Elektra  0.9.5
How-To: Write a Plugin

This file serves as a tutorial on how to write a storage plugin (which includes all information to write filter plugins).

First, there are a few basic points to understand about Elektra plugins. This first section will explain the basic layout of a plugin and what various methods exists within one.

All plugins use the same basic interface. This interface consists of five basic functions:

. The developer replaces Plugin with the name of their plugin. So in the case of the line plugin, the names of these functions would be elektraLineOpen(), elektraLineGet(), elektraLineSet(), elektraLineError(), and elektraLineClose(). Additionally, there is one more function called ELEKTRA_PLUGIN_EXPORT, where once again Plugin should be replaced with the name of the plugin, this time in uppercase. So for the line plugin this function would be ELEKTRA_PLUGIN_EXPORT(line). The developer may also define elektraPluginCheckConf() if configuration validation at mount-time is desired.

The KDB relies on the first five functions for interacting with configuration files stored in the key database. Calls to kdbGet() and kdbClose() will call the functions elektraPluginGet() and elektraPluginClose() respectively for the plugin that was used to mount the configuration data. kdbSet() calls elektraPluginSet() but also elektraPluginError() when an error occurs. elektraPluginOpen() is called before the first call to elektraPluginGet() or elektraPluginSet(). These functions serve different purposes that allow the plugin to work:

Most simply put: most plugins consist of five major functions, elektraPluginOpen(), elektraPluginClose(), elektraPluginGet(), elektraPluginSet(), and ELEKTRA_EXPORT_PLUGIN.

Because remembering all these functions can be cumbersome, we provide a skeleton plugin in order to easily create a new plugin. The skeleton plugin is called `template` and a new plugin can be created by calling the copy-template script . For example, the author of the line plugin used the command scripts/dev/copy-template line to create the initial version of the plugin. Afterwards two important things are left to be done:

After these two steps your plugin is ready to be compiled, installed and mounted for the first time. Have a look at How-To: kdb mount

If you want to use C++ instead of C for plugin development you can use `copy-template` to create a plugin based on `cpptemplate`. For example, to create a new plugin called pluginbaby use the command:

scripts/dev/copy-template -p pluginbaby

In Elektra, multiple plugins form a backend. If every plugin would do whatever it likes to do, there would be chaos and backends would be unpredictable.

To avoid this situation, plugins export a so called contract. In this contract the plugin states how nicely it will behave and what other plugins can depend on.

Because the contracts also contain information for humans, these parts are written in a README.md files of the plugins. To make the contracts machine-readable, the following CMake command exists:

generate_readme(pluginname)

It will generate a readme_plugginname.c (in the build-directory) out of the README.md of the plugin’s source directory.

But prefer to use

add_plugin(pluginname)

where the generation of the readme (among many other things) are already done for you. More details about how to write the CMakeLists.txt will be discussed later in the tutorial.

The README.md will be used by:

The first lines must look like:

- infos = Information about YAJL plugin is in keys below
- infos/author = Markus Raab <elektra@libelektra.org>
- infos/licence = BSD
- infos/provides = storage/json
- infos/needs = directoryvalue
- infos/recommends = rebase comment type
- infos/placements = getstorage setstorage
- infos/status = maintained coverage unittest
- infos/description = JSON using YAJL

Every of these line represents a clause of the contract. All these clauses need to be present for every plugin.

The information of clauses are limited to a single line, starting with - (so that the file renders nicely in Markdown), followed by the clause itself separated by =. Only for the description an unlimited amount of lines can be used (until the end of the file).

For the meaning (semantics) of these clauses, please refer to contract specification. The only difference for filter plugins is that their infos/provides and infos/placements are different.

The already mentioned generate_readme will produce a list of Keys using the information in README.md. It would look like (for the third key):

keyNew ("system:/elektra/modules/yajl/infos/licence",
KEY_VALUE, "BSD", KEY_END);

In your plugin, specifically in your elektraPluginGet() implementation, you have to return the contract whenever configuration below system:/elektra/modules/plugin is requested:

if (!strcmp (keyName(parentKey), "system:/elektra/modules/plugin"))
{
KeySet *moduleConf = elektraPluginContract();
ksAppend(returned, moduleConf);
ksDel(moduleConf);
return 1;
}

The elektraPluginContract() is a method implemented by the plugin developer containing the parts of the contract not specified in README.md. An example of this function (taken from the `yajl` plugin):

static inline KeySet *elektraYajlContract()
{
return ksNew (30,
keyNew ("system:/elektra/modules/yajl",
KEY_VALUE, "yajl plugin waits for your orders", KEY_END),
keyNew ("system:/elektra/modules/yajl/exports", KEY_END),
keyNew ("system:/elektra/modules/yajl/exports/get",
KEY_FUNC, elektraYajlGet,
keyNew ("system:/elektra/modules/yajl/exports/set",
KEY_FUNC, elektraYajlSet,
#include "readme_yourplugin.c"
keyNew ("system:/elektra/modules/yajl/infos/version",
KEY_VALUE, PLUGINVERSION, KEY_END),
keyNew ("system:/elektra/modules/yajl/config", KEY_END),
keyNew ("system:/elektra/modules/yajl/config/",
KEY_VALUE, "system",
keyNew ("system:/elektra/modules/yajl/config/below",
KEY_VALUE, "user",
}

It basically only contains the symbols to be exported (these symbols depend on the functions the plugin provides) and the plugin version information that is always defined by the macro PLUGINVERSION.

As already said, readme_yourplugin.c is generated in the binary directory, so make sure that your CMakeLists.txt contains (prefer to use add_plugin where this is already done correctly):

include_directories (${CMAKE_CURRENT_BINARY_DIR})

For every plugin you have to write a CMakeLists.txt. If your plugin has no dependencies, you can skip this section. The full documentation of add_plugin is available here.

In order to understand how to write the CMakeLists.txt, you need to know that the same file is included multiple times for different reasons.

  1. The first time, only the name of plugins and directories are enquired. In this phase, only the add_plugin should be executed.
  2. The second time (if the plugin is actually requested), the CMakeLists.txt is used to detect if all dependencies are actually available.

This means that in the first time, only the add_plugin should be executed and in the second time the detection code together with add_plugin.

So that you can distinguish the first and second phase, the variable DEPENDENCY_PHASE is set to ON iff you should search for all needed CMake packages. You should avoid to search for packages otherwise, because this would:

So usually you would have:

if (DEPENDENCY_PHASE)
find_package (MyLib QUIET)
if (MYLIB_FOUND)
# add testdata, test cases...
else ()
remove_plugin (myplugin "mylib not found")
endif ()
endif ()

So if you are in the second phase (DEPENDENCY_PHASE), you will search for all dependencies, in this case MyLib. If all dependencies are satisfied, you add everything needed for the plugin, except the plugin itself. This happens after endif ():

add_plugin (myplugin
SOURCES
...
LINK_LIBRARIES
${LIBXML2_LIBRARIES}
DEPENDENCIES
${LIBXML2_FOUND}
)

Important is that you pass the information which packages are found as boolean. The plugin will actually be added iff all of the DEPENDENCIES are true.

Note that no code should be outside of if (DEPENDENCY_PHASE). It would be executed twice otherwise. The only exception is add_plugin which must be called twice to successfully add a plugin.

Please note that the parameters passed to add_plugin need to be constant between all invocations. Some find_package cache their variables, others do not, which might lead to toggling variables. To avoid problems, create a variable containing all LINK_LIBRARIES or DEPENDENCIES within DEPENDENCY_PHASE.

If your plugin makes use of compilation variants you should also read the information there.

This section will focus on an overview of the kind of code you would use to develop a plugin. It gives examples from real plugins and should serve as a rough guide on how to write a storage plugin that can read and write configuration data into an Elektra KeySet.

elektraPluginGet is the function responsible for turning information from a file into a usable KeySet. This function usually differs pretty greatly between each plugin. This function should be of type int, it returns either 1 or on 0 on success.

Any other return value indicates an error (ELEKTRA_PLUGIN_STATUS_ERROR). The function will take in a Key, usually called parentKey which contains a string containing the path to the file that is mounted. For instance, if you run the command kdb mount /etc/linetest system:/linetest line then keyString(parentKey) should be equal to /etc/linetest. At this point, you generally want to open the file so you can begin saving it into keys. Here is the trickier part to explain. Basically, at this point you will want to iterate through the file and create keys and store string values inside of them according to what your plugin is supposed to do. I will give a few examples of different plugins to better explain.

The line plugin was written to read files into a KeySet line by line using the newline character as a delimiter and naming the keys by their line number such as #1, #2, .. #_22 for a file with 22 lines. So once I open the file given by parentKey, every time as I read a line I create a new key, let's call it new_key using dupKey(parentKey). Then I set new_key's name to lineNN (where NN is the line number) using keyAddBaseName and store the string value of the line into the key using keySetString. Once the key is initialized, I append it to the KeySet that was passed into the elektraPluginGet function, let's call it returned for now, using ksAppendKey(returned, new_key). Now the KeySet will contain new_key with the name #N properly saved where it should be according to the kdb mount command (in this case, system:/linetest/#N), and a string value equal to the contents of that line in the file. The line plugin repeats these steps as long as it hasn't reached end of file, thus saving the whole file into a KeySet line by line.

The simpleini plugin works similarly, but it parses for ini files instead of just line-by-line. At their most simple level, ini files are in the format of name=value with each pair taking one line. So for this plugin, it makes a lot of sense to name each Key in the KeySet by the string to the left of the = sign and store the value into each key as a string. For instance, the name of the key would be name and keyGetString(name) would return value.

As you may have noticed, simpleini and line plugins work very similarly. However, they just parse the files differently. The simpleini plugin parses the file in a way that is more natural to ini file (setting the key's name to the left side of the equals sign and the value to the right side of the equals sign). The elektraPluginGet function is the heart of a storage plugin, it’s what allows Elektra to store configurations in its database. This function isn't just run when a file is first mounted, but whenever a file gets updated, this function is run to update the Elektra Key Database to match.

We also give a brief overview of the elektraPluginSet function. This function is basically the opposite of elektraPluginGet. Where elektraPluginGet reads information from a file into the Elektra Key Database, elektraPluginSet writes information from the database back into the mounted file.

First have a look at the signature of elektraLineSet:

int elektraLineSet(Plugin *handle ELEKTRA_UNUSED, KeySet *toWrite, Key *parentKey);

Lets start with the most important parameters, the KeySet and the parentKey. The KeySet supplied is the KeySet that is going to be persisted in the file. In our case it would contain the Keys representing the lines. The parentKey is the topmost Key of the KeySet and serves several purposes. First, it contains the filename of the destination file as its value. Second, errors and warnings can be emitted via the parentKey. We will discuss error handling in more detail later. The Plugin handle can be used to persist state information in a thread-safe way with elektraPluginSetData. As our plugin is not stateful and therefore does not use the handle, it is marked as unused in order to suppress compiler warnings.

Basically the implementation of elektraLineSet can be described with the following pseudocode:

// open the file
if (error)
{
ELEKTRA_SET_RESOURCE_ERROR(parentKey, keyString(parentKey));
}
for (/* each key */)
{
// write the key value together with a newline
}
// close the file

The full-blown code can be found at line plugin.

As you can see, all elektraLineSet does is open a file, take each Key from the KeySet (remember they are named #1, #2 ... #_22) in order, and write each key as its own line in the file. Since we don't care about the name of the Key in this case (other than for order), we just write the value of keyString for each Key as a new line in the file. That's it. Now, each time the mounted KeySet is modified, elektraPluginSet will be called and the mounted file will be updated.

We haven't discussed ELEKTRA_SET_<CONCRETE_TYPE>_ERROR yet. Because Elektra is a library, printing errors to stderr wouldn't be a good idea. Instead, errors and warnings can be appended to a key in the form of metadata. This is what ELEKTRA_SET_<CONCRETE_TYPE>_ERROR does. The <CONCRETE_TYPE> in the text means the concrete error type such as RESOURCE, INSTALLATION, etc. There are also abstract error types which are not instantiable. You can read more about concrete and abstract error types in the error-categorization.md guideline. Note that you also have a varargs macro with ...ERRORF that allows you to insert a string and substitute parts with variables. You can see all available error types as well as their categorization guidelines here. Because the parentKey always exists even if a critical error occurs, we write the error to the parentKey. The error does not necessarily have to be in a configuration. If there are multiple errors in a configuration, only the first occurrence will be written to the metadata of the parentKey.

The second parameter can be used to provide additional information about the error. In our case we simply supply the filename of the file that caused the error. The kdb tools will interpret this error and print it in a pretty way. Notice that this can be used in any plugin function where the parentKey is available.

The elektraPluginOpen and elektraPluginClose functions are not commonly used for storage plugins, but they can be useful and are worth reviewing. elektraPluginOpen function runs before elektraPluginGet and is useful to do initialization if necessary for the plugin. On the other hand elektraPluginClose is run after other functions of the plugin and can be useful for freeing up resources.

The elektraPluginCheckConf function may be used for validation of the plugin configuration during mount-time. The signature of the function is:

int elektraLineCheckConf (Key * errorKey, KeySet * conf);

The configuration of the plugin is provided as conf. The function may report an error or warnings using the errorKey and the return value.

The following convention was established for the return value of elektraPluginCheckConf:

The following example demonstrates how to limit the length of the values within the plugin configuration to 3 characters.

int elektraLineCheckConf (Key * errorKey, KeySet * conf)
{
Key * cur;
ksRewind (conf);
while ((cur = ksNext (conf)) != 0)
{
const char * value = keyString (cur);
if (strlen (value) > 3)
{
ELEKTRA_SET_VALIDATION_SYNTACTIC_ERRORF ( errorKey,
"Value '%s' is more than 3 characters long",
value);
return -1; // The configuration was not OK and could not be fixed
}
}
return 0; // The configuration was OK and has not been changed
}

The elektraPluginCheckConf function is exported via the plugin's contract. The following example demonstrates how to export the checkconf function (see section Contract for further details):

keyNew ("system:/elektra/modules/" ELEKTRA_PLUGIN_NAME "/exports/checkconf", KEY_FUNC, elektraLineCheckConf, KEY_END);

Within the checkconf function all of the plugin configuration values should be validated. Errors should be reported via Elektra's error handling mechanism (see section ELEKTRA*SET*_ERROR for further details). If checkconf encounters a configuration value, that is not strictly invalid but can not be parsed by the plugin (e.g. a parameter which is not part of the plugin configuration), then a warning should be appended to errorKey, using ELEKTRA_ADD_<CONCRETE_TYPE>_WARNING. You also have a ...WARNINGF vararg macro that allows you to substitute parts of the message with variables.

A function that is always needed in a plugin, is ELEKTRA_PLUGIN_EXPORT. This functions is responsible for letting Elektra know that the plugin exists and which methods it implements. The code from the line plugin is a good example and pretty self-explanatory:

Plugin *ELEKTRA_PLUGIN_EXPORT
{
return elektraPluginExport("line",
ELEKTRA_PLUGIN_GET, &elektraLineGet,
ELEKTRA_PLUGIN_SET, &elektraLineSet,
}

For further information see the API documentation.

In order to enable communication between plugins which is more complex than what can be done with metadata, Elektra provides a global keyset which plugins can read from and modify.

The keyset is initialized and closed by a KDB handle and can be accessed by all plugins of a single handle except for plugins created manually (e.g. with elektraPluginOpen). It is not shared between different KDB handles.

It can be accessed by calling the elektraPluginGetGlobalKeySet function, which returns a handle to the global keyset.

Plugins using the global keyset are responsible for cleaning up the parts of the keyset they no longer need.

To make sure there is no collision between plugins, each plugin should use a unique prefix for its keys, e.g. system:/elektra/<plugin> for <plugin>.

To improve performance, the cache plugin also caches parts of the global keyset. If your plugin uses non-cacheable data, you don't have to do anything special. However, if you want your plugin's keys to be cached you should put them below system:/elektra/cached (to avoid collisions, use e.g. system:/elektra/cached/<plugin> for <plugin>). The cache plugin also caches keys below system:/elektra/cache, but those are reserved for use by the plugin itself.

Some applications want to call Elektra methods directly via native access. A KeySet is a data structure over which functions can iterate. If you want to start again from to first element, you have to explicitly call rewind() to set the internal pointer to the start. Any plugin expects the passed KeySet to be rewinded.