Contents

Introduction

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.

Changes in 0.2.10

Python bindings

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 ...

Changes in 0.2.11

Python bindings

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.

Changes in 0.3.0

Core

Query: key for reponame filtering

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.

Repo initialization

hy_repo_create() for Repo object initialization now needs to be passed a name of the repository.

Query installs obsoleted

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 Selectors are not Queries.

Python bindings

Query: filtering by repository with the reponame key

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).

Package: removed methods for direct EVR comparison

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:


    ...

Repo initialization

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")

Query installs obsoleted

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")

Changes in 0.3.1

Query: hy_query_filter_package_in() takes a new parameter

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.

Removed hy_query_filter_obsoleting()

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)

Changes in 0.3.2

Removed hy_packagelist_of_obsoletes.

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)

Changes in 0.3.3

Renamed hy_package_get_nvra to hy_package_get_nevra

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.

Changes in 0.3.4

Python bindings

pkg.__repr__() is more verbose now

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.

Changes in 0.3.8

Core

New parameter rootdir to hy_sack_create()

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);

Python bindings

Forms recognized by Subject are no longer an instance-scope setting

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))

Changes in 0.3.9

Core

Flags for hy_sack_create

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);

hy_sack_get_cache_path is renamed to hy_sack_get_cache_dir

Update your code by mechanically replacing the name.

Python bindings

make_cache_dir for Sack's constructor

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)

cache_path property of Sack renamed to cache_dir

Reflects the similar change in C API.

Changes in 0.3.11

Core

hy_goal_package_obsoletes() removed, hy_goal_list_obsoleted_by_package() provided instead

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.

hy_goal_list_erasures() does not report obsoletes

In other words, hy_goal_list_erasures() and hy_goal_list_obsoleted() return disjoint sets.

Python bindings

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]

Changes in 0.4.5

Core

Query: hy_query_filter_latest() now filter latest packages ignoring architecture

For old function behavior use new function hy_query_filter_latest_per_arch()

Python bindings

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']

Changes in 0.4.13

Core

Deprecated hy_package_get_update_*

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.

Changes in 0.4.15

Core

hy_goal_write_debugdata() takes a directory parameter

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");

Python bindings

Goal.write_debugdata() takes a directory parameter

Analogous to core changes.

Package: string attributes are represented by Unicode object

Attributes baseurl, location, sourcerpm, version, release, name, arch, description, evr, license, packager, reponame, summary and url of Package object return Unicode string.

Changes in 0.4.18

Core

Deprecated hy_advisory_get_filenames

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.

Python bindings

Repo() does not accept cost keyword argument

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.

Deprecated _hawkey.Advisory.filenames

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.

Changes in 0.4.19

Python bindings

Advisory attributes in Unicode

All string attributes of Advisory and AdvisoryRef objects (except the deprecated filenames attribute) are Unicode objects now.

Changes in 0.5.2

Core

hy_chksum_str returns NULL

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.

Changes in 0.5.3

Core

New parameter logfile to hy_sack_create()

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);

Deprecated hy_create_cmdline_repo()

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.

Python bindings

New optional parameter logfile to Sack constructor

This addition lets user specify log file path from Sack.__init__()

cache_path property of Sack renamed to cache_dir

This change was already announced but it actually never happened.

Deprecated Sack method create_cmdline_repo()

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.

Changes in 0.5.4

Python bindings

Goal: install() takes a new optional parameter

If the optional parameter is set to True, hawkey silently skips packages that can not be installed.

Changes in 0.5.5

Core

Renamed hy_sack_load_yum_repo to hy_sack_load_repo

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.

Python bindings

Sack method load_yum_repo has been renamed to Sack.load_repo()

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.

Changes in 0.5.7

Python bindings

Package: file attribute is represented by list of Unicode objects

Sack: list_arches method returns list of Unicode objects

Changes in 0.5.9

Core

Deprecated hy_goal_req_has_distupgrade(), hy_goal_req_has_erase() and hy_goal_req_has_upgrade() functions

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

Python bindings

Deprecated Goal methods Goal.req_has_distupgrade_all(), Goal.req_has_erase() and Goal.req_has_upgrade_all()

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

Changes in 0.6.2

Core

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().

Python bindings

Aside from the one change below, the Python bindings should be unaffected by the C API changes.

Advisory: The filename property is removed along with the C API

Contents

Getting Started

How do I build it?

See the README.

Are there examples using hawkey?

Yes, look at:

Using Hawkey

How do I obtain the repo metadata files to feed to Hawkey?

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.

Why is a tool to do the downloads not integrated into Hawkey?

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.

Contents

IMPORTANT:

Setup

First of, make sure hawkey is installed on your system, this should work from your terminal:

>>> import hawkey

The Sack Object

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.

Loading RPMDB

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.

Loading Repositories

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).

Case for Loading the Filelists

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:

Building and Reusing the Repo Cache

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.

Queries

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:

Resolving things with Goals

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.

Selector Installs

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)

Contents

Introduction

This reference manual describes Python API to the library. For a quick start take a look at python-hawkey Tutorial. To be sure that you are familiar with our deprecation policy, see API Changes.

NOTE:

WARNING:

Contents

API Documentation Contents

Sack---The fundamental hawkey structure

Error handling

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

Module level constants

Repositories

Indices:

Selectors are not Queries

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