Note that it is possible to have __DATA__ tokens in the same package in multiple files, and that the last __DATA__ token in a given package that is encountered by the compiler is the one accessible by the filehandle. This also applies to __END__ and main, i.e. if the 'main' program has an __END__, but a module 'require'd (_not_ 'use'd) by that program has a 'package main;' declaration followed by an '__DATA__', then the DATA filehandle is set to access the data after the __DATA__ in the module, _not_ the data after the __END__ token in the 'main' program, since the compiler encounters the 'require'd file later.
The SelfLoader will read from the FOOBAR::DATA filehandle to load in the data after __DATA__, and load in any subroutine when it is called. The costs are the one-time parsing of the data after __DATA__, and a load delay for the _first_ call of any autoloaded function. The benefits (hopefully) are a speeded up compilation phase, with no need to load functions which are never used.
The SelfLoader will stop reading from __DATA__ if it encounters the __END__ token - just as you would expect. If the __END__ token is present, and is followed by the token DATA, then the SelfLoader leaves the FOOBAR::DATA filehandle open on the line after that token.
The SelfLoader exports the AUTOLOAD subroutine to the package using the SelfLoader, and this loads the called subroutine when it is first called.
There is no advantage to putting subroutines which will _always_ be called after the __DATA__ token.
So specifically, autoloaded functions cannot see package lexicals (this applies to both the SelfLoader and the Autoloader).
There is no need to inherit from the SelfLoader.
The SelfLoader works similarly to the AutoLoader, but picks up the subs from after the __DATA__ instead of in the 'lib/auto' directory. There is a maintainance gain in not needing to run AutoSplit on the module at installation, and a runtime gain in not needing to keep opening and closing files to load subs. There is a runtime loss in needing to parse the code after the __DATA__.
Data after the __DATA__ token in a module is read using the FOOBAR::DATA filehandle. __END__ can still be used to denote the end of the __DATA__ section if followed by the token DATA - this is supported by the SelfLoader. The FOOBAR::DATA filehandle is left open if an __END__ followed by a DATA is found, with the filehandle positioned at the start of the line after the __END__ token. If no __END__ token is present, or an __END__ token with no DATA token on the same line, then the filehandle is closed.
The SelfLoader reads from wherever the current position of the FOOBAR::DATA filehandle is, until the EOF or __END__. This means that if you want to use that filehandle (and ONLY if you want to), you should either
1. Put all your subroutine declarations immediately after the __DATA__ token and put your own data after those declarations, using the __END__ token to mark the end of subroutine declarations. You must also ensure that the SelfLoader reads first by calling 'SelfLoader->load_stubs();', or by using a function which is selfloaded;
2. You should read the FOOBAR::DATA filehandle first, leaving the handle open and positioned at the first line of subroutine declarations.
You could conceivably do both.
A subroutine stub (or forward declaration) looks like
i.e. it is a subroutine declaration without the body of the subroutine. For modules which are not classes, there is no real need for stubs as far as autoloading is concerned.
For modules which ARE classes, and need to handle inherited methods, stubs are needed to ensure that the method inheritance mechanism works properly. You can load the stubs into the module at 'require' time, by adding the statement 'SelfLoader->load_stubs();' to the module to do this.
The alternative is to put the stubs in before the __DATA__ token BEFORE releasing the module, and for this purpose the Devel::SelfStubber module is available. However this does require the extra step of ensuring that the stubs are in the module. If this is done I strongly recommend that this is done BEFORE releasing the module - it should NOT be done at install time in general.
Fully qualified subroutine names are also supported. For example,
will all be loaded correctly by the SelfLoader, and the SelfLoader will ensure that the packages 'foo' and 'baz' correctly have the SelfLoader AUTOLOAD method when the data after __DATA__ is first parsed.