpm, version, release, name, arch, description, evr, license, packager, reponame, summary and url of Package object return Unicode string.
The function was deprecated because we need more information about packages listed in an advisory than just file names. Please use the new function hy_advisory_get_packages() in combination with hy_advisorypkg_get_string() to obtain the data originally provided by the deprecated function.
The old function will be dropped after 2014-10-15 AND no sooner than in 0.4.21.
Instead of:
r = hawkey.Repo('name', cost=30)
use:
r = hawkey.Repo('name')
r.cost = 30
Also previously when no cost was given it defaulted to 1000. Now the default is 0. Both these aspects were present by mistake and the new interface is consistent with the C library.
The attribute was deprecated because the underlying C function was also deprecated. Please use the new attribute packages and the attribute filename of the returned objects to obtain the data originally provided by the deprecated attribute.
The old attribute will be dropped after 2014-10-15 AND no sooner than in 0.4.21.
All string attributes of Advisory and AdvisoryRef objects (except the deprecated filenames attribute) are Unicode objects now.
Previously, the function hy_chksum_str would cause a segmentation fault when it was used with incorrect type value. Now it correctly returns NULL if type parameter does not correspond to any of expected values.
hy_sack_create() now accepts fifth argument, logfile to customize log file path. If NULL parameter as logfile is given, then all debug records are written to hawkey.log in cachedir. To make your code compile in 0.5.3 without changing functionality, change:
HySack sack = hy_sack_create(cachedir, arch, rootdir, 0);
to:
HySack sack = hy_sack_create(cachedir, arch, rootdir, NULL, 0);
The function will be removed since hy_add_cmdline_package creates cmdline repository automatically.
The function will be dropped after 2015-06-23 AND no sooner than in 0.5.8.
This addition lets user specify log file path from Sack.__init__()
This change was already announced but it actually never happened.
The method will be removed since Sack.add_cmdline_package() creates cmdline repository automatically.
The method will be dropped after 2015-06-23 AND no sooner than in 0.5.8.
If the optional parameter is set to True, hawkey silently skips packages that can not be installed.
Hawkey is package manager agnostic and the yum phrase could be misleading.
The function will be dropped after 2015-10-27 AND no sooner than in 0.5.8.
Hawkey is package manager agnostic and the yum phrase could be misleading.
The method will be dropped after 2015-10-27 AND no sooner than in 0.5.8.
To make your code compile in 0.5.9 without changing functionality, change:
hy_goal_req_has_distupgrade_all(goal) hy_goal_req_has_erase(goal) hy_goal_req_has_upgrade_all(goal)
to:
hy_goal_has_actions(goal, HY_DISTUPGRADE_ALL) hy_goal_has_actions(goal, HY_ERASE) hy_goal_has_actions(goal, HY_UPGRADE_ALL)
respectively
To make your code compatible with hawkey 0.5.9 without changing functionality, change:
goal.req_has_distupgrade_all() goal.req_has_erase() goal.req_has_upgrade_all()
to:
goal.actions & hawkey.DISTUPGRADE_ALL goal.actions & hawkey.ERASE goal.actions & hawkey.UPGRADE_ALL
respectively
The hy_advisory_get_filenames() API call, the corresponding Python property filenames of class Advisory are removed. Instead, iterate over hy_advisory_get_packages() with hy_advisorypkg_get_string() and HY_ADVISORYPKG_FILENAME. No known hawkey API consumers were using this call.
Hawkey now has a dependency on GLib. Aside from the above hy_advisory_get_filenames() call, the Python API is fully preserved. The C API has minor changes, but the goal is to avoid causing a significant amount of porting work for existing consumers.
The hy_package_get_files API call now returns a char **, allocated via g_malloc. Free with g_strfreev.
The HyStringArray type is removed, as nothing now uses it.
HyPackageList is now just a GPtrArray, though the existing API is converted into wrappers. Notably, this means you can now use g_ptr_array_unref().
Aside from the one change below, the Python bindings should be unaffected by the C API changes.
See the README.
Yes, look at:
It is entirely up to you. Hawkey does not provide any means to do this automatically, for instance from your /etc/yum.repos.d configuration. Use or build tools to do that. For instance, both Yum and DNF deals with the same problem and inside they employ urlgrabber to fetch the files. A general solution if you work in C is for instance libcurl. If you are building a nice downloading library that integrates well with hawkey, let us know.
Because downloading things from remote servers is a differnt domain full of its own complexities like HTTPS, parallel downloads, error handling and error recovery to name a few. Downloading is a concern that can be naturally separated from other parts of package metadata managing.
IMPORTANT:
First of, make sure hawkey is installed on your system, this should work from your terminal:
>>> import hawkey
Sack is an abstraction for a collection of packages. Sacks in hawkey are toplevel objects carrying much of hawkey's of functionality. You'll want to create one:
>>> sack = hawkey.Sack() >>> len(sack) 0
Initially, the sack contains no packages.
hawkey is a lib for listing, querying and resolving dependencies of packages from repositories. On most linux distributions you always have at least the system repo (in Fedora it is the RPM database). To load it:
>>> sack.load_system_repo() >>> len(sack) 1683
Hawkey always knows the name of every repository. The system repository is always set to hawkey.SYSTEM_REPO_NAME. and the client is responsible for naming the available repository metadata.
Let's be honest here: all the fun in packaging comes from packages you haven't installed yet. Information about them, their metadata, can be obtained from different sources and typically they are downloaded from an HTTP mirror (another possibilities are FTP server, NFS mount, DVD distribution media, etc.). Hawkey does not provide any means to discover and obtain the metadata locally: it is up to the client to provide valid readable paths to the repository metadata XML files. Structures used for passing the information to hawkey are the hawkey Repos. Suppose we somehow obtained the metadata and placed it in /home/akozumpl/tmp/repodata. We can then load the metadata into hawkey:
>>> path = "/home/akozumpl/tmp/repodata/%s"
>>> repo = hawkey.Repo("experimental")
>>> repo.repomd_fn = path % "repomd.xml"
>>> repo.primary_fn = path % "f7753a2636cc89d70e8aaa1f3c08413ab78462ca9f48fd55daf6dedf9ab0d5db-primary.xml.gz"
>>> repo.filelists_fn = path % "0261e25e8411f4f5e930a70fa249b8afd5e86bb9087d7739b55be64b76d8a7f6-filelists.xml.gz"
>>> sack.load_repo(repo, load_filelists=True)
>>> len(sack)
1685
The number of packages in the Sack will increase by the number of packages found in the repository (two in this case, it is an experimental repo after all).
What the load_filelists=True argument to load_repo() above does is instruct hawkey to process the <hash>filelists.xml.gz file we passed in and which contains structured list of absolute paths to all files of all packages within the repo. This information can be used for two purposes:
Some files provided by a package (e.g those in /usr/bin) are always visible even without loading the filelists. Well-behaved packages requiring only those can be thus resolved directly. Unortunately, there are packages that don't behave and it is hard to tell in advance when you'll deal with one.
The strategy for using load_filelists=True is thus:
Internally to hold the package information and perform canonical resolving hawkey uses Libsolv. One great benefit this library offers is providing writing and reading of metadata cache files in libsolv's own binary format (files with .solv extension, typically). At a cost of few hundreds of milliseconds, using the solv files reduces repo load times from seconds to tens of milliseconds. It is thus a good idea to write and use the solv files every time you plan to use the same repo for more than one Sack (which is at least every time your hawkey program is run). To do that use build_cache=True with load_repo() and load_system_repo():
>>> sack = hawkey.Sack(make_cache_dir=True) >>> sack.load_system_repo(build_cache=True)
By default, Hawkey creates @System.cache under the /var/tmp/hawkey-<your_login>-<random_hash> directory. This is the hawkey cache directory, which you can always delete later (deleting the cache files in the process). The .solv files are picked up automatically the next time you try to create a hawkey sack. Except for a much higher speed of the operation this will be completely transparent to you:
>>> s2 = hawkey.Sack() >>> s2.load_system_repo()
By the way, the cache directory (if not set otherwise) also contains a logfile with some boring debugging information.
Query is the means in hawkey of finding a package based on one or more criteria (name, version, repository of origin). Its interface is loosely based on Django's QuerySets, the main concepts being:
For instance, let's say I want to find all installed packages which name ends with gtk:
>>> q = hawkey.Query(sack).filter(reponame=hawkey.SYSTEM_REPO_NAME, name__glob='*gtk') >>> for pkg in q: ... print str(pkg) ... NetworkManager-gtk-1:0.9.4.0-9.git20120521.fc17.x86_64 authconfig-gtk-6.2.1-1.fc17.x86_64 clutter-gtk-1.2.0-1.fc17.x86_64 libchamplain-gtk-0.12.2-1.fc17.x86_64 libreport-gtk-2.0.10-3.fc17.x86_64 pinentry-gtk-0.8.1-6.fc17.x86_64 python-slip-gtk-0.2.20-2.fc17.noarch transmission-gtk-2.50-2.fc17.x86_64 usermode-gtk-1.109-1.fc17.x86_64 webkitgtk-1.8.1-2.fc17.x86_64 xdg-user-dirs-gtk-0.9-1.fc17.x86_64
Or I want to find the latest version of all python packages the Sack knows of:
>>> q.clear() >>> q = q.filter(name='python', latest_per_arch=True) >>> for pkg in q: ... print str(pkg) ... python-2.7.3-6.fc17.x86_64
You can also test a Query for its truth value. It will be true whenever the query matched at least one package:
>>> q = hawkey.Query(sack).filter(file='/boot/vmlinuz-3.3.4-5.fc17.x86_64') >>> if q: ... print 'match' ... match >>> q = hawkey.Query(sack).filter(file='/booty/vmlinuz-3.3.4-5.fc17.x86_64') >>> if q: ... print 'match' ... >>> if not q: ... print 'no match' ... no match
NOTE:
Many Sack sessions culminate in a bout of dependency resolving, that is answering a question along the lines of "I have a package X in a repository here, what other packages do I need to install/update to have X installed and all its dependencies recursively satisfied?" Suppose we want to install the RTS game Spring. First let's locate the latest version of the package in repositories:
>>> q = hawkey.Query(sack).filter(name='spring', latest_per_arch=True) >>> pkg = hawkey.Query(sack).filter(name='spring', latest_per_arch=True)[0] >>> str(pkg) 'spring-88.0-2.fc17.x86_64' >>> pkg.reponame 'fedora'
Then build the Goal object and tell it our goal is installing the pkg. Then we fire off the libsolv's dependency resolver by running the goal:
>>> g = hawkey.Goal(sack) >>> g.install(pkg) >>> g.run() True
True as a return value here indicates that libsolv could find a solution to our goal. This is not always the case, there are plenty of situations when there is no solution, the most common one being a package should be installed but one of its dependencies is missing from the sack.
The three methods Goal.list_installs(), Goal.list_upgrades() and Goal.list_erasures() can show which packages should be installed/upgraded/erased to satisfy the packaging goal we set out to achieve (the mapping of str() over the results below ensures human readable package names instead of numbers are presented):
>>> map(str, g.list_installs()) ['spring-88.0-2.fc17.x86_64', 'spring-installer-20090316-10.fc17.x86_64', 'springlobby-0.139-3.fc17.x86_64', 'spring-maps-default-0.1-8.fc17.noarch', 'wxBase-2.8.12-4.fc17.x86_64', 'wxGTK-2.8.12-4.fc17.x86_64', 'rb_libtorrent-0.15.9-1.fc17.x86_64', 'GeoIP-1.4.8-2.1.fc17.x86_64'] >>> map(str, g.list_upgrades()) [] >>> map(str, g.list_erasures()) []
So what does it tell us? That given the state of the given system and the given repository we used, 8 packages need to be installed, spring-88.0-2.fc17.x86_64 itself included. No packages need to be upgraded or erased.
For certain simple and commonly used queries we can do installs directly. Instead of executing a query however we instantiate and pass the Goal.install() method a Selector:
>>> g = hawkey.Goal(sack) >>> sltr = hawkey.Selector(sack).set(name='emacs-nox') >>> g.install(select=sltr) >>> g.run() True >>> map(str, g.list_installs()) ['spring-88.0-2.fc17.x86_64', 'spring-installer-20090316-10.fc17.x86_64', 'springlobby-0.139-3.fc17.x86_64', 'spring-maps-default-0.1-8.fc17.noarch', 'wxBase-2.8.12-4.fc17.x86_64', 'wxGTK-2.8.12-4.fc17.x86_64', 'rb_libtorrent-0.15.9-1.fc17.x86_64', 'GeoIP-1.4.8-2.1.fc17.x86_64'] >>> len(g.list_upgrades()) 0 >>> len(g.list_erasures()) 0
Notice we arrived at the same result as before, when a query was constructed and iterated first. What Selector does when passed to Goal.install() is tell hawkey to examine its settings and without evaluating it as a Query it instructs libsolv to find the best matching package for it and add that for installation. It saves user some decisions like which version should be installed or what architecture (this gets very relevant with multiarch libraries).
So Selectors usually only install a single package. If you mean to install all packages matching an arbitrarily complex query, just use the method describe above:
>>> map(goal.install, q)
This reference manual describes Python API to the library. For a quick start take a look at tutorial-py. To be sure that you are familiar with our deprecation policy, see changes.
NOTE:
WARNING:
API Documentation Contents
WARNING:
cachedir is a string giving a path of a cache location.
arch is a string specifying an architecture.
rootdir is a string giving a path to an installroot.
pkgcls is a class of packages retrieved from the sack. The class' __init__ method must accept two arguments. The first argument is a tuple of the sack and the ID of the package. The second argument is the pkginitval argument. pkginitval cannot be None if pkgcls is specified.
make_cache_dir is a boolean that specifies whether the cache should be used to speedup loading of repositories or not (see building_and_reusing_the_repo_cache-label).
logfile is a string giving a path of a log file location.
This is the inverse operation of add_excludes(). Any package that is not in the union of all the included packages is excluded. This works in conjunction with exclude and doesn't override it. So, if you both include and exclude the same package, the package is considered excluded no matter of the order.
WARNING:
repo is an optional Repo object that represents the system repository. The object is updated during the loading.
build_cache is a boolean that specifies whether the information should be written to the cache (see building_and_reusing_the_repo_cache-label).
repo is the Repo object to be processed. At least its Repo.repomd_fn must be set. If the cache has to be updated, Repo.primary_fn is needed too. Some information about the loading process and some results of it are written into the internal state of the repository object.
build_cache is a boolean that specifies whether the information should be written to the cache (see building_and_reusing_the_repo_cache-label).
load_filelists, load_presto and load_updateinfo are booleans that specify whether the Repo.filelists_fn, Repo.presto_fn and Repo.updateinfo_fn files of the repository should be processed. These files may contain information needed for dependency solving, downloading or querying of some packages. Enable it if you are not sure (see case_for_loading_the_filelists-label).
When an error or an unexpected event occurs during a Hawkey routine, an exception is raised:
The class hierarchy for Hawkey exceptions is:
+-- hawkey.Exception
+-- hawkey.ValueException
| +-- hawkey.QueryException
| +-- hawkey.ArchException
+-- hawkey.RuntimeException
+-- hawkey.ValidationException
name is a string giving the name of the repository. The name should not be equal to hawkey.SYSTEM_REPO_NAME nor hawkey.CMDLINE_REPO_NAME
Indices:
Since both a Query and a Selector work to limit the set of all Sack's packages to a subset, it can be suggested the two concepts should be the same and e.g. Queries should be used for Goal specifications instead of Selectors:
// create sack, goal, ... HyQuery q = hy_query_create(sack); hy_query_filter(q, HY_PKG_NAME, HY_EQ, "anaconda") hy_goal_install_query(q)
This arrangment was in fact used in hawkey prior to version 0.3.0, just because Queries looked like a convenient structure to hold this kind of information. It was unfortunately confusing for the programmers: notice how evaluating the Query q would generally produce several packages (anaconda for different architectures and then different versions) but somehow when the same Query is passed into the goal methods it always results in up to one pacakge selected for the operation. This is a principal discrepancy. Further, Query is universal and allows one to limit the package set with all sorts of criteria, matched in different ways (substrings, globbing, set operation) while Selectors only support few. Finally, while a fresh Query with no filters applied corresponds to all packages of the Sack, a fresh Selector with no limits set is of no meaning.
An alternative to introducing a completely different concept was adding a separate constructor function for Query, one that would from the start designate the Query to only accept settings compatible with its purpose of becoming the selecting element in a Goal operation (in Python this would probably be implemented as a subclass of Query). But that would break client's assumptions about Query (the unofficial C++ FAQ takes up the topic).
Implementation note: Selectors reflect the kind of specifications that can be directly translated into Libsolv jobs, without actually searching for a concrete package to put there. In other words, Selectors are specifically designed not to iterate over the package data (with exceptions, like glob matching) like Queries do. While Hawkey mostly aims to hide any twists and complexities of the underlying library, in this case the combined reasons warrant a concession.
Indices and tables
Aleš Kozumplík
2012-2021, Red Hat, Licensed under GPLv2+
| April 14, 2021 | 0.55.2 |