Skip to content

Latest commit

 

History

History
555 lines (417 loc) · 26.2 KB

command-line-options.md

File metadata and controls

555 lines (417 loc) · 26.2 KB
Raw

Command-line Options

Introduction

Many applications use command-line options and environment variables as a way to override configuration values. In Elektra this can be automated by providing a specification that maps command-line options and environment variables to keys in the KDB.

The recommended way to do this is via the gopts plugin. This plugin internally calls the actual parser elektraGetOpts. However, since there are some downsides to calling the parser manually, we cannot generally recommend doing so. If you think you have a use case for calling elektraGetOpts directly, take a look at the section Advanced Use: Calling elektraGetOpts directly below.

The parser uses a specification together with argc/argv and a list of environment variables to create keys in the proc:/ namespace. Because the keys are in the proc:/ namespace, they will be preferred over all other namespaces in a standard cascading lookup. This allows us to use command-line options or environment variables to override standard configuration keys.

Setup

While you could manually mount and configure gopts, it is not recommended doing so. Instead, you should use elektraGOptsContract to create a contract for use with kdbOpen(). This contract ensures that gopts is automatically mounted and correctly configured. To use elektraGOptsContract, include kdbgopts.h. This gives you access to these two functions:

int elektraGOptsContract (KeySet * contract, int argc, const char * const * argv, const char * const * envp, const Key * parentKey, KeySet * goptsConfig);

int elektraGOptsContractFromStrings (KeySet * contract, size_t argsSize, const char * args, size_t envSize, const char * env, const Key * parentKey, KeySet * goptsConfig);

Whenever possible, we recommend that you use elektraGOptsContract since it has less memory and processing overhead than elektraGOptsContractFromStrings. However, to use elektraGOptsContract the pointers given for argv and envp must remain valid until after calling kdbClose(), because the gopts plugin will directly use these pointers. In the standard use case (i.e. using the argv from main() and the global environ for envp), this restriction is not a problem.

An example for using elektraGOptsContract could look like this:

extern char ** environ;

int main (int argc, char ** argv)
{
  Key * parentKey = keyNew ("/sw/org/example/#0/current", KEY_END);

  KeySet * contract = ksNew (0, KS_END);
  KeySet * goptsConfig = ksNew (0, KS_END);

  // error handling omitted for brevity
  elektraGOptsContract (contract, argc, argv, environ, parentKey, goptsConfig);

  KDB * kdb = kdbOpen (contract, parentKey);
}

If you cannot provide pointers that meet the requirements, you may use elektraGOptsContractFromStrings. This function copies its arguments args and env into separate memory, so the pointers need only be valid for the duration of the function call. This is mainly useful for language bindings, since in many programming languages manual memory management is not possible and there is no (easy) way to ensure the argv and envp pointers meet the necessary requirements.

The gopts plugin can also use operating system specific functions to retrieve the command-line arguments and environment variables internally. We recommend that you do not rely on this behavior whenever possible, since it can be a bit flaky (especially for command-line arguments). However, if for example you are writing a library or for some other reason do not have access to the necessary data, you can use this fallback.

If you pass argc=0 and argv=NULL to elektraGOptsContract or argsSize=0 and args=NULL to elektraGOptsContractFromStrings, gopts will fallback to the internal lookup of command-line options. Similarly, if you pass envp=NULL to elektraGOptsContract or envSize=0 and env=NULL to elektraGOptsContractFromStrings, environment variables will be retrieved internally.

The other parameters are the same for both functions. The contract will be written into the contract KeySet. The parentKey indicates where to find the specification. The actual namespace that parentKey uses is irrelevant (we recommend a cascading key, so that it can be re-used for kdbGet()). The parser will use the keys below the spec:/ namespace key equivalent to parentKey as the specification, and it will write keys to the equivalent key in proc:/.

The last parameter goptsConfig can be used to provide additional configuration values to gopts. For example, this can be used to configure the auto-generated help message. The keys that gopts accepts in this KeySet will be explained throughout the document . A full list can be found in the gopts README.

Specification

This section describes the specification used by the command-line option and environment variable parser.

Options

To define a command-line option either set the opt metakey to the short option you want to use, or set opt/long to the long option you want to use. For short options, only the first character of the given value will be used ('\0' is ignored). Short and long options can be used simultaneously.

Additionally, a key can also be associated with multiple short/long options. To achieve this treat opt as an array. For example for two options -a and -b you would set opt=#1, opt/#0=a and opt/#1=b. If not explicitly stated otherwise, you can replace opt with any opt/# array element in all meta-keys mentioned in this document. This of course includes long options (i.e. opt/#0/long, etc.).

While you can specify multiple options (or environment variables, see below) for a single key, only one of them can be used at the same time. Using two or more options (or variables) that are all linked to the same key, will result in an error.

Option Arguments

Per default an option is expected to have an argument. Arguments to short and long options are given in the same way as with getopt_long(3) (i.e. -oarg, -o arg, --option arg or --option=arg).

To change whether an option expects an argument set opt/arg to either "none" or "optional" (the default is "required").

  • If you choose "none", the corresponding key will be set to "1", if the option is used. This value can be changed by setting opt/flagvalue.
  • An option that is set to "optional" is treated the same as with "none", except that you can also set the value with the long option form --option=value. This also means that opt/flagvalue is used, if no argument is given. Contrary to getopt_long(3) options with optional arguments can still have short forms. They just cannot have an argument in this form.

Environment Variables

Elektra also supports parsing environment variables in a similar manner. For these there are however, less configuration options. You can simply specify one or more environment variables for a key using the env metakey (or env/# meta-array for multiple).

Arrays

Both, options and environment variables, expose special behavior, if used in combination with arrays.

If an option is specified on a key with basename #, the option can be used repeatedly. All occurrences will be collected into the array.

Environment variables obviously cannot be repeated, instead a behavior similar to that used for PATH is adopted. On Windows the variable will be split at each ';' character. On all other systems ':' is used as a separator.

Parameter Arguments

All unused elements of argv are collected into an array. You can access this array by specifying args=remaining on a key with basename #. The array will be copied into this key. As is the case with getopt(3) processing of options will stop, if -- is encountered in argv.

If we parse command-line options like the POSIX version of getopt(3) does, then we would also stop processing options at the first non-option argument. This is not the case by default, but we can enable this behavior. To do so, you need to pass a Key /posixly with value 1 in the goptsConfig KeySet of the gopts contract.

Additionally, there is args=indexed. If it is specified on a key, the key must also have the metakey args/index=N set to an integer N. Such a key will be set to the unused element at index N. If a key has args=indexed and args/index=N, then there must also be keys for all integers 0 <= X < N with args=indexed and args/index=N set. For example, you cannot use args/index=0 and args/index=2 without args/index=1.

Combining args=indexed and args=remaining in the same specification (on different keys) is also possible. The key with args=remaining will only contain those elements not used via args=indexed. For example, if there are keys with args/index=0 and args/index=1 then the args=remaining array will start with the third (index 2) parameter argument. Note however, the args=remaining array always starts with index #0, even if it doesn't contain the first parameter argument.

Example

[from]
args = indexed
args/index = 0

[to]
args = indexed
args/index = 1

[more/#]
args = remaining

If an application app with the specification above is called as ./app apple banana cherry date, then the keys will be assigned as follows:

  • from = apple
  • to = banana
  • more/#0 = cherry
  • more/#1 = date

Sub-Commands

The parser also supports sub-commands. Explaining sub-commands is easiest through the help of an example: add and commit are both sub-commands of git, since we can call git add and git commit and they do entirely different things. The most important impact of using sub-commands is their effect on option arguments. For example calling git -p add and git add -p result in different behavior, since the -p option is interpreted differently. The options that git understands are separate from the options that its sub-command add knows. However, the option -p is understood by both. In git it is short for --paginate and in add it is short for --patch.

An important thing to know about sub-commands is that they automatically turn on POSIX mode. This means all options for a specific sub-command must be given before any non-option arguments (such as parameters or sub-commands). Otherwise, we couldn't distinguish between git -p add and git add -p. In other words, an option argument is always assigned to the first sub-command to its left. Any element of argv that is not an argument for a sub-command, either switches to a new sub-command or is the start of the parameter arguments.

A sub-command is created by specifying command on a key. To enable sub-command processing the parent key of the whole specification must have command set to an empty string. All keys marked with command directly below another key K marked with command (e.g. the parent key) are sub-commands of K. It is an error, if the immediate parent of a key X marked with command is not marked with command and X is not the parent of the whole specification.

To inform the application about the invoked sub-commands, the parser sets each command key to one of two values:

  • The basename of the key, whose command was invoked.
  • An empty string otherwise.

For example consider ./app add more: The parent key will be set to the basename of whatever key command=add was specified on, the key for add will be set to the basename corresponding to more and the key for more is set to an empty string, because none of its sub-commands were invoked. A more detailed example is shown below.

Every key considered by the parser is assigned either to the root command, or a single sub-command. Specifically, each key is assigned to the command of its immediate parent. If sub-commands are used and the immediate parent of an opt or args key has no command metadata an error occurs. The value of a key will only be set, if the corresponding sub-command was invoked.

Lastly, it is allowed to have keys with args and command below the same parent. If a matching sub-command is found among the command keys, processing will continue there. Otherwise, the args keys will be considered. This allows an application to implement dynamic commands (like git or kdb) by using the args=remaining array to invoke another application.

If an unknown sub-command is encountered without an args key, an error is returned.

All of this is best understood with an example:

[kdb]
command = ""

[kdb/printversion]
description = "print version information and exit (ignoring all other options/commands/parameters)"
opt = v
opt/long = version
opt/arg = none

[kdb/getter]
description = "get a key's value"
command = get

[kdb/getter/verbose]
description = "print additional information about where the value comes from"
opt = v
opt/long = verbose
opt/arg = none

[kdb/getter/keyname]
description = "name of the key to read"
args = indexed
args/index = 0

[kdb/setter]
description = "set a key's value"
command = set

[kdb/setter/verbose]
description = "print additional information about where the value will be stored"
opt = v
opt/long = verbose
opt/arg = none

[kdb/setter/keyname]
description = "name of the key to write"
args = indexed
args/index = 0

[kdb/setter/value]
description = "value to be written"
args = indexed
args/index = 1

[kdb/dynamic/#]
description = "dynamically call a user-supplied command"
args = remaining
  • If we invoke kdb -v, keys below kdb/getter, kdb/setter and kdb/dynamic are not touched. The result is:
    • kdb = ""
    • kdb/printversion = 1
    • kdb/getter = ""
    • kdb/setter = ""
  • If we invoke kdb get -v name, keys below kdb/setter and kdb/dynamic are not touched. The result is:
    • kdb = getter
    • kdb/getter = ""
    • kdb/getter/verbose = 1
    • kdb/getter/keyname = name
    • kdb/setter = ""
  • If we invoke kdb -v set -v, keys below kdb/getter and kdb/dynamic are not touched. The result is:
    • kdb = setter
    • kdb/printversion = 1
    • kdb/setter = ""
    • kdb/setter/verbose = 1
    • kdb/getter = ""
  • If we invoke kdb -v custom -v -x z, keys below kdb/getter and kdb/setter are not touched. The result is:
    • kdb = ""
    • kdb/printversion = 1
    • kdb/getter = ""
    • kdb/setter = ""
    • kdb/dynamic/#0 = custom
    • kdb/dynamic/#1 = -v
    • kdb/dynamic/#2 = -x
    • kdb/dynamic/#3 = z

To determine what code to execute, an application would just start with the parent key kdb in the example above. It would then repeatedly look at the current key's value, append that to the current key and continue, until the current key's value was the empty string "". Each of the examined keys corresponds to one of the sub-commands in the invocation and the keys directly below those, contain the relevant options (and for the last sub-command also parameters).

The C code for this example is located in examples/optsCommands.c.

Precedence

The order of precedence is simple:

  • If a short option for a key is found, it will always be used.
  • If none of the short options for a key are found, we look for long options.
  • If neither short nor long options are found, environment variables are considered.

Limitations

  • Both options and environment variables can only be specified on a single key. If you need to have the value of one option/environment variable in multiple keys, you may use fallbacks.
  • - cannot be used as a short option, because it would collide with the "option end marker".
  • help cannot be used as a long option, because it would collide with the help option.

Help Message

When the help option --help is encountered in argv, the parser only reads the specification, but does not create any keys in the proc:/ namespace. It will however, generate a standard help message that you can print.

To find you, whether --help was encountered, check if the KeySet returned by kdbGet() contains the special key proc:/elektra/gopts/help with value 1. If it does, the auto-generated help message is stored in the key proc:/elektra/gopts/help/message.

int main (int argc, char ** argv)
{
  // setup ommitted for brevity
  KeySet * ks = ksNew (0, KS_END);

  // error handling ommitted for brevity
  kdbGet (kdb, ks, parentKey);

  Key * help = ksLookupByName (ks, "proc:/elektra/gopts/help");
  if (help != NULL && strcmp (keyString (help), "1") == 0)
  {
    printf ("%s\n", keyString (ksLookupByName (ks, "proc:/elektra/gopts/help/message")));
    // cleanup omitted for brevity
    return 0;
  }
}

Note: The key proc:/elektra/gopts/help will always be generated. Only if its value is set to 1, the --help option was encountered.

Structure of the Help Message

The help message consists of a usage line and an options list. The program name for the usage line is taken from argv[0]. If the value contains a slash (/) it will be considered a path and only the part after the last slash will be used.

The options list will contain exactly one entry for each key that has at least one option. Each entry has two parts. First all the options for the key are listed and then (possibly on the next line, if there are a lot of options), the description for the key is listed. The description is taken from the opt/help or alternatively the description metakey.

Note: opt/help is specified only once per key. That means even if the key uses opt/#0, opt/#1, etc. (unlike most other metadata) the description will always be taken from opt/help directly, because there can only be one description. In general, we recommend using description, because it is used by other parts of Elektra as well. opt/help is intended to provide a less verbose description more suitable for the command-line.

Sub-Commands

If sub-commands are in use, the generated help message will apply to the invoked sub-command only. For example ./app --help generates the general help message for ./app containing only options valid for the root command. But ./app more --help generates the help message for the sub-command more and contains options valid to this sub-command.

Modifying the Help Message

The standard help message can be modified in a few different ways:

  • The usage line can be replaced by a custom string (see below).
  • A custom string can be inserted between the usage line and the options list (see below).
  • An option can be hidden from the help message by setting opt/hidden to "1". This hides both the long and short form of the option. If you want to hide just one form, use an array of two options and hide just one index.
  • If the option has an "optional" or "required" argument, the string ARG will be used as a placeholder by default. You can change this, by setting opt/arg/help for the corresponding option.

Custom Usage Line

To use a custom usage line, you can either manually generate the help message or you can pass the key /help/usage in the goptsConfig KeySet. The value of /help/usage will be used to replace the default usage line.

Please note, that this will replace the whole usage line including the program name taken from argv[0].

Adding a Prefix Text

If you just want to add information above the options list, but not replace the whole usage line, you can do so by adding a prefix text. This can be done by manually generating the help message or by passing the key /help/prefix in the goptsConfig KeySet. The value of /help/prefix will be inserted between the usage line and the options list.

Examples

The following specification describes a command line interface similar to the one used by rm. (It is based on rm (GNU coreutils) 8.30).

[force]
opt = f
opt/long = force
opt/arg = none
description = ignore nonexistent files and arguments, never prompt

[interactive]
opt = #1
opt/#0 = i
opt/#0/long = interactive
opt/#0/arg = optional
opt/#0/flagvalue = always
opt/#0/arg/help = WHEN
opt/#1 = I
opt/#1/flagvalue = once
opt/#1/arg = none
description = prompt according to WHEN: never, once (-I), or always (-i); without WHEN, prompt always

[singlefs]
opt/long = one-file-system
opt/arg = none
description = when removing a hierarchy recursively, skip any directory that is on a file system different from that of the corresponding line argument

[nopreserve]
opt/long = no-preserve-root
opt/arg = none
description = do not treat '/' specially

[preserve]
opt/long = preserve-root
opt/arg = optional
opt/arg/help = all
opt/flagvalue = root
description = do not remove '/' (default); with 'all', reject any command line argument on a separate device from its parent

[recursive]
opt = #1
opt/#0 = r
opt/#0/long = recursive
opt/#0/arg = none
opt/#1 = R
opt/#1/arg = none
description = remove directories and their contents recursively

[emptydirs]
opt = d
opt/long = dir
opt/arg = none
description = remove empty directories

[verbose]
opt = v
opt/long = verbose
opt/arg = none
env = VERBOSE
description = explain what is being done

[showversion]
opt/long = version
opt/arg = none
description = output version information and exit

[files/#]
args = remaining
env = FILES
description = the files that shall be deleted

If this specification is used in a program called erm (for Elektra rm), which is called like this:

FILES="one.txt:other.log" VERBOSE=1 erm -fi --recursive

The following keys will be created (assuming the specification is mounted at spec:/sw/org/erm/#0/current):

  • proc:/sw/org/erm/#0/current/force = "1"
  • proc:/sw/org/erm/#0/current/interactive = "always"
  • proc:/sw/org/erm/#0/current/recursive = "1"
  • proc:/sw/org/erm/#0/current/verbose = "1"
  • proc:/sw/org/erm/#0/current/files [array] = "#1"
  • proc:/sw/org/erm/#0/current/files/#0 = "one.txt"
  • proc:/sw/org/erm/#0/current/files/#1 = "other.log"

Calling FILES="abcd.txt" erm 123.txt 456.txt meanwhile will result in:

  • proc:/sw/org/erm/#0/current/files [array] = "#1"
  • proc:/sw/org/erm/#0/current/files/#0 = "123.txt"
  • proc:/sw/org/erm/#0/current/files/#1 = "456.txt"

NOTE: proc:/sw/org/erm/#0/current/files [array] = "#1" means the array metadata of proc:/sw/org/erm/#0/current/files is #1.

You can find a full working example here. However, it uses a hard coded specification which is a bit harder to read.

Advanced Use: Calling elektraGetOpts directly

The actual command line parser is implemented in elektraGetOpts.

int elektraGetOpts (KeySet * ks, int argc, const char ** argv, const char ** envp, Key * parentKey);

To access this function, you first need to link your application against libelektra-opts and then include the header kdbopts.h.

Note: libelektra-opts is an internal library and kdbopts.h an internal header. We do not make and guarantees to the API stability of the functions declared in kdbopts.h. We will do our best to keep the API compatible, but the only way to have guaranteed API stability and backwards compatibility is to only use the parser via gopts.

Calling elektraGetOpts directly has some disadvantages. The main one being that there is no way to validate the values of command-line options. When you use gopts, you can for example add the metadata type=long to a key with a command-line option specification and the type plugin will validate that the value generated by the command-line option parser is actually of type long. But since elektraGetOpts has no way of delegating to plugins (as it is independent of kdbGet), you need to do all validation manually.

So why would you want to call elektraGetOpts directly? The advantage of calling elektraGetOpts directly is that you have more control over where the specification comes from. With gopts we need to mount the specification before starting our application (at least before calling kdbOpen). But if we call elektraGetOpts directly, we can pass whatever KeySet we want. This can be useful, if you are writing a custom configuration tool (like the standard kdb tool, but specific to your use case). Such tools normally don't need (or want) to be configured via persistent config files, but often have an advanced command-line interface. Using the fairly feature-rich parser implemented in elektraGetOpts could be a good option here.

Advanced Use: Manually Generating the Help Message

When using gopts you automatically get a generated help message via the key proc:/elektra/gopts/help/message, whenever the --help option is used. But if you want to show the help message in another case, or you cannot use the auto-generated message for some reason (e.g. you want to add a prefix text, but the content of this text is not known before calling kdbOpen()), then you need to generate the help message manually.

If you are using elektraGetOpts directly, manually generating the help message is the only option.

The help message can be generated with elektraGetOptsHelpMessage.

char * elektraGetOptsHelpMessage (Key * helpKey, const char * usage, const char * prefix);

To access this function, you first need to link your application against libelektra-opts and then include the header kdbopts.h.

Note: libelektra-opts is an internal library and kdbopts.h an internal header. We do not make and guarantees to the API stability of the functions declared in kdbopts.h. We will do our best to keep the API compatible, but the only way to have guaranteed API stability and backwards compatibility is to only use the parser via gopts.

Calling elektraGetOptsHelpMessage allocates a new string that will contain the generated help message. You need to free the string with elektraFree, once you are done with it.

The parameter helpKey has to be the same key as passed to elektraGetOpts as parentKey, if you called elektraGetOpts directly. If you used gopts, you should pass the key proc:/elektra/gopts/help.

The parameter usage is used to replace the default usage line, if it is not NULL. The string prefix is inserted between the usage line and the options list, if it is not NULL.

With gopts this could look something like this:

int main (int argc, char ** argv)
{
  // setup omitted for brevity
  kdbGet (kdb, ks, parentKey);

  if (/* custom condition */)
  {
    Key * help = ksLookupByName (ks, "proc:/elektra/gopts/help", 0);
    char * helpMessage = elektraGetOptsHelpMessage (help, "custom usage", "custom prefix");
    printf ("%s\n", helpMessage);
    elektraFree (helpMessage);
  }
}