I have 4 files:
A.mli A.ml B.mli B.ml
Using ocamlc -c I have compiled A.mli then A.ml then B.mli with no errors
How can I use code form files A in B.ml
At the begining of file B.ml I have
#load "A.cmo"
and this line fails to compile using ocamlc -c
How should I load module(compiled code) in my code?
The #load command only works in the toplevel (the OCaml REPL).
To use code from A.ml in B.ml, you just need to prefix names with A. in the code in B.ml. The compiler will find the compiled version of A.mli and will know what to do.
Later you need to link all the files together to produce an executable. You need to list them in dependency order (a module appears before the other modules that depend on it).
Here's a small example:
$ head a.mli a.ml b.mli b.ml main.ml
==> a.mli <==
val f : int -> int
==> a.ml <==
let f x = x + 10
==> b.mli <==
val g : int -> int
==> b.ml <==
let g x = A.f (x * 10)
==> main.ml <==
let main () = Printf.printf "%d\n" (B.g 12)
let () = main ()
$ ocamlc -c a.mli a.ml b.mli b.ml main.ml
$ ocamlc -o main a.cmo b.cmo main.cmo
$ ./main
130
(There are tools that will automate much of this, which makes OCaml development less tedious. But this should show the basic idea.)
Related
I am trying to build a simple program using Yojson: the main.ml file is just
let () = exit 1;
and an mod.ml file contains
open Yojson
open Yojson.Basic.Util
let rec json_to_tree json =
let _ = json |> member "key" in
()
I have a Makefile which first compiles .ml files, then links. For simplicity, let assume we run the three following commands:
ocamlfind ocamlc -package yojson -g -c mod.ml
ocamlfind ocamlc -package yojson -g -c main.ml
ocamlfind ocamlc -package yojson -g -o inter mod.cmo main.cmo
The problem is that the linking part produces the following error:
File "_none_", line 1:
Error: Required module `Yojson' is unavailable
which I have no idea to solve. Using the Unix module can lead to similar error which are resolved adding unix.cma when linking, but this doesn't help for Yojson.
For the sake of completeness, I installed yojson via opam, and ocamlfind list | grep yojson returns
yojson (version: 1.3.3)
yojson.biniou (version: 1.3.3)
This works.
ocamlfind ocamlc -package yojson -linkpkg -g -o inter mod.cmo main.cmo
While this works, https://github.com/ocaml/dune is the better option to building ocaml code.
I can take the following ocaml module test.ml and compile it with ocamlopt
(* test.ml *)
Printf.printf "hi\n"
By doing
$ ocamlopt -o test.native test.ml
$ ./test.native
hi
I can also compile it without linking.
$ ocamlopt -o test.o -c test.ml
$ ls
test.cmi test.cmx test.ml test.o
However, if I then try to link the object I just created with the OCaml runtime to get an executable, no files are created (even though ocamlopt exits normally)
$ ocamlopt -o test.native test.o
$ ls
test.cmi test.cmx test.ml test.o
How do I direct ocamlopt to link in the OCaml runtime and produce an executable "as if" I had run ocamlopt -o test.native test.ml?
You need to give the .cmx file to the compiler, not the .o file.
$ ocamlopt -o test.native test.cmx
$ ./test.native
hi
I have three files:
$ ls
lib.ml desk.ml test.ml
$ cat lib.ml
let myfunction () = print_endline "Hello world"
$ cat desk.ml
module Liberty = Lib
$ cat test.ml
Desk.Liberty.myfunction ()
I want to compile a desk.cma that does not expose the module Lib, but is able to make use of it. For example, I've tried:
$ ocamlc -a lib.ml -o lib.cma
$ ocamlc -a lib.cma desk.ml -o desk.cma
$ ocamlc desk.cma test.ml -o test.byte
Unfortunately, if you change test.ml to be Lib.myfunction (), the same compilation steps work just fine. I want it so that Lib is not exposed to test.ml and only exposed to desk.ml. Does anyone know how to do this? Thanks very much!
It looks like you're looking for the -pack mechanism of OCaml. This allows you to create a pack.cmo (or pack.cmx for native compilation) from several compilation units, like a pack.cma library, but with the added advantage that you can have a corresponding pack.mli interface with which you can hide the parts that should be internal to your pack. In your example, you could for instance have:
-- lib.ml --
let myfunction () = print_endline "Hello world"
-- liberty.ml --
include Lib
-- desk.mli --
(* No module Lib: Lib will not be seen outside of the pack. *)
module Liberty: sig
val myfunction: unit -> unit
end
Then you can compile that with
ocamlc -for-pack Desk -c lib.ml
ocamlc -for-pack Desk -c liberty.ml
ocamlc desk.mli
ocamlc -pack -o desk.cmo lib.cmo liberty.cmo
This will give you a module Desk, that contains only Liberty a sub-module. Of course, desk.mli can be used to get more fine-grained restriction, e.g. with
module Lib: sig end
module Liberty: sig val myfunction: unit -> unit end
you export Lib with an empty signature, thus hiding a single (albeit the only one in the example ;-P) function from Lib.
You're asking that Lib be a module of desk.cma that Desk can use, but no other modules. I'm pretty sure the structure of a cma file doesn't offer this kind of control.
If you define Lib inside Desk you can make it a hidden detail.
$ ls
desk.ml desk.mli test1.ml test2.ml test3.ml
$ cat desk.ml
module Lib = struct
let myfunction () = print_endline "Hello world"
end
module Liberty = Lib
$ cat desk.mli
module Liberty : sig
val myfunction : unit -> unit
end
$ cat test1.ml
Desk.Liberty.myfunction ()
$ cat test2.ml
Lib.myfunction ()
$ cat test3.ml
Desk.Lib.myfunction ()
The Liberty module is visible to the test programs, but the Lib module isn't:
$ ocamlc -c desk.mli
$ ocamlc -a desk.ml -o desk.cma
$ ocamlc desk.cma test1.ml -o test1.byte
$ ./test1.byte
Hello world
$ ocamlc desk.cma test2.ml -o test2.byte
File "test2.ml", line 1, characters 0-14:
Error: Unbound module Lib
$ ocamlc desk.cma test3.ml -o test3.byte
File "test3.ml", line 1, characters 0-19:
Error: Unbound module Desk.Lib
You can implement Lib in a separate file, then include it into the definition of the Desk.Lib module.
I suspect this isn't a fully satisfying answer, but I hope it's helpful.
I am new to the OCaml and I'm confused with the file of .cma, .cmo and .cmx. Sometimes I have to include a .cma file in the compile command but sometimes I have to include a .cmo file.
Why is there a such difference for library? Is it the same concept in C++ as library and include path?
Example: ocamlc -o executable str.cma extstring.cmo mycode.ml
Thanks
You may think that .cmo is library, but it's not.
.cmo is bytecode object file. It's like .class in Java.
.cma is bytecode library. It's produce by archiving some .cmo into .cma
.cmx is produced by ocamlopt (the native-code compiler). Main output files of ocamlopt is .o but ocamlopt also produce it.
To create executable files, we arrange .cmo and .cma like ocamlc -o executable lib.cma a.cmo b.cmo ... to link them.
You can write .ml here instead of .cmo but it's the same as compiling .ml with -c and link the .cmo and other files.
For your deep undestanding, it's better to check how each files (related to ocaml) are produced.
Let's see what files are produce by ocamlc and ocamlopt.
[/tmp/test] ls
test.ml
[/tmp/test] cat ./test.ml
let id x = x
[/tmp/test] ocamlc -c /tmp/test/test.ml
[/tmp/test] ls
test.cmi test.cmo test.ml
[/tmp/test]
Now I compiled test.ml file and compile it with ocamlc with -c option
(the content of test.ml is not matter here).
You see ocamlc outputs two files:
test.cmi: Compiled interface file. This file includes type information of functions, variables in test.ml for separate compilation.
test.cmo: Bytecode object file: It's like .class file in Java.
We use .cmo files to create executable files.
[/tmp/test] ocamlc -c test.ml
[/tmp/test] ocamlc -o a.out test.cmo
[/tmp/test] ls
a.out test.cmi test.cmo test.ml
You see a.out file are produced via the .cmo file.
.cma are library files. These are produced by composing multiple .cmo files.
[/tmp/test] ls
test.ml lib.ml
[/tmp/test] cat lib.ml
let i = Test.id 1
let j = Test.id 2
[/tmp/test] ocamlc -c test.ml; ocamlc -c lib.ml
[/tmp/test] ls
lib.cmi lib.cmo lib.ml test.cmi test.cmo test.ml
[/tmp/test] ocamlc -a -o testlib.cma ./test.cmo lib.cmo
[/tmp/test] ls
lib.cmi lib.cmo lib.ml test.cmi test.cmo test.ml testlib.cma
Now I create lib.ml (which use id function in test.ml) and compile test.ml and lib.ml, then link them to create testlib.cma (option -a means creating a library).
You can see .cma is just packed with .cmo files.
To use the library, we just arrange it with other object files.
[/tmp/test] cat user.ml
let _ = print_int (Lib.i + Lib.j)
[/tmp/test] ocamlc -c user.ml
[/tmp/test] ocamlc -o a.out testlib.cma user.cmo
Finally, let's check what files are produce by ocamlopt.
[/tmp/test] ocamlopt -c ./test.ml
[/tmp/test] ls
test.cmi test.cmx test.ml test.o
ocamlopt produces
test.o: Native object file
test.cmi: Compiled interaface file
test.cmx: Also native object file, but it is mainly used for inlining functions across files!!!
So, the difference is here (when there's a x in the file extension, it means it's a native-code compiled object so that it runs faster. The cmx files are obtained with ocamlopt and the cmo files with ocamlc) :
.cma / .cmxa -> libraries that are already available for the programmer (standard libraries and libraries you installed)
.cmo / .cmx -> object files corresponding to your .ml files
TL;DR : .cm(x)a files were not created by you, .cm(o|x) files were.
What's the simplest command line to produce a native OCaml executable from a set of OCaml and C sources which use a C library that needs to be included via -l<lib>, such as -lm?
For instance, the following code illustrates a (contrived) example where it would be necessary:
// test.c
#include <math.h>
#include <caml/alloc.h>
value my_sqrt(value x) { return copy_double(sqrt(Double_val(x))); }
// test.ml
external sqrt_c: float -> float = "my_sqrt"
let () =
Printf.printf "srqt(5) = %g\n" (sqrt_c 5.0)
In this example, ocamlc -o next -custom test.c test.ml -cclib -lm produces OCaml bytecode, but -custom is not available for ocamlopt.
You have to use a different filename for your C-File:
ocamlopt test-native.c test.ml -cclib -lm -o test