NAME

parse_args - Core-style argument parsing for scripts

SYNOPSIS

package require parse_args ?0.7.2?

parse_args::parse_args args argspec ?varname?

DESCRIPTION

The commands provided by the Tcl core follow a pattern of optional, named arguments (named with a leading “-” to indicate an option) in addition to positional arguments. They strongly prefer named arguments when the number of arguments and particularly optional arguments grows beyond about 3. Tcl scripts however have no access to standard argument parsing facilities to implement this pattern and have to build their named argument parsing machinery from scratch. This is slow to do in script, error prone, and creates noise in the code that distracts from the core task of the command implementation. This means that script-defined commands generally present half-baked argument handling APIs and tend to over-use positional arguments.

This package aims to address this weakness by providing a robust argument parsing mechanism to script code that can implement the full set of patterns established by the core Tcl commands, is terse and readable to specify the arguments, and is fast enough to use in any situation.

COMMANDS

parse_args::parse_args args argspec
Parse the arguments supplied in args against the specification given in argspec, setting variables in the current scope as defined in the argspec. If the optional varname argument is supplied, it names a variable that will have a dictionary written to it containing the parsed arguments and no variables will be created for the arguments. See ARGUMENT SPECIFICATION below for the format of argspec.

ARGUMENT SPECIFICATION

The syntax of the argument specification (given in the argspec argument to the parse_args command) is a dictionary, with the keys corresponding to the names by which the arguments will be available to the caller, either as variables created by the parse_args command or keys in the dictionary written to varname if that was specified. If the first character of the key is a “-” character it defines a named option, otherwise a positional parameter. Named options and positional parameters may be freely interleaved in the argspec: positional parameters are bound in the order they appear in it, and the order of the named options is irrelevant. (The arguments being parsed still need to supply any named options before the positional arguments - see NAMED OPTION SCAN.) Each key may only appear once: a repeated key is an error. Named options must be supplied exactly as they appear in the argspec: unlike Tcl_GetIndexFromObj, unique prefixes are not accepted. Like in proc argument definitions the name args is special and will consume all remaining arguments (after the leading positionals are bound and modulo any reservation for trailing -required positionals - see below for details); a positional may use -args all to opt into the same behaviour under a different name. The one difference is that args, like a proc’s args, has an implicit -default of the empty list, so it is always set, whereas a -args all positional that receives no arguments is left unset unless it has a -default.

The values of the dictionary contain the definition of the corresponding argument, such as -required, which marks that argument as being required (whether a named or positional argument). The options that are available to define the argument are:

-default val
Provides a default value for the argument if it wasn’t supplied. Without this an argument that wasn’t supplied won’t have its corresponding variable (or dictionary key) set. val is permitted to be a value that is outside of the range of possible values that would be accepted from the caller, by design.

-required
If this boolean option is given, the argument is flagged as required. Failing to supply the arg causes the parse_args to throw an exception similar to what would happen if a required positional argument to a proc was not provided.

-validate cmdprefix
Append the value supplied for this argument to the cmdprefix and run it. The value is rejected if an exception is thrown or the value returned isn’t a Tcl boolean true value.

-name newname
Change the name of the variable / dictionary key that will receive the value for this argument to newname. By default the name will be the argspec key (without the leading “-” for named arguments).

-boolean
Flag this argument as a boolean, similar to -nocase to regexp. The variable will always be defined and be a boolean, true if the argument was supplied and false if it wasn’t. Only valid for named options; using -boolean on a positional argument raises an error at spec compile time.

-args count
By default named arguments consume a single following argument as the value for that argument. This setting changes that to count, which will cause the following count arguments to be gathered into the variable holding this argument. If count is the special value “all”, then all remaining arguments are gathered into a list as the value of this argument. A negative count is an error, and a count of 0 is only valid for named options (equivalent to -boolean); using -args 0 on a positional argument raises an error at spec compile time.

-enum possible_values
Restrict the values that will be accepted for this argument to those in the possible_values list. Similar to using a validator that checks for inclusion in that list but with additional internal performance optimizations.

-# free_form_text
Embed a comment for this argument. The supplied free_form_text is ignored by parse_args and serves only to document the argument to programmers reading the code. In some future version this text may be incorporated into a usage error message that is generated when argument parsing fails.

-multi
Group a set of defined arguments together, so they act like mutually exclusive flags. Used to implement patterns like the arguments -ascii, -dictionary, -integer and -real arguments to the core lsort command. name will be the name of the variable that holds the flag that was supplied, without its leading “-” (so -integer gives integer). A -default setting on any of the linked -multi arguments (those that share a name) will supply a default value in name if none were passed. A -required setting on any of the linked -multi arguments will require that the caller supply at least one of them. If multiple different flags were given, the one last in the argument list applies, unless -all was specified (on any linked option), in which case all instances are grouped as a list in name, in the order they were specified. -multi is only valid for named options, and can’t be combined with -enum; either raises an error at spec compile time.

-alias
Treat the value supplied for this argument as a variable name and bind the resulting variable for this argument to the variable of that name in the call frame above this one (upvar 1). Not supported when a varname is given (dictionary mode), where it raises an error.

-all
Collect all the instances of this argument as a list, rather than the default behaviour of using the last instance as the value.

-end
Treat the remaining arguments as if -- directly followed this option - that is: all words in args after those consumed by the current option are treated as positional arguments.

NAMED OPTION SCAN

Named options are recognised by scanning the arguments from the start. The scan ends, and all the remaining arguments are bound to positional parameters, at the first of:

An argument that doesn’t start with “-” (including the empty string).
The argument --, which is consumed (it isn’t bound to any positional). Only the first -- is special: a later one is an ordinary value.
The arguments consumed by an option marked -end.
The arguments reserved for trailing -required positionals. These are the last n arguments, where n is the total number of arguments the -required positionals (without a -default) need. They are never treated as options, even if they look like one. For example, with the argspec {-x {} name {-required}}, the arguments -foo and -x 1 -foo both set name to -foo, but -foo bar is an unknown option error.

If the argspec defines no named options, there is no scan at all: every argument is positional, including any that start with “-”, and -- is an ordinary value rather than a separator. So parse_args {-foo} {name {-required}} sets name to -foo, and parse_args {--} {name {}} sets name to --.

An argument within the scan that starts with “-” and isn’t -- must be a named option from the argspec, otherwise it’s an error (see ERRORS). To pass a positional value that may start with “-” to a command that has named options, callers should precede it with --.

POSITIONAL BINDING

Positional parameters are bound left-to-right with one element of forward lookahead. Each positional p is processed in order:

If p is args (or has -args all): p consumes all remaining input arguments, except that arguments needed to satisfy any later -required positionals are reserved for them. If after that reservation p would consume zero arguments and it has no -default (args always has one), the variable for p is left unset.
If p is -required (without a -default): p consumes its max(1, arg_count) arguments. If too few input arguments remain to satisfy this it is a usage error.
Otherwise (optional, or -required with a -default): p consumes its arg_count arguments only when at least arg_count + (count needed for later -required positionals) arguments remain. Otherwise p’s -default is applied (if any) and consumption skips to the next positional.

This means optional positionals between or before -required positionals are skipped automatically when the caller supplies just enough arguments to satisfy the -required ones, rather than greedily filling the optional and erroring on the missing required. An argspec may contain at most one args/-args all positional, and any positional that follows it must be -required without a -default.

The string representation of an input argument is generated only for arguments that fall within the named-option scan (see NAMED OPTION SCAN). Arguments reserved for trailing -required positionals, and all arguments when the argspec has no named options, are bound by position only and never have their first character inspected. So shimmer-hostile values, such as large pure bytearrays whose string form is expensive to generate, pass through without being converted.

ERRORS

An invalid argspec is reported when it is first compiled, which is the first time it is used to parse arguments (or the first time after it has lost its internal representation). Usage errors are reported with these errorcodes:

TCL LOOKUP INDEX option name
name (which starts with “-” and is in the named-option scan) isn’t a named option in the argspec. The message lists the valid options, as for Tcl_GetIndexFromObj.

TCL LOOKUP INDEX option value
value isn’t one of the -enum values for option.

TCL WRONGARGS
More arguments were supplied than the argspec can bind, or a named option was missing its value(s). The message is a usage generated from the argspec, in the usual wrong # args: should be form but without a command name, since parse_args can’t know which name is right (it may be parsing a proc’s args, values passed into a resumed coroutine, a script’s argv, …). A caller that wants one can add it when handling the error.

PARSE_ARGS REQUIRED name
The -required argument name wasn’t supplied.

PARSE_ARGS REQUIRED_ONE_OF names
None of the -multi group names, one of which is -required, was supplied.

PARSE_ARGS WRONGARGS name
Too few arguments remained to fill the -required positional name, which has -args count greater than 1.

PARSE_ARGS VALIDATION name
The -validate command prefix rejected the value supplied for name.

C API

The same argument parsing is available to C code, through a Tcl stubs table. A separately compiled extension depends on parse_args (pkg-config or meson dependency(‘parse_args’), which supply -DUSE_PARSE_ARGS_STUBS=1 and the stub library), includes parse_args.h and calls ParseArgs_InitStubs(interp) in its Init. Code compiled by jitc with a package {parse_args 0.7.2} part calls the functions directly.

int ParseArgs_BindDict(interp, argspec, objc, objv, dictPtr)
Bind objc/objv against the argspec Tcl_Obj, as the parse_args command does with a varname, leaving the result dict in *dictPtr (which must point to NULL or an object reference, which is replaced: replace_tclobj semantics).

int ParseArgs_Bind(interp, argspec, objc, objv, cb, cdata)
Bind objc/objv against the argspec Tcl_Obj, calling cb(cdata, interp, decl_idx, name, val) for each value bound. decl_idx is the index of the argspec entry that the value is for, counting both options and positional parameters in the order they appear in argspec (or -1 for the value of a -multi group), and name is its output name. val is only guaranteed to be valid during the callback.

int ParseArgs_BindSlots(interp, cachePtr, specstr, objc, objv, slotc, slots)
Bind objc/objv against the argspec in *cachePtr, which is created from the string specstr when *cachePtr is NULL (and kept there, with a reference held, for the caller to release), or for this call only if cachePtr is NULL. Each value is converted as directed by slots[decl_idx], which must have an entry for each argspec entry, and stored at its addr. A reference to the value is kept in the slot’s hold member (which must initially be NULL), keeping pointers into it valid until the caller releases it with ParseArgs_ReleaseSlots(slotc, slots). -multi and -alias are not supported. This is the mechanism behind parseargs_bind.def.

All three leave an error message (and errorcode, as for parse_args) in the interp result on failure.

parseargs_bind.def

parseargs_bind.def declares and binds a struct local in a C command implementation, from an X-macro that declares the argspec:

static int foo(void* cdata, Tcl_Interp* interp, int objc, Tcl_Obj*const objv[])
{
    #define PARSEARGS_SPEC(OPT, ARG) \
        OPT( special,  "-boolean",     BOOL    ) \
        ARG( a1,       "-required",    STRING  ) \
        ARG( a2,       "-default 42",  INT     ) \
        ARG( a3,       "-required",    WIDEINT ) \
        /* line intentionally left blank */
    #include <parseargs_bind.def>

    // args.special, args.a1, args.a2, args.a3 ...
    return TCL_OK;
}

Each entry is OPT(name, settings, type) for an option (-_name_) or ARG(name, settings, type) for a positional parameter. name is both the C struct member and the argspec name (so options are limited to names that are valid C identifiers), settings is a string literal with the argspec settings for the entry, and type selects the member’s C type and the conversion:

type	Member(s)	Conversion
OBJ	Tcl_Obj* name	The value itself (NULL if not supplied)
BOOL	bool name	Tcl_GetBooleanFromObj
INT	int name	Tcl_GetIntFromObj
WIDEINT	Tcl_WideInt name	Tcl_GetWideIntFromObj
DOUBLE	double name	Tcl_GetDoubleFromObj
STRING	const char* name	Tcl_GetString
BYTES	const unsigned char* name, Tcl_Size name_len	Tcl_GetBytesFromObj
LIST	Tcl_Size namec, Tcl_Obj** namev	Tcl_ListObjGetElements (for -args n and args)
INDEX	int name	Index of the value in the entry’s -enum choices
SKIP	(none)	Not stored

Further types can be added by defining PARSEARGS_OUTTYPE__type_(name) (the member declaration) and PARSEARGS_SLOTFOR__type_(out, name) (the struct ParseArgs_Slot initializer, using PARSEARGS_SLOT_CUSTOM and a converter function). Members for entries that are not supplied and have no -default are left zeroed.

Values are held until the struct goes out of scope, so pointer members stay valid until then. The release is done with defer, so the code including parseargs_bind.def must be compiled as C23 with defer support (as jitc cdefs are). These macros may be defined before the include to override the defaults, and are all undefined afterwards:

PARSEARGS_OUT
The name of the struct local, default args.

PARSEARGS_INTERP, PARSEARGS_OBJC, PARSEARGS_OBJV
Defaults interp, objc-1 and objv+1.

PARSEARGS_CACHE
A Tcl_Obj** in which to cache the compiled argspec across calls, or NULL to build it for each call. The default, ((Tcl_Obj**)cdata), is the per-binding slot that jitc::bind provides as the ClientData of a jitc OBJCMD (and jitc releases it when the command is deleted). Under jitc::capply cdata is NULL, so the argspec is built for each call. Other code must supply its own slot, and release it.

PARSEARGS_ON_ERROR
The statement to run when binding fails, default return TCL_ERROR.

EXAMPLES

Mimic the argument handling of the core glob command:

proc glob args {
    parse_args::parse_args $args {
        -directory  {}
        -join       {-boolean}
        -nocomplain {-boolean}
        -path       {}
        -tails      {-boolean}
        -types      {-default {}}
        args        {-name patterns}
    }

    if {$join} {
        set patterns    [list [file join {*}$patterns]]
    }

    if {[llength $patterns] == 0 && $nocomplain} return

    foreach pattern $patterns {
        if {[info exists directory]} {
            ...
        }
    }
    ...
}

Mimic regex:

proc regexp args {
    parse_args::parse_args $args {
        -about      {-boolean}
        -expanded   {-boolean}
        -indices    {-boolean}
        -line       {-boolean}
        -linestop   {-boolean}
        -lineanchor {-boolean}
        -nocase     {-boolean}
        -all        {-boolean}
        -inline     {-boolean}
        -start      {-default 0}
        exp         {-required}
        string      {-required}
        matchvar    {}
        args        {-name submatchvars}
    }
    ...
}

lsort:

proc lsort args {
    parse_args::parse_args $args {
        -ascii      {-name compare_as -multi -default ascii}
        -dictionary {-name compare_as -multi}
        -integer    {-name compare_as -multi}
        -real       {-name compare_as -multi}

        -command    {}

        -increasing {-name order -multi -default increasing}
        -decreasing {-name order -multi}

        -indices    {-boolean}
        -index      {}
        -stride     {-default 1}
        -nocase     {-boolean}
        -unique     {-boolean}
        list        {-required}
    }

    if {![info exists command]} {
        switch -- $compare_as {
            ascii {
                set command {string compare}
                if {$nocase} {
                    lappend command -nocase
                }
            }
            dictionary {
                ...
            }
            integer - real {
                set command tcl::mathop::-
            }
        }
    }
}

lsearch:

proc lsearch args {
    parse_args::parse_args $args {
        -exact      {-name matchtype -multi}
        -glob       {-name matchtype -multi -default glob}
        -regexp     {-name matchtype -multi}

        -sorted     {-boolean}
        -all        {-boolean}
        -inline     {-boolean}
        -not        {-boolean}
        -start      {-default 0}

        -ascii      {-name compare_as -multi -default ascii}
        -dictionary {-name compare_as -multi}
        -integer    {-name compare_as -multi}
        -real       {-name compare_as -multi}

        -nocase     {-boolean}

        -decreasing {-name order -multi}
        -increasing {-name order -multi -default increasing}
        -bisect     {-boolean}

        -index      {}
        -subindices {-boolean}
    }

    if {$sorted && $matchtype in {glob regexp}} {
        error "-sorted is mutually exclusive with -glob and -regexp"
    }
}

A Tk widget - entry:

proc entry {widget args} {
    parse_args::parse_args $args {
        -disabledbackground {-default {}}
        -disabledforeground {-default {}}
        -invalidcommand     {-default {}}
        -readonlybackground {-default {}}
        -show               {}
        -state              {-default normal -enum {
            normal disabled readonly
        }}
        -validate           {-default none -enum {
            none focus focusin focusout key all
        }}
        -validatecommand    {-default {}}
        -width              {-default 0}
        -textvariable       {}
    }
}

BUGS

Please open an issue on the github tracker for the project if you encounter any problems: https://github.com/RubyLane/parse_args/issues

Navigation Menu

Repository navigation

README.md

README.md

NAME

SYNOPSIS

DESCRIPTION

COMMANDS

ARGUMENT SPECIFICATION

NAMED OPTION SCAN

POSITIONAL BINDING

ERRORS

C API

parseargs_bind.def

EXAMPLES

SEE ALSO

BUGS

LICENSE

FilesExpand file tree

README.md

Latest commit

History

README.md

File metadata and controls

NAME

SYNOPSIS

DESCRIPTION

COMMANDS

ARGUMENT SPECIFICATION

NAMED OPTION SCAN

POSITIONAL BINDING

ERRORS

C API

parseargs_bind.def

EXAMPLES

SEE ALSO

BUGS

LICENSE