Apart from discussing the C interface provided by Perl for writing callbacks the document uses a series of examples to show how the interface actually works in practice. In addition some techniques for coding callbacks are covered.
Examples where callbacks are necessary include
A fairly common feature in applications is to allow you to define a C function that will be called whenever something nasty occurs. What we would like is to be able to specify a Perl subroutine that will be called instead.
Before you launch yourself head first into the rest of this document, it would be a good idea to have read the following two documents - the perlxs manpage and the perlguts manpage .
Perl has a number of C functions that allow you to call Perl subroutines. They are
The key function is perl_call_sv . All the other functions are fairly simple wrappers which make it easier to call Perl subroutines in special cases. At the end of the day they will all call perl_call_sv to actually invoke the Perl subroutine.
All the perl_call_* functions have a flags parameter which is
used to pass a bit mask of options to Perl. This bit mask operates
identically for each of the functions. The settings available in the
bit mask are discussed in
FLAG VALUES
.
Each of the functions will now be discussed in turn.
sv, is an SV*.
This allows you to specify the Perl subroutine to be called either as a
C string (which has first been converted to an SV) or a reference to a
subroutine. The section,
Using perl_call_sv
, shows how you can make
use of
perl_call_sv
.
``pkg::fred''.
methname corresponds to the name of the method
to be called. Note that the class that the method belongs to is passed
on the Perl stack rather than in the parameter list. This class can be
either the name of the class (for a static method) or a reference to an
object (for a virtual method). See the perlobj manpage
for more information on
static and virtual methods and
Using perl_call_method
for an example
of using
perl_call_method
.
subname parameter. It also takes the usual flags
parameter. The final parameter, argv, consists of a NULL terminated
list of C strings to be passed as parameters to the Perl subroutine.
See
Using perl_call_argv
.
As a general rule you should always check the return value from these functions. Even if you are expecting only a particular number of values to be returned from the Perl subroutine, there is nothing to stop someone from doing something unexpected - don't say you haven't been warned.
flags parameter in all the perl_call_* functions is a bit mask
which can consist of any combination of the symbols defined below,
OR'ed together.
This flag has 2 effects:
If 0, then you have specified the G_DISCARD flag.
If 1, then the item actually returned by the Perl subroutine will be
stored on the Perl stack - the section
Returning a Scalar
shows how
to access this value on the stack. Remember that regardless of how
many items the Perl subroutine returns, only the last one will be
accessible from the stack - think of the case where only one value is
returned as being a list with only one element. Any other items that
were returned will not exist by the time control returns from the
perl_call_* function. The section I
As with G_SCALAR, this flag has 2 effects:
If 0, then you have specified the G_DISCARD flag.
If not 0, then it will be a count of the number of items returned by the subroutine. These items will be stored on the Perl stack. The section Returning a list of values gives an example of using the G_ARRAY flag and the mechanics of accessing the returned items from the Perl stack.
If you do not set this flag then it is very important that you make
sure that any temporaries (i.e. parameters passed to the Perl
subroutine and values returned from the subroutine) are disposed of
yourself. The section
Returning a Scalar
gives details of how to
explicitly dispose of these temporaries and the section I
@_ array for the Perl subroutine.
Although the functionality provided by this flag may seem straightforward, it should be used only if there is a good reason to do so. The reason for being cautious is that even if you have specified the G_NOARGS flag, it is still possible for the Perl subroutine that has been called to think that you have passed it parameters.
In fact, what can happen is that the Perl subroutine you have called
can access the @_ array from a previous Perl subroutine. This will
occur when the code that is executing the perl_call_* function has
itself been called from another Perl subroutine. The code below
illustrates this
This will print
What has happened is that fred accesses the @_ array which
belongs to joe.
Whenever control returns from the perl_call_* function you need to
check the $@ variable as you would in a normal Perl script.
The value returned from the perl_call_* function is dependent on what other flags have been specified and whether an error has occurred. Here are all the different cases that can occur:
$@ and
you want the program to continue, you must remember to pop the undef
from the stack.
$@ variable and set it to a string describing
the error iff there was an error in the called code. This unqualified
resetting of $@ can be problematic in the reliable identification of
errors using the eval {} mechanism, because the possibility exists
that perl will call other code (end of block processing code, for
example) between the time the error causes $@ to be set within
eval {}, and the subsequent statement which checks for the value of
$@ gets executed in the user's script.
This scenario will mostly be applicable to code that is meant to be
called from within destructors, asynchronous callbacks, signal
handlers, __DIE__ or __WARN__ hooks, and tie functions. In
such situations, you will not want to clear $@ at all, but simply to
append any new errors to any existing value of $@.
The G_KEEPERR flag is meant to be used in conjunction with G_EVAL in perl_call_* functions that are used to implement such code. This flag has no effect when G_EVAL is not used.
When G_KEEPERR is used, any errors in the called code will be prefixed
with the string ``\t(in cleanup)'', and appended to the current value
of $@.
The G_KEEPERR flag was introduced in Perl version 5.002.
See Using G_KEEPERR for an example of a situation that warrants the use of this flag.
GIMME macro. This will return
G_SCALAR
if you have been called in a scalar context and
G_ARRAY
if in an
array context. An example of using the GIMME macro is shown in
section
Using GIMME
.
Specifically, if the two flags are used when calling a subroutine and that subroutine does not call die, the value returned by perl_call_* will be wrong.
The symptom of this problem is that the called Perl sub will continue to completion, but whenever it attempts to pass control back to the XSUB, the program will immediately terminate.
For example, say you want to call this Perl sub
via this XSUB
When Call_fred is executed it will print
As control never returns to Call_fred, the ``back in Call_fred''
string will not get printed.
To work around this problem, you can either upgrade to Perl 5.002 (or later), or use the G_EVAL flag with perl_call_* as shown below
Perl provides many macros to assist in accessing the Perl stack. Wherever possible, these macros should always be used when interfacing to Perl internals. Hopefully this should make the code less vulnerable to any changes made to Perl in the future.
Another point worth noting is that in the first series of examples I have made use of only the perl_call_pv function. This has been done to keep the code simpler and ease you into the topic. Wherever possible, if the choice is between using perl_call_pv and perl_call_sv , you should always try to use perl_call_sv . See Using perl_call_sv for details.
and here is a C function to call it
Simple, eh.
A few points to note about this example.
dSP and PUSHMARK(sp) for now. They will be discussed in
the next example.
LeftString, which will take 2 parameters - a
string ($s) and an integer ($n). The subroutine will simply
print the first $n characters of the string.
So the Perl subroutine would look like this
The C function required to call LeftString would look like this.
Here are a few notes on the C function call_LeftString.
dSP and
ending with the line PUTBACK.
dSP - it declares
and initializes a local copy of the Perl stack pointer.
All the other macros which will be used in this example require you to have used this macro.
The exception to this rule is if you are calling a Perl subroutine
directly from an XSUB function. In this case it is not necessary to
explicitly use the dSP macro - it will be declared for you
automatically.
PUSHMARK and PUTBACK macros. The purpose of these two macros, in
this context, is to automatically count the number of parameters you
are pushing. Then whenever Perl is creating the @_ array for the
subroutine, it knows how big to make it.
The PUSHMARK macro tells Perl to make a mental note of the current
stack pointer. Even if you aren't passing any parameters (like the
example shown in the section
No Parameters, Nothing returned
) you
must still call the PUSHMARK macro before you can call any of the
perl_call_* functions - Perl still needs to know that there are no
parameters.
The PUTBACK macro sets the global copy of the stack pointer to be
the same as our local copy. If we didn't do this
perl_call_pv
wouldn't know where the two parameters we pushed were - remember that
up to now all the stack pointer manipulation we have done is with our
local copy, not the global copy.
See the ``XSUBs and the Argument Stack'' for details on how the XPUSH macros work.
Here is a Perl subroutine, Adder, which takes 2 integer parameters and simply returns their sum.
Since we are now concerned with the return value from Adder, the C function required to call it is now a bit more complex.
Points to note this time are
@_
array will be created and that the value returned by Adder will
still exist after the call to
perl_call_pv
.
at the start of the function, and
at the end. The ENTER/SAVETMPS pair creates a boundary for any
temporaries we create. This means that the temporaries we get rid of
will be limited to those which were created after these calls.
The FREETMPS/LEAVE pair will get rid of any values returned by
the Perl subroutine, plus it will also dump the mortal SV's we have
created. Having ENTER/SAVETMPS at the beginning of the code
makes sure that no other mortals are destroyed.
Think of these macros as working a bit like using { and } in Perl
to limit the scope of local variables.
See the section Using Perl to dispose of temporaries for details of an alternative to using these macros.
SPAGAIN is to refresh the local copy of the
stack pointer. This is necessary because it is possible that the memory
allocated to the Perl stack has been re-allocated whilst in the
perl_call_pv
call.
If you are making use of the Perl stack pointer in your code you must always refresh the your local copy using SPAGAIN whenever you make use of the perl_call_* functions or any other Perl internal function.
Expecting a single value is not quite the same as knowing that there will be one. If someone modified Adder to return a list and we didn't check for that possibility and take appropriate action the Perl stack would end up in an inconsistent state. That is something you really don't want to ever happen.
POPi macro is used here to pop the return value from the stack.
In this case we wanted an integer, so POPi was used.
Here is the complete list of POP macros available, along with the types they return.
PUTBACK is used to leave the Perl stack in a consistent
state before exiting the function. This is necessary because when we
popped the return value from the stack with POPi it updated only our
local copy of the stack pointer. Remember, PUTBACK sets the global
stack pointer to be the same as our local copy.
Here is the Perl subroutine
and this is the C function
If call_AddSubtract is called like this
then here is the output
Notes
POPi is used twice this time because we were
retrieving 2 values from the stack. The important thing to note is that
when using the POP* macros they come off the stack in reverse
order.
The other modification made is that call_AddSubScalar will print the number of items returned from the Perl subroutine and their value (for simplicity it assumes that they are integer). So if call_AddSubScalar is called
then the output will be
In this case the main point to note is that only the last item in the list returned from the subroutine, Adder actually made it back to call_AddSubScalar.
The Perl subroutine, Inc, below takes 2 parameters and increments each directly.
and here is a C function to call it.
To be able to access the two parameters that were pushed onto the stack
after they return from
perl_call_pv
it is necessary to make a note
of their addresses - thus the two variables sva and svb.
The reason this is necessary is that the area of the Perl stack which held them will very likely have been overwritten by something else by the time control returns from perl_call_pv .
and some C to call it
If call_Subtract is called thus
the following will be printed
Notes
is the direct equivalent of this bit of Perl
errgv is a perl global of type GV * that points to the
symbol table entry containing the error. GvSV(errgv) therefore
refers to the C equivalent of $@.
POPs in the block where
SvTRUE(GvSV(errgv)) is true. This is necessary because whenever a
perl_call_* function invoked with G_EVAL|G_SCALAR returns an error,
the top of the stack holds the value undef. Since we want the
program to continue after detecting this error, it is essential that
the stack is tidied up by removing the undef.
This example will fail to recognize that an error occurred inside the
eval {}. Here's why: the call_Subtract code got executed while perl
was cleaning up temporaries when exiting the eval block, and since
call_Subtract is implemented with
perl_call_pv
using the G_EVAL
flag, it promptly reset $@. This results in the failure of the
outermost test for $@, and thereby the failure of the error trap.
Appending the G_KEEPERR flag, so that the perl_call_pv call in call_Subtract reads:
will preserve the error and restore reliable error handling.
Consider the Perl code below
Here is a snippet of XSUB which defines CallSubPV.
That is fine as far as it goes. The thing is, the Perl subroutine can be specified only as a string. For Perl 4 this was adequate, but Perl 5 allows references to subroutines and anonymous subroutines. This is where perl_call_sv is useful.
The code below for CallSubSV is identical to CallSubPV except
that the name parameter is now defined as an SV* and we use
perl_call_sv
instead of
perl_call_pv
.
Since we are using an SV to call fred the following can all be used
As you can see, perl_call_sv gives you much greater flexibility in how you can specify the Perl subroutine.
You should note that if it is necessary to store the SV (name in the
example above) which corresponds to the Perl subroutine so that it can
be used later in the program, it not enough to just store a copy of the
pointer to the SV. Say the code above had been like this
The reason this is wrong is that by the time you come to use the
pointer rememberSub in CallSavedSub1, it may or may not still refer
to the Perl subroutine that was recorded in SaveSub1. This is
particularly true for these cases
By the time each of the SaveSub1 statements above have been executed,
the SV*'s which corresponded to the parameters will no longer exist.
Expect an error message from Perl of the form
for each of the CallSavedSub1 lines.
Similarly, with this code
you can expect one of these messages (which you actually get is dependent on the version of Perl you are using)
The variable $ref may have referred to the subroutine fred
whenever the call to SaveSub1 was made but by the time
CallSavedSub1 gets called it now holds the number 47. Since we
saved only a pointer to the original SV in SaveSub1, any changes to
$ref will be tracked by the pointer rememberSub. This means that
whenever CallSavedSub1 gets called, it will attempt to execute the
code which is referenced by the SV* rememberSub. In this case
though, it now refers to the integer 47, so expect Perl to complain
loudly.
A similar but more subtle problem is illustrated with this code
This time whenever CallSavedSub1 get called it will execute the Perl
subroutine joe (assuming it exists) rather than fred as was
originally requested in the call to SaveSub1.
To get around these problems it is necessary to take a full copy of the
SV. The code below shows SaveSub2 modified to do that
In order to avoid creating a new SV every time SaveSub2 is called,
the function first checks to see if it has been called before. If not,
then space for a new SV is allocated and the reference to the Perl
subroutine, name is copied to the variable keepSub in one
operation using newSVsv. Thereafter, whenever SaveSub2 is called
the existing SV, keepSub, is overwritten with the new value using
SvSetSV.
and here is an example of perl_call_argv which will call PrintList.
Note that it is not necessary to call PUSHMARK in this instance.
This is because
perl_call_argv
will do it for you.
It just implements a very simple class to manage an array. Apart from
the constructor, new, it declares methods, one static and one
virtual. The static method, PrintID, simply prints out the class
name and a version number. The virtual method, Display, prints out a
single element of the array. Here is an all Perl example of using it.
will print
Calling a Perl method from C is fairly straightforward. The following things are required
PrintID and Display methods from C.
So the methods PrintID and Display can be invoked like this
The only thing to note is that in both the static and virtual methods, the method name is not passed via the stack - it is used as the first parameter to perl_call_method .
and here is some Perl to test it
The output from that will be
ENTER/SAVETMPS -
FREETMPS/LEAVE pairing.
sequence in the callback (and not, of course, specifying the G_DISCARD flag).
If you are going to use this method you have to be aware of a possible memory leak which can arise under very specific circumstances. To explain these circumstances you need to know a bit about the flow of control between Perl and the callback routine.
The examples given at the start of the document (an error handler and an event driven program) are typical of the two main sorts of flow control that you are likely to encounter with callbacks. There is a very important distinction between them, so pay attention.
In the first example, an error handler, the flow of control could be as follows. You have created an interface to an external library. Control can reach the external library like this
Whilst control is in the library, an error condition occurs. You have previously set up a Perl callback to handle this situation, so it will get executed. Once the callback has finished, control will drop back to Perl again. Here is what the flow of control will be like in that situation
After processing of the error using perl_call_* is completed, control reverts back to Perl more or less immediately.
In the diagram, the further right you go the more deeply nested the scope is. It is only when control is back with perl on the extreme left of the diagram that you will have dropped back to the enclosing scope and any temporaries you have left hanging around will be freed.
In the second example, an event driven program, the flow of control will be more like this
In this case the flow of control can consist of only the repeated sequence
for the practically the complete duration of the program. This means that control may never drop back to the surrounding scope in Perl at the extreme left.
So what is the big problem? Well, if you are expecting Perl to tidy up those temporaries for you, you might be in for a long wait. For Perl to actually dispose of your temporaries, control must drop back to the enclosing scope at some stage. In the event driven scenario that may never happen. This means that as time goes on, your program will create more and more temporaries, none of which will ever be freed. As each of these temporaries consumes some memory your program will eventually consume all the available memory in your system - kapow!
So here is the bottom line - if you are sure that control will revert back to the enclosing Perl scope fairly quickly after the end of your callback, then it isn't absolutely necessary to explicitly dispose of any temporaries you may have created. Mind you, if you are at all uncertain about what to do, it doesn't do any harm to tidy up anyway.
Potentially one of the trickiest problems to overcome when designing a callback interface can be figuring out how to store the mapping between the C callback function and the Perl equivalent.
To help understand why this can be a real problem first consider how a
callback is set up in an all C environment. Typically a C API will
provide a function to register a callback. This will expect a pointer
to a function as one of its parameters. Below is a call to a
hypothetical function register_fatal which registers the C function
to get called when a fatal error occurs.
The single parameter cb1 is a pointer to a function, so you must
have defined cb1 in your code, say something like this
Now change that to call a Perl subroutine instead
where the Perl equivalent of register_fatal and the callback it
registers, pcb1, might look like this
The mapping between the C callback and the Perl equivalent is stored in
the global variable callback.
This will be adequate if you ever need to have only 1 callback
registered at any time. An example could be an error handler like the
code sketched out above. Remember though, repeated calls to
register_fatal will replace the previously registered callback
function with the new one.
Say for example you want to interface to a library which allows asynchronous file i/o. In this case you may be able to register a callback whenever a read operation has completed. To be of any use we want to be able to call separate Perl subroutines for each file that is opened. As it stands, the error handler example above would not be adequate as it allows only a single callback to be defined at any time. What we require is a means of storing the mapping between the opened file and the Perl subroutine we want to be called for that file.
Say the i/o library has a function asynch_read which associates a C
function ProcessRead with a file handle fh - this assumes that it
has also provided some routine to open the file and so obtain the file
handle.
This may expect the C ProcessRead function of this form
To provide a Perl interface to this library we need to be able to map
between the fh parameter and the Perl subroutine we want called. A
hash is a convenient mechanism for storing this mapping. The code
below shows a possible implementation
and asynch_read_if could look like this
For completeness, here is asynch_close. This shows how to remove
the entry from the hash Mapping.
So the Perl interface would look like this
The mapping between the C callback and Perl is stored in the global
hash Mapping this time. Using a hash has the distinct advantage that
it allows an unlimited number of callbacks to be registered.
What if the interface provided by the C callback doesn't contain a
parameter which allows the file handle to Perl subroutine mapping? Say
in the asynchronous i/o package, the callback function gets passed only
the buffer parameter like this
Without the file handle there is no straightforward way to map from the C callback to the Perl subroutine.
In this case a possible way around this problem is to pre-define a series of C functions to act as the interface to Perl, thus
In this case the functions fn1, fn2 and fn3 are used to
remember the Perl subroutine to be called. Each of the functions holds
a separate hard-wired index which is used in the function Pcb to
access the Map array and actually call the Perl subroutine.
There are some obvious disadvantages with this technique.
Firstly, the code is considerably more complex than with the previous example.
Secondly, there is a hard-wired limit (in this case 3) to the number of callbacks that can exist simultaneously. The only way to increase the limit is by modifying the code to add more functions and then re-compiling. None the less, as long as the number of functions is chosen with some care, it is still a workable solution and in some cases is the only one available.
To summarize, here are a number of possible methods for you to consider for storing the mapping between C and the Perl callback
Although I have made use of only the POP* macros to access values
returned from Perl subroutines, it is also possible to bypass these
macros and read the stack using the ST macro (See the perlxs manpage
for a
full description of the ST macro).
Most of the time the POP* macros should be adequate, the main
problem with them is that they force you to process the returned values
in sequence. This may not be the most suitable way to process the
values in some cases. What we want is to be able to access the stack in
a random order. The ST macro as used when coding an XSUB is ideal
for this purpose.
The code below is the example given in the section IPOP*.
Notes
ax. This is
because the ST macro expects it to exist. If we were in an XSUB it
would not be necessary to define ax as it is already defined for
you.
sets the stack up so that we can use the ST macro.
ST(0) refers to the
first value returned by the Perl subroutine and ST(count-1)
refers to the last.
Special thanks to the following people who assisted in the creation of the document.
Jeff Okamoto, Tim Bunce, Nick Gianniotis, Steve Kelem, Gurusamy Sarathy and Larry Wall.