From d04baabafa2ebbad92741d1f87e6ff32999f894a Mon Sep 17 00:00:00 2001 From: Judd Vinet Date: Tue, 15 Mar 2005 01:51:43 +0000 Subject: Initial revision --- README | 288 +++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 288 insertions(+) create mode 100644 README (limited to 'README') diff --git a/README b/README new file mode 100644 index 00000000..db345460 --- /dev/null +++ b/README @@ -0,0 +1,288 @@ +========================================================================== +README: + +Overview and internals of the ALPM library and the PACMAN frontend. + +This document describes the state of the implementation before its CVS +import. +At this stage, the code is in pre-alpha state, but the design should not +change that much. +There's still need for some work to get the current code properly working. +The tag "ORE" was added in various places in the code, each time a point +remains unclear or is not yet implemented. + +========================================================================== + + +ALPM library overview & internals +================================= + +Here is a list of the main objects and files from the ALPM (i.e. Arch +Linux Package Management) library. +This document, whilst not exhaustive, also indicates some limitations +(on purpose, or sometimes due to its poor design) of the library at the +present time. + +Note: there is one special file ("alpm.h") which is the public interface +that should be distributed and installed on systems with the library. +Only structures, data and functions declared within this file are made +available to the frontend. +Lots of structures are of an opaque type and their fields are only +accessible in read-only mode, through some clearly defined functions. + +Note: several structures and functions have been renamed compared to +pacman 2.9 code. +This was done at first for the sake of naming scheme consistency, and +then primarily because of potential namespace conflicts between library +and frontend spaces. +Indeed, it is not possible to have two different functions with the same +name declared in both spaces. +To avoid such conflicts, some function names have been prepended with +"_alpm_". +In a general manner, public library functions are named +"alpm__" (examples: alpm_trans_commit(), +alpm_lib_release(), alpm_pkg_getinfo(), ...). +Internal (and thus private) functions should be named "_alpm_XXX" for +instance (examples: _alpm_needbackup(), _alpm_runscriplet(), ...). +As of now, this scheme is only applied to most sensitive functions +(mainly the ones from util.c), which have generic names, and thus, which +are likely to be redefined in the frontend. +One can consider that the frontend should have the priority in function +names choice, and that it is up to the library to hide its symbols to +avoid conflicts with the frontend ones. +Finally, functions defined and used inside a single file should be +defined as "static". + + +[HANDLE] (see handle.c) + +The "handle" object is the heart of the library. It is a global +structure available from almost all other objects (althought some very +low level objects should not be aware of the handle object, like chained +list, package or groups structures. + +There is only one instance, created by the frontend upon +"alpm_lib_init()" call, and destroyed upon "alpm_lib_release()" call. + +alpm_lib_init() is used to initialize library internals and to create +the handle object (handle != NULL). +Before its call, the library can't be used. +alpm_lib_release() just does the opposite (memory used by the library is +freed, and handle is set to NULL). +After its call, the library is no more available. + +The aim of the handle is to provide a central placeholder for essential +library parameters (filesystem root, pointers to database objects, +configuration parameters, ...) + +The handle also allows to register a log callback usable by the frontend +to catch all sort of notifications from the library. +The frontend can choose the level of verbosity (i.e. the mask), or can +simply choose to not use the log callback. +A friendly frontend should care at least for WARNING and ERROR +notifications. +Other notifications can safely be ignored and are mainly available for +troubleshooting purpose. + +Last, but not least, the handle holds a _unique_ transaction object. + + +[TRANSACTION] (see trans.c, and also alpm.c) + +The transaction sturcture permits easy manipulations of several package +at a time (i.e. adding, upgrade and removal operations). + +A transaction can be initiatied with a type (ADD, UPGRADE or REMOVE), +and some flags (NODEPS, FORCE, CASCADE, ...). + +Note: there can only be one type at a time: a transaction is either +created to add packages to the system, or either created to remove packages. +The frontend can't request for mixed operations: it has to run several +transactions, one at a time, in such a case. + +The flags allow to tweak the library behaviour during its resolution. +Note, that some options of the handle can also modify the behavior of a +transaction (NOUPGRADE, IGNOREPKG, ...). + +Note: once a transaction has been initiated, it is not possible anymore +to modify its type or its flags. + +One can also add some targets to a transaction (alpm_trans_addtarget()). +These targets represent the list of packages to be handled. + +Then, a transaction needs to be prepared (alpm_trans_prepare()). It +means that the various targets added, will be inspected and challenged +against the set of alreayd installed packages (dependency checkings, + +Last, a callback is associated with each transaction. During the +transaction resolution, each time a new step is started or done (i.e +dependency or conflict checkings, package adding or removal, ...), the +callback is called, allowing the frontend to be aware of the progress of +the resolution. Can be useful to implement a progress bar. + + +[CONFIGURATION/OPTIONS] (see handle.c) + +The library does not use any configuration file. The handle holds a +number of configuration options instead (IGNOREPKG, SYSLOG usage, +log file name, registered databases, ...). +It is up to the frontend to set the options of the library. +Options can be manipulated using calls to +alpm_set_option()/alpm_get_option(). + +Note: the file system root is a special option which can only be defined +when calling alpm_lib_init(). It can't be modified afterwards. + + +[CACHE] (see cache.c) + +Compared to pacman 2.9, there is now one cache object connected to each +database object. +There are both a package and a group cache. +The cache is loaded only on demand (i.e the cache is loaded the first +time data from it should used). + +Note: the cache of a database is always destroyed by the library after +an operation changing the database content (adding and/or removal of +packages). +Beware frontends ;) + + +[PACKAGE] (see package.c, and also db.c) + +The package structure is using three new fields, namely: origin, data, +infolevel. +The purpose of these fields is to know some extra info about data stored +in package structures. + +For instance, where is the package coming from (i.e origin)? +Was it loaded from a file or loaded from the cache? +If it's coming from a file, then the field data holds the full path and +name of the file, and infolevel is set to the highest possible value +(all package fields are reputed to be known). +Otherwise, if the package comes from a database, data is a pointer to +the database structure hosting the package, and infolevel is set +according to the db_read() infolevel parameter (it is possible using +db_read() to only read a part of the package datas). + +Indeed, to reduce database access, all packages data requested by the +frontend are comming from the cache. As a consequence, the library needs +to know exactly the level of information about packages it holds, and +then decide if more data needs to be fetched from the database. + +In file alpm.c, have a look at alpm_pkg_getinfo() function to get an +overview. + + +[ERRORS] (error.c) + +The library provides a global variable pm_errno. +It aims at being to the library what errno is for C system calls. + +Almost all public library functions are returning an integer value: 0 +indicating success, whereas -1 would indicate a failure. +If -1 is returned, the variable pm_errno is set to a meaningful value +(not always yet, but it should improve ;). +Wise frontends should always care for these returned values. + +Note: the helper function alpm_strerror() can also be used to translate +the error code into a more friendly sentence. + + +[LIST] (see list.c, and especially list wrappers in alpm.c) + +It is a double chained list structure, use only for the internal needs +of the library. +A frontend should be free to use its own data structures to manipulate +packages. +For instance, consider a graphical frontend using the gtk toolkit (and +as a consequence the glib library). The frontend will make use of the +glib chained lists or trees. +As a consequence, the library only provides a simple and very small +interface to retrieve pointers to its internal data (see functions +alpm_list_first(), alpm_list_next() and alpm_list_getdata()), giving to +the frontend the responsibility to copy and store the data retrieved +from the library in its own data structures. + + +PACMAN frontend overview & internals +==================================== + +Here are some words about the frontend responsibilities. +The library can operate only a small set of well defined operations and +dumy operations. + +High level features are left to the frontend ;) + +For instance, during a sysupgrade, the library returns the whole list of +packages to be upgraded, without any care for its content. +The frontend can inspect the list and perhaps notice that "pacman" +itself has to be upgraded. In such a case, the frontend can choose to +perform a special action. + + +[MAIN] (see pacman.c) + +Calls for alpm_lib_init(), and alpm_lib_release(). +Read the configuration file, and parse command line arguments. +Based on the action requested, it initiates the appropriate transactions +(see pacman_add(), pacman_remove(), pacman_sync() in files add.c, +remove.c and sync.c). + + +[CONFIGURATION] (see conf.c) + +The frontend is using a configuration file, usually "/etc/pacman.conf". +Part of these options are only usefull for the frontend only (mainly, +the download stuffs, and some options like HOLDPKG). +The rest is used to configure the library. + + +[ADD/UPGRADE/REMOVE/SYNC] + +Nothing new here, excepted some reorganization. + +The file pacman.c has been divided into several smaller files, namely +add.c, remove.c, sync.c and query.c, to hold the big parts: pacman_add, +pacman_remove, pacman_sync. +These 3 functions have been splitted too to ease the code reading. + + +[DONWLOAD] (see download.c) + +The library is not providing download facilities. As a consequence, it +is up the the frontend to retrieve packages from Arch Linux servers. +To do so, pacman is linked against an improved version of libftp +supporting both http and ftp donwloads. +As a consequence, the frontend is repsonsible for the directory +/var/cache/pacman/pkgs. +One can consider that this cache is a facility provided by pacman. + +Note: other frontends have to download packages by themselves too, +although the cache directory can be shared by several frontends. + + +[LIST] (see list.c) + +Single chained list. +A minimalistic chained list implementation to store options from the +configuration file, and targets passed to pacman on the command line. + + +LIMITATIONS/BEHAVIOR CHANGES COMPARED TO PACMAN 2.9 +=================================================== + +Excepted missing features still needing to be implemented, one can +notice the following limitations: + +- When trying to add a package that conflicts with an already installed +one, pacman won't ask for removing the latter one prior to install the +former. +It will stop with an error code mentionning a conflict. + +The library can handle only one transaction at a time, and as a consequence, +it is not easily possible to remove a conflicting package while holding +still the on-going transaction... + +- ... + -- cgit v1.2.3-24-g4f1b