specification of an individual parameter to the command's main function
the help text for the command
the subcommand path of the command
the subcommand path of the command
the arguments passed to the command
flag name spec ~doc
specifies a command that, among other things, takes a flag
named name
on its command line. doc
indicates the meaning of the flag.
All flags must have a dash at the beginning of the name. If name
is not prefixed
by "-", it will be normalized to "-" ^ name
.
Unless full_flag_required
is used, one doesn't have to pass name
exactly on the
command line, but only an unambiguous prefix of name
(i.e., a prefix which is not
a prefix of any other flag's name).
NOTE: the doc
for a flag which takes an argument should be of the form
arg_name ^ " " ^ description
where arg_name
describes the argument and
description
describes the meaning of the flag.
NOTE: flag names (including aliases) containing underscores will be rejected. Use dashes instead.
NOTE: "-" by itself is an invalid flag name and will be rejected.
Superceded by return
, preserved for backwards compatibility
composable command-line specifications
Ultimately one forms a basic command by combining a spec of type
('main, unit -> unit) t
with a main function of type 'main
; see the basic
function below. Combinators in this library incrementally build up the type of main
according to what command-line parameters it expects, so the resulting type of
main
is something like:
arg1 -> ... -> argN -> unit -> unit
It may help to think of ('a, 'b) t
as a function space 'a -> 'b
embellished with
information about:
One can view a value of type ('main_in, 'main_out) t
as function that transforms a
main function from type 'main_in
to 'main_out
, typically by supplying some
arguments. E.g. a value of type Spec.t
might have type:
(arg1 -> ... -> argN -> 'r, 'r) Spec.t
Such a value can transform a main function of type arg1 -> ... -> argN -> 'r
by
supplying it argument values of type arg1
, ..., argn
, leaving a main function
whose type is 'r
. In the end, Command.basic
takes a completed spec where
'r = unit -> unit
, and hence whose type looks like:
(arg1 -> ... -> argN -> unit -> unit, unit -> unit) Spec.t
A value of this type can fully apply a main function of type
arg1 -> ... -> argN -> unit -> unit
to all its arguments.
The final unit argument allows the implementation to distinguish between the phases of (1) parsing the command line and (2) running the body of the command. Exceptions raised in phase (1) lead to a help message being displayed alongside the exception. Exceptions raised in phase (2) are displayed without any command line help.
The view of ('main_in, main_out) Spec.t
as a function from 'main_in
to
'main_out
is directly reflected by the step
function, whose type is:
val step : ('m1 -> 'm2) -> ('m1, 'm2) t
spec1 ++ spec2 ++ ... ++ specN
composes spec1 through specN.
For example, if spec_a
and spec_b
have types:
spec_a: (a1 -> ... -> aN -> 'ra, 'ra) Spec.t
spec_b: (b1 -> ... -> bM -> 'rb, 'rb) Spec.t
then spec_a ++ spec_b
has the following type:
(a1 -> ... -> aN -> b1 -> ... -> bM -> 'rb, 'rb) Spec.t
So, spec_a ++ spec_b
transforms a main function it by first supplying spec_a
's
arguments of type a1
, ..., aN
, and then supplying spec_b
's arguments of type
b1
, ..., bm
.
One can understand ++
as function composition by thinking of the type of specs
as concrete function types, representing the transformation of a main function:
spec_a: \/ra. (a1 -> ... -> aN -> 'ra) -> 'ra
spec_b: \/rb. (b1 -> ... -> bM -> 'rb) -> 'rb
Under this interpretation, the composition of spec_a
and spec_b
has type:
spec_a ++ spec_b : \/rc. (a1 -> ... -> aN -> b1 -> ... -> bM -> 'rc) -> 'rc
And the implementation is just function composition:
sa ++ sb = fun main -> sb (sa main)
the empty command-line spec
combinator for patching up how parameters are obtained or presented
Here are a couple examples of some of its many uses
step (fun m v -> m ~foo:v) +> flag "-foo" no_arg : (foo:bool -> 'm, 'm) t
step (fun m user -> match user with | Some user -> m user | None -> print_string "enter username: "; m (read_line ())) +> flag "-user" (optional string) ~doc:"USER to frobnicate" : (string -> 'm, 'm) t
A use of step
might look something like:
step (fun main -> let ... in main x1 ... xN) : (arg1 -> ... -> argN -> 'r, 'r) t
Thus, step
allows one to write arbitrary code to decide how to transform a main
function. As a simple example:
step (fun main -> main 13.) : (float -> 'r, 'r) t
This spec is identical to const 13.
; it transforms a main function by supplying
it with a single float argument, 13.
. As another example:
step (fun m v -> m ~foo:v) : (foo:'foo -> 'r, 'foo -> 'r) t
This spec transforms a main function that requires a labeled argument into a main function that requires the argument unlabeled, making it easily composable with other spec combinators.
combinator for defining a class of commands with common behavior
Here are two examples of command classes defined using wrap
wrap (fun ~run ~main -> Exn.handle_uncaught ~exit:true (fun () -> run main) ) : ('m, unit) t -> ('m, unit) t
wrap (fun ~run ~main -> In_channel.iter_lines stdin ~f:(fun line -> run (main line)) ) : ('m, unit) t -> (string -> 'm, unit) t
Beware that an anonymous argument of type int
cannot be specified as negative,
as it is ambiguous whether -1 is a negative number or a flag. If you need to pass
a negative number to your program, make it a parameter to a flag.
Beware that an anonymous argument of type int
cannot be specified as negative,
as it is ambiguous whether -1 is a negative number or a flag. If you need to pass
a negative number to your program, make it a parameter to a flag.
a flag specification
no_arg
flags may be passed at most once. The boolean returned
is true iff the flag is passed on the command line
no_arg_register ~key ~value
is like no_arg
, but associates value
with key
in the in the auto-completion environment
no_arg_abort ~exit
is like no_arg
, but aborts command-line parsing
by calling exit
. This flag type is useful for "help"-style flags that
just print something and exit.
escape
flags may be passed at most once. They cause the command line parser to
abort and pass through all remaining command line arguments as the value of the
flag.
A standard choice of flag name to use with escape
is "--"
.
flags_of_args_exn args
creates a spec from Caml.Arg.t
s, for compatibility with
ocaml's base libraries. Fails if it encounters an arg that cannot be converted.
NOTE: There is a difference in side effect ordering between Caml.Arg
and
Command
. In the Arg
module, flag handling functions embedded in Caml.Arg.t
values will be run in the order that flags are passed on the command line. In the
Command
module, using flags_of_args_exn flags
, they are evaluated in the order
that the Caml.Arg.t
values appear in flags
.
a specification of some number of anonymous arguments
(name %: typ)
specifies a required anonymous argument of type typ
.
The name
must not be surrounded by whitespace, if it is, an exn will be raised.
If the name
is surrounded by a special character pair (<>, {}, [] or (),)
name
will remain as-is, otherwise, name
will be uppercased.
In the situation where name
is only prefixed or only suffixed by one of the
special character pairs, or different pairs are used, (e.g. "<ARG") an exn will
be raised.
The (possibly transformed) name
is mentioned in the generated help for the
command.
t2
, t3
, and t4
each concatenate multiple anonymous argument
specs into a single one. The purpose of these combinators is to allow
for optional sequences of anonymous arguments. Consider a command with
usage:
main.exe FOO [BAR BAZ]
where the second and third anonymous arguments must either both be there or both not be there. This can be expressed as:
t2 ("FOO" %: foo) (maybe (t2 ("BAR" %: bar) ("BAZ" %: baz)))]
Sequences of 5 or more anonymous arguments can be built up using nested tuples:
maybe (t3 a b (t3 c d e))