Cppo is an equivalent of the C preprocessor for OCaml programs. It allows the definition of simple macros and file inclusion.
Cppo is:
- more OCaml-friendly than cpp
- easy to learn without consulting a manual
- reasonably fast
- simple to install and to maintain
- Author: Martin Jambon
- OCaml-community maintainers:
- Martin Jambon (@mjambon)
- Yishuai Li (@liyishuai)
- License: BSD 3-Clause "New" or "Revised" License
- Compatible OCaml versions: 4.02.3 or later
- Additional dependencies:
- Dune 1.10 or later
- OCamlbuild and Findlib, for Ocamlbuild plugin
The easiest way to install the latest released version of cppo is via OPAM:
opam install cppo
To instead build and install manually, do:
git clone https://github.com/ocaml-community/cppo.git
cd cppo
make
make install
Cppo is a preprocessor for programming languages that follow lexical rules
compatible with OCaml including OCaml-style comments (* ... *)
. These include Ocamllex, Ocamlyacc, Menhir, and extensions of OCaml based on Camlp4, Camlp5, or ppx. Cppo should work with Bucklescript as well. It won't work so well with Reason code because Reason uses C-style comment delimiters /*
and */
.
Cppo supports a number of directives. A directive is a #
sign placed
at the beginning of a line, possibly preceded by some whitespace, and followed
by a valid directive name or by a number:
BLANK* "#" BLANK* ("def"|"enddef"|"define"|"undef"
|"scope"|"endscope"
|"if"|"ifdef"|"ifndef"|"else"|"elif"|"endif"
|"include"
|"warning"|"error"
|"ext"|"endext") ...
A macro definition that is delimited by #def
and #enddef
can span
several lines. There is no need for protecting line endings with
backslash characters \
.
A directive (other than #def ... #enddef
)
can be split into multiple lines by placing a backslash character \
at
the end of the line to be continued. In general, any special character
can be used as a normal character by preceding it with backslash.
#include "hello.ml"
This is how a source file hello.ml
can be included.
Relative paths are searched first in the directory of the current file
and then in the search paths added on the command line using -I
, if any.
This is a simple macro that doesn't take an argument ("object-like macro" in the cpp jargon):
#define Ms Mississippi
match state with
Ms -> true
| _ -> false
After preprocessing by cppo, the code above becomes:
match state with
Mississippi -> true
| _ -> false
If needed, defined macros can be undefined. This is required prior to redefining a macro:
#undef X
An important distinction with cpp is that only previously-defined macros are accessible. Defining, undefining or redefining a macro has no effect on how previous macros will expand.
Macros can take arguments. That is, a macro can be parameterized;
this is known as a "function-like macro" in cpp
jargon.
When a parameterized macro is defined
and when it is applied,
the opening parenthesis must stick to the macro's identifier:
that is, there must be no space in between.
For example, this text:
#define debug(args) if !debugging then Printf.eprintf args else ()
debug("Testing %i" (1 + 1))
is expanded into:
if !debugging then Printf.eprintf "Testing %i" (1 + 1) else ()
An ordinary macro, which takes no arguments, can be viewed as a parameterized macro that takes zero arguments. However, the syntax differs: when there is no argument, no parentheses are used; when there is at least one argument, parentheses must be used. Here is a summary of the valid syntaxes:
#define FOO 42 (* Definition of an ordinary macro *)
FOO (* A use of an ordinary macro *)
#define BAR() 42 (* Invalid! When parentheses are used,
there must be at least one parameter *)
#define BAR(x) 42+x (* Definition of a parameterized macro *)
BAR(0) (* A use of this parameterized macro *)
BAR() (* Another valid use -- the argument is empty *)
All user-definable macros are constant. There are however two
predefined variable macros: __FILE__
and __LINE__
which take the value
of the position in the source file where the macro is being expanded.
#define loc (Printf.sprintf "File %S, line %i" __FILE__ __LINE__)
Macros can be defined on the command line as follows:
# preprocessing only
cppo -D 'VERSION 1.0' example.ml
# preprocessing and compiling
ocamlopt -c -pp "cppo -D 'VERSION 1.0'" example.ml
A macro definition that begins with #define
can span several lines.
In that case, the end of each line must be protected with a backslash
character, as in this example:
#define repeat_until(action,condition) \
action; \
while not (condition) do \
action \
done
In other words, at the first line ending that is not preceded by a \
character, an #enddefine
token is implicitly generated,
and the definition ends.
This convention, which is inherited from C, causes two problems. First,
protecting every line ending with a \
character is painful. Second, more
seriously, this convention does not allow macro definitions to be nested.
Indeed, if one attempts to nest two definitions that begin with #define
,
then only one #enddefine
token is generated; it is generated at the first
unprotected line ending. So, the beginnings and ends of definitions cannot
be correctly balanced.
These problems are avoided by using an alternative syntax where the beginning
and end of a macro definition are explicitly marked by #def
and #enddef
.
Here is an example:
#def repeat_until(action,condition)
action;
while not (condition) do
action
done
#enddef
With this syntax, a macro can span several lines:
there is no need to protect line endings with \
characters.
Furthermore, this syntax allows macro definitions to be nested:
inside a macro definition that is delimited by #def
and #enddef
,
both #def
and #define
can be used.
A parameterized macro can take a parameterized macro as a parameter: this is known as a higher-order macro.
To enable this feature, some annotations are required: when a macro parameter is itself a parameterized macro, it must be annotated with its type.
A macro takes n arguments (where n can be zero) and returns a piece of text. So, to describe the type of a macro, it suffices to describe the types of its n arguments.
Thus, the syntax of types is
τ ::= [τ ... τ]
.
That is, a type is a sequence of n types,
without separators,
surrounded with square brackets.
An ordinary macro,
which takes zero parameters,
has type []
.
This is the base type: in other words, it is the type of text.
For greater readability,
this type can also be written in the form of a single period, .
.
Here are a few examples of types:
. (* An ordinary unparameterized macro: in other words, text *)
[] (* Same as above. *)
[.] (* A parameterized macro that expects one piece of text *)
[..] (* A parameterized macro that expects two pieces of text *)
[[.].] (* A parameterized macro
whose first parameter is a parameterized macro of type [.]
and whose second parameter is a piece of text *)
In the definition of a parameterized macro M
,
each parameter X
can be annotated with a type
by writing X : τ
.
This is optional: if no annotation is provided,
the base type .
is assumed.
If a parameter X
is annotated with a type τ
other than the base type,
then, when the parameterized macro M
is applied,
the actual argument Y
that is supplied as an instance for X
must be the name of a macro of type τ
.
This is more easily explained via an example. In the following code,
#define TWICE(e) (e + e)
#define APPLY(F : [.], e) (let x = (e) in F(x))
let forty_two =
APPLY(TWICE,1+2+3+4+5+6)
TWICE
is a parameterized macro of type [.]
, and
APPLY
is a higher-order macro, whose type is [[.].]
.
Thus, the application APPLY(TWICE, ...)
is valid.
This code is expanded into:
let forty_two =
(let x = (1+2+3+4+5+6) in (x + x))
When a block of text is delimited by #scope ... #endscope
,
all macro definitions (#define
, #def ... #enddef
)
and undefinitions (#undef
)
become local:
they take effect only within this block.
(* Here, assume that the macro FOO is not defined. *)
#scope
#define FOO "FOO is now defined"
let x = FOO (* FOO expands to "FOO is now defined" *)
#endscope
(* Here, the macro FOO is again not defined. *)
#define FOO 42
let y = FOO (* FOO expands to 42 *)
Scopes can be nested, as illustrated by this example:
#scope
#define HELLO "Hello, "
#scope
#define MAN "man"
let message1 = HELLO ^ MAN
#endscope
(* Here, MAN is no longer defined, but HELLO still is. *)
let message2 = HELLO ^ "world"
#endscope
Here is a quick reference on conditionals available in cppo. If you
are not familiar with #ifdef
, #ifndef
, #if
, #else
and #elif
, please
refer to the corresponding section in the cpp manual.
#ifndef VERSION
#warning "VERSION is undefined"
#define VERSION "n/a"
#endif
#ifndef VERSION
#error "VERSION is undefined"
#endif
#if OCAML_MAJOR >= 3 && OCAML_MINOR >= 10
...
#endif
#ifdef X
...
#elif defined Y
...
#else
...
#endif
The boolean expressions following #if
and #elif
may perform arithmetic
operations and tests over 64-bit ints.
Boolean expressions:
defined
... followed by an identifier, returns true if such a macro existstrue
false
(
...)
- ...
&&
... - ...
||
... not
...
Arithmetic comparisons used in boolean expressions:
- ...
=
... - ...
<
... - ...
>
... - ...
<>
... - ...
<=
... - ...
>=
...
Arithmetic operators over signed 64-bit ints:
(
...)
- ...
+
... - ...
-
... - ...
*
... - ...
/
... - ...
mod
... - ...
lsl
... - ...
lsr
... - ...
asr
... - ...
land
... - ...
lor
... - ...
lxor
... lnot
...
Macro identifiers can be used in place of ints as long as they expand to an int literal or a tuple of int literals, e.g.:
#define one 1
#if one + one <> 2
#error "Something's wrong."
#endif
#define VERSION (1, 0, 5)
#if VERSION <= (1, 0, 2)
#error "Version 1.0.2 or greater is required."
#endif
Version strings (http://semver.org/) can also be passed to cppo on the
command line. This results in multiple variables being defined, all
sharing the same prefix. See the output of cppo -help
(copied at the
bottom of this page).
$ cppo -V OCAML:`ocamlc -version`
#if OCAML_VERSION >= (4, 0, 0)
(* All is well. *)
#else
#error "This version of OCaml is not supported."
#endif
Output:
# 2 "<stdin>"
(* All is well. *)
Location directives are the same as in OCaml and are echoed in the output. They consist of a line number optionally followed by a file name:
# 123
# 456 "source"
Warnings and error messages can be produced by the preprocessor:
#ifndef X
#warning "Assuming default value for X"
#define X 1
#elif X = 0
#error "X may not be null"
#endif
Cppo provides a mechanism for converting sections of a file using
and external program. Such a section must be placed between #ext
and
#endext
directives.
$ cat foo
ABC
#ext lowercase
DEF
#endext
GHI
#ext lowercase
KLM
NOP
#endext
QRS
$ cppo -x lowercase:'tr "[A-Z]" "[a-z]"' foo
# 1 "foo"
ABC
def
# 5 "foo"
GHI
klm
nop
# 10 "foo"
QRS
In the example above, lowercase
is the name given on the
command-line to external command 'tr "[A-Z]" "[a-z]"'
that reads
input from stdin and writes its output to stdout.
The following characters can be escaped by a backslash when needed:
(
)
,
#
In OCaml #
is used for method calls. It is usually not a problem
because in order to be interpreted as a preprocessor directive, it
must be the first non-blank character of a line and be a known
directive. If an object has a define method and you want #
to appear
first on a line, you would have to use \#
instead:
obj
\#define
Line directives in the usual format supported by OCaml are correctly interpreted by cppo.
Comments and string literals constitute single tokens even when they span across multiple lines. Therefore newlines within string literals and comments should remain as-is (no preceding backslash) even in a macro body:
#define welcome \
"**********
*Welcome!*
**********
"
CONCAT()
is a predefined macro that takes two arguments, removes any
whitespace between and around them and fuses them into a single identifier.
The result of the concatenation must be a valid identifier of the
form [A-Za-z_][A-Za-z0-9_]+ or [A-Za-z], or empty.
For example,
#define x 123
CONCAT(z, x)
expands into:
z123
However the following is illegal:
#define x 123
CONCAT(x, z)
because 123z does not form a valid identifier.
CONCAT(a,b)
is roughly equivalent to a##b
in cpp syntax.
CAPITALIZE()
is a predefined macro that takes one argument,
removes any leading and trailing whitespace, reduces each internal
whitespace sequence to a single space character and produces
a valid OCaml identifer with first character.
For example,
#define EVENT(n,ty) external CONCAT(on,CAPITALIZE(n)) : ty = STRINGIFY(n) [@@bs.val]
EVENT(exit, unit -> unit)
is expanded into:
external onExit : unit -> unit = "exit" [@@bs.val]
STRINGIFY()
is a predefined macro that takes one argument,
removes any leading and trailing whitespace, reduces each internal
whitespace sequence to a single space character and produces
a valid OCaml string literal.
For example,
#define TRACE(f) Printf.printf ">>> %s\n" STRINGIFY(f); f
TRACE(print_endline) "Hello"
is expanded into:
Printf.printf ">>> %s\n" "print_endline"; print_endline "Hello"
STRINGIFY(x)
is the equivalent of #x
in cpp syntax.
An ocamlbuild plugin is available. To use it, you can call ocamlbuild
with the argument -plugin-tag package(cppo_ocamlbuild)
(only since
ocaml 4.01 and cppo >= 0.9.4).
Starting from cppo >= 1.6.0, the cppo_ocamlbuild
plugin is in a
separate OPAM package (opam install cppo_ocamlbuild
).
With Oasis :
OCamlVersion: >= 4.01
AlphaFeatures: ocamlbuild_more_args
XOCamlbuildPluginTags: package(cppo_ocamlbuild)
After that, you need to add in your myocamlbuild.ml
:
let () =
Ocamlbuild_plugin.dispatch
(fun hook ->
Ocamlbuild_cppo.dispatcher hook ;
)
By default the plugin will apply cppo on all files ending in .cppo.ml
cppo.mli
, and cppo.mlpack
, in order to produce .ml
, .mli
,
and.mlpack
files. The following tags are available:
cppo_D(X)
≡-D X
cppo_U(X)
≡-U X
cppo_q
≡-q
cppo_s
≡-s
cppo_n
≡-n
cppo_x(NAME:CMD_TEMPLATE)
≡-x NAME:CMD_TEMPLATE
- The tag
cppo_I(foo)
can behave in two way:- If
foo
is a directory, it's equivalent to-I foo
. - If
foo
is a file, it addsfoo
as a dependency and apply-I parent(foo)
.
- If
cppo_V(NAME:VERSION)
≡-V NAME:VERSION
cppo_V_OCAML
≡-V OCAML:VERSION
, whereVERSION
is the version of OCaml that ocamlbuild uses.
All delimiters,
including scope delimiters (#scope
and #endscope
),
delimiters of macro definitions (#def
and #enddef
),
and delimiters of conditional constructs (#if
, #endif
, etc.),
must be used in a well-balanced manner.
This requirement does not apply separately to each category of delimiters.
instead, it applies to all categories of delimiters at once.
This is a stricter requirement.
Thus, for example, #scope
cannot be followed with #endif
,
and #if
cannot be followed with #endscope
.
In other words,
a scope cannot contain a fragment of a conditional construct,
and a conditional construct cannot contain a fragment of a macro definition.
Usage: ./cppo [OPTIONS] [FILE1 [FILE2 ...]]
Options:
-D DEF
Equivalent of interpreting '#define DEF' before processing the
input
-U IDENT
Equivalent of interpreting '#undef IDENT' before processing the
input
-I DIR
Add directory DIR to the search path for included files
-V VAR:MAJOR.MINOR.PATCH-OPTPRERELEASE+OPTBUILD
Define the following variables extracted from a version string
(following the Semantic Versioning syntax http://semver.org/):
VAR_MAJOR must be a non-negative int
VAR_MINOR must be a non-negative int
VAR_PATCH must be a non-negative int
VAR_PRERELEASE if the OPTPRERELEASE part exists
VAR_BUILD if the OPTBUILD part exists
VAR_VERSION is the tuple (MAJOR, MINOR, PATCH)
VAR_VERSION_STRING is the string MAJOR.MINOR.PATCH
VAR_VERSION_FULL is the original string
Example: cppo -V OCAML:4.02.1
-o FILE
Output file
-q
Identify and preserve camlp4 quotations
-s
Output line directives pointing to the exact source location of
each token, including those coming from the body of macro
definitions. This behavior is off by default.
-n
Do not output any line directive other than those found in the
input (overrides -s).
-version
Print the version of the program and exit.
-x NAME:CMD_TEMPLATE
Define a custom preprocessor target section starting with:
#ext "NAME"
and ending with:
#endext
NAME must be a lowercase identifier of the form [a-z][A-Za-z0-9_]*
CMD_TEMPLATE is a command template supporting the following
special sequences:
%F file name (unescaped; beware of potential scripting attacks)
%B number of the first line
%E number of the last line
%% a single percent sign
Filename, first line number and last line number are also
available from the following environment variables:
CPPO_FILE, CPPO_FIRST_LINE, CPPO_LAST_LINE.
The command produced is expected to read the data lines from stdin
and to write its output to stdout.
-help Display this list of options
--help Display this list of options
See our contribution guidelines at https://github.com/mjambon/documents/blob/master/how-to-contribute.md