Packages in Erlang

Introduction

Packages are simply namespaces for modules. All old Erlang modules automatically belong to the top level ("empty-string") namespace, and do not need any changes.

The full name of a packaged module is written as e.g. "fee.fie.foe.foo", i.e., as atoms separated by periods, where the package name is the part up to but not including the last period; in this case "fee.fie.foe". A more concrete example is the module erl.lang.term, which is in the package erl.lang. Package names can have any number of segments, as in erl.lang.list.sort. The atoms in the name can be quoted, as in foo.'Bar'.baz, or even the whole name, as in 'foo.bar.baz' but the concatenation of atoms and periods must not contain two consecutive period characters or end with a period, as in 'foo..bar', foo.'.bar', or foo.'bar.'. The periods must not be followed by whitespace.

The code loader maps module names onto the file system directory structure. E.g., the module erl.lang.term corresponds to a file .../erl/lang/term.beam in the search path. Note that the name of the actual object file corresponds to the last part only of the full module name. (Thus, old existing modules such as lists simply map to .../lists.beam, exactly as before.)

A packaged module in a file "foo/bar/fred.erl" is declared as:

This can be compiled and loaded from the Erlang shell using c(fred), if your current directory is the same as that of the file. The object file will be named fred.beam.

The Erlang search path works exactly as before, except that the package segments will be appended to each directory in the path in order to find the file. E.g., assume the path is ["/usr/lib/erl", "/usr/local/lib/otp/legacy/ebin", "/home/barney/erl"]. Then, the code for a module named foo.bar.fred will be searched for first as "/usr/lib/erl/foo/bar/fred.beam", then "/usr/local/lib/otp/legacy/ebin/foo/bar/fred.beam" and lastly "/home/barney/erl/foo/bar/fred.beam". A module like lists, which is in the top-level package, will be looked for as "/usr/lib/erl/lists.beam", "/usr/local/lib/otp/legacy/ebin/lists.beam" and "/home/barney/erl/lists.beam".

Programming

Normally, if a call is made from one module to another, it is assumed that the called module belongs to the same package as the source module. The compiler automatically expands such calls. E.g., in:
    -module(foo.bar.m1).
    -export([f/1]).

    f(X) -> m2:g(X).
m2:g(X) becomes a call to foo.bar.m2

If this is not what was intended, the call can be written explicitly, as in

    -module(foo.bar.m1).
    -export([f/1]).

    f(X) -> fee.fie.foe.m2:g(X).
Because the called module is given with an explicit package name, no expansion is done in this case.

If a module from another package is used repeatedly in a module, an import declaration can make life easier:

    -module(foo.bar.m1).
    -export([f/1, g/1]).
    -import(fee.fie.foe.m2).

    f(X) -> m2:g(X).
    g(X) -> m2:h(X).
will make the calls to m2 refer to fee.fie.foe.m2. More generally, a declaration -import(Package.Module). will cause calls to Module to be expanded to Package.Module.

Old-style function imports work as normal (but full module names must be used); e.g.:

    -import(fee.fie.foe.m2, [g/1, h/1]).
however, it is probably better to avoid this form of import altogether in new code, since it makes it hard to see what calls are really "remote".

If it is necessary to call a module in the top-level package from within a named package, the module name can be written either with an initial period as in e.g. ".lists", or with an empty initial atom, as in "''.lists". However, the best way is to use an import declaration - this is most obvious to the eye, and makes sure we don't forget adding a period somewhere:

    -module(foo.bar.fred).
    -export([f/1]).
    -import(lists).

    f(X) -> lists:reverse(X).
The dot-syntax for module names can be used in any expression. All segments must be constant atoms, and the result must be a well-formed package/module name. E.g.:
    spawn(foo.bar.fred, f, [X])
is equivalent to spawn('foo.bar.fred', f, [X]).

The Erlang Shell

The shell also automatically expands remote calls, however currently no expansions are made by default. The user can change the behaviour by using the import/1 shell function (or its abbreviation use/1). E.g.:
        1> import(foo.bar.m).
        ok
        2> m:f().
will evaluate foo.bar.m:f(). If a new import is made of the same name, this overrides any previous import. (It is likely that in the future, some system packages will be pre-imported.)

In addition, the function import_all/1 (and its alias use_all/1) imports all modules currently found in the path for a given package name. E.g., assuming the files ".../foo/bar/fred.beam", ".../foo/bar/barney.beam" and ".../foo/bar/bambam.beam" can be found from our current path,

        1> import_all(foo.bar).
will make fred, barney and bambam expand to foo.bar.fred, foo.bar.barney and foo.bar.bambam, respectively.

Note: The compiler does not have an "import all" directive, for the reason that Erlang has no compile time type checking. E.g. if the wrong search path is used at compile time, a call m:f(...) could be expanded to foo.bar.m:f(...) without any warning, instead of the intended frob.ozz.m:f(...), if package foo.bar happens to be found first in the path. Explicitly declaring each use of a module makes for safe code.