[% setvar title Omnibus Structured Exception/Error Handling Mechanism %]
Note: these documents may be out of date. Do not use as reference! |
To see what is currently happening visit http://www.perl6.org/
Omnibus Structured Exception/Error Handling Mechanism
Maintainer: Tony Olekshy <olekshy@avrasoft.com> Date: 8 Aug 2000 Last Modified: 1 Oct 2000 Mailing List: perl6-language-errors@perl.org Number: 88 Version: 3 Status: Frozen
RFC 88 as HTML www.avrasoft.com RFC 88 as Text www.avrasoft.com RFC 88 as POD www.avrasoft.com Perl 5 Try.pm www.avrasoft.com Regression Test www.avrasoft.com
"The Encyclopedia of Software Engineering" [ESE-1994] says (p.847):
Inevitably, no matter how carefully a programmer behaves when writing a program and no matter how thoroughly its verification is carried out, errors remain in the program and program execution may result in a failure. [...] The programming language may provide a framework for detecting and then handling faults, so that the program either fails gracefully or continues to work after some remedial action has been taken to recover from the error. Such a linguistic framework is usually called exception handling.
The structured exception handling mechanism described in this RFC satisfies the following requirements.
It does not, by default, interfere with traditional Perl programming styles. When explicitly used, it simply adds functionality to better support other programming styles, which currently have to undertake some contortions to get the desired effect.
It is simple in the cases in which it is commonly used.
It is capable enough to handle the needs of production applications, frameworks, and modules, including the ability to hook into the mechanism itself.
It is suitable for "error" handling via exceptions.
It is suitable for light-weight exceptions that may not involve errors at all.
This RFC describes a collection of changes and additions to Perl, which together support a built-in base class for Exception objects, and exception/error handling code like this:
exception 'Alarm'; try { throw Alarm "a message", tag => "ABC.1234", ... ; } catch Alarm => { ... } catch Error::DB, Error::IO => { ... } catch $@ =~ /divide by 0/ => { ... } catch { ... } finally { ... }
Any exceptions that are raised within an enclosing try, catch, or finally block, where the enclosing block can be located anywhere up the subroutine call stack, are trapped and processed according to the semantics described in this RFC.
The new built-in Exception base class is designed to be used by Perl for raising exceptions for failed operators or functions, but this RFC can be used with the Exception base class whether or not that happens.
Readers who are not familiar with the technique of using exception handling to handle errors should refer to the CONVERSION section of this document first.
It is not the intent of this RFC to interfere with traditional Perl scripts; the intent is only to facilitate the availability of a more controllable, pragmatic, and yet robust mechanism when such is found to be appropriate.
eval {die "Can't foo."}; print $@;
continues to work as before.try
, throw
, catch
, or finally
at
all, if one doesn't want to.raise
An exception is raised to begin call-stack unwinding according to the
semantics described herein. This provides for controlled non-local
flow-control. This is what die
does.
propagate
The passing of an exception up the call stack for further processing is called propagation. Raising an exception starts propagation. Propagation stops when the exception is trapped.
unwinding
The overall process of handling the propagation of an exception, from the point it is raised until the point it is trapped, is called unwinding.
trap
The termination of unwinding, for the purpose of attempting further
processing using local flow-control semantics, is called trapping.
This is what eval
does, as do try
, catch
, and finally
.
cleanly caught
This means the trapping and handling of an exception did not itself
raise an exception. Cleanly handling exceptions raised while handling
exceptions is difficult, tedious, and error-prone given only eval
.
exception
An exception is a collection of informaton about a particular non-local
goto, captured at raise-time, for use by trap-time handling based on
said informaton. This is what $@
is. It can be any structured data;
a reference to a blessed hash works well in practice, uses minimal OO
concepts, and can be made to stringify just like the Perl 5 $@
.
error
A fuzzy concept, an error is essentially an exception with a negative connotation. A traditional definition might be, "an exception raised to signal an assertion failure that typically indicates the inability of an algorithm to complete the request it has been given". Ultimately, whether or not an exception is considered to be an error depends on the trapper, not the raiser.
The most common forms of structured exception handling are straight- forward. Here they are:
try { ... } catch { ... }
Invoke the catch block only if the try block raises an exception (aka dies), otherwise, there is nothing to catch.
Continue unwinding (propagate $@
) only if the catch block was invoked
and it raises an exception too.
Otherwise, execute the next statement according to local flow-control, because either no exception has been raised by either block, or one was raised in try but it was cleanly caught.
try { ... } finally { ... }
Invoke the finally block whether or not the try block raises an exception.
Continue unwinding (propagate $@
) if either the try block or the
finally block raised an exception.
Otherwise, execute the next statement according to local flow control, because no exception has been raised.
try { ... } catch Exception::Foo => { ... } catch Exception::Bar => { ... } catch { ... } # everything else finally { ... }
In this case if the try raises an exception then the first matching catch is invoked, and then whether or not the try (or any catch) raised any exception, the finally is invoked.
Once a catch clause matches, subsequent catch clauses are skipped (like elsif/elsif/else) up to the next finally clause.
This try statement unwinds only if try raises an exception and the exception's matching catch block raises an exception, or finally raises an exception. Otherwise no-one raised an exception, or it was in the try and it was cleanly caught.
This means that all exceptions propagate unless they are cleanly caught, just as in Perl 5. To prevent this, use:
try { fragile(); } catch { } # Go on no matter what.
More complicated constructs should be avoided unless the rules in "Unwinding Semantics" make sense to you and your target audience.
throw Exception::IO "a message", tag => "ABC.1234", ... ;
Throw is both a class method and an instance method of the built-in Exception class. The indirect object syntax is used to make the throw imperative. As a class method, it is syntactic sugar for:
die Exception::IO->new( message => "a message", tag => "ABC.1234", ...);
As an instance method it is syntactic sugar for copying over any values
given as arguments, and then effecting die $self
. This allows
throw $@
to be used to re-raise exceptions.
Note that a derived class can override its constructor to preprocess the optional arguments, so that (for example) tags are parsed out of the message, which allows something like this to work for developers who prefer it (such as the author):
throw MyError "ABC.1234: A message.";
This also illustrates why the message is a required argument to the
throw method. It should not have to be more complicated than that to
raise an exception of a given type with a given annotation, in common
use. One should not have to always add message =
> just for that.
If throw
is followed by a <string>, rather than a class, like this:
throw "A Message.", ...;
it is the equivalent of:
throw Exception "A Message.", ...;
try { ... } catch <test> => { ... } finally { ... }
A try statement starts with a block and is followed by zero or more catch and/or finally clauses.
The <test> argument of the catch clause is optional, and is described below.
try
, catch
, and finally
blocks should share the same lexical
scope, in the way that while
and continue
do. This is so that
variables defined in the try
block can be operated on in the other
blocks, which allows one to say:
try { my $fh = open $file } finally { $fh and close $fh }
Note that try
is a keyword, not a function. This is so that a
;
is not needed at the end of the last block. This is because a
try/catch/finally now looks more like an if/elsif/else, which does not
require such a ;
, than like an eval, which does).
catch { ... }
Traps all exceptions, according to the unwind semantics described below.
It is a syntax error for a catch all clause like this to be immediately followed by another catch clause, because that would be dead code that could never be executed.
Otherwise, it is syntactic sugar for:
catch 1 => { ... }
To signal failure in a catch block, throw an exception.
catch Exception::DB => { ... }
When catch is followed by a class name, the catch block is invoked only if the current error is an instance of said class. It is syntactic sugar for:
catch $@->isa($string) => { ... }
catch Exception::DB, Exception::IO => { ... }
When catch is followed by a comma-separated list of class names, the catch block is invoked only if the current is an instance of one of the given classes. It is syntactic sugar for:
catch grep { $@->isa($_) } @list => { ... }
catch <expr> => { ... }
Traps exceptions for which the expression <expr> returns true, when evaluated at the time the catch clause is attempted, according to the unwinding semantics described below.
If multiple statements are required to compute the value of the
<expr>, use this form: catch do { a; b } => { ... }
In order to prevent developers from accidentally writing
catch "Exception:DB" => { ... }
(which would be seen as a
test that is always true, while they almost certainly were looking
for a class name), the compiler should detect that special literal
case and issue a compilation warning.
finally { ... }
Once the try block is entered, every finally block is guaranteed to be entered before the try statement completes, whether or not any exceptions have been raised since the try block was entered.
die
If passed a single argument that isa Exception
, raise it as the
new exception and dies in the fashion that Perl 5 does.
Otherwise, the arguments are stringified and joined with ''
(as
in Perl 5), the resulting string is wrapped up in a new Exception
object (setting the message
instance variable to said string),
the original unstringified arguments are saved in a list ref in the
object's data
instance variable, and the new Exception
object
is raised.
The above guarantees that $@ isa Exception
(so developers can
depend on that when writing short conditional catch expressions),
but because of Exception object stringification, in simple scripts
one can still write:
open F, $file or die "Can't open $file";
When an exception $e
is raised, the equivalent of the following
is automatically done, using the variables described below:
unshift @@, $@ = $e;
This provides the mechanism we use to keep track of raised exceptions while unwinding.
$@ and @@
$@
contains the current exception, and @@
contains the current
exception stack, as defined above under die
. The unshift
rule
guarantees that $@ == $@[0]
.
$@
and @@
can only be set by die, which guarantees that $@
and $@[$i]
are instances of Exception, as described above.
Unlike eval, try does not start by clearing $@
. The successful
completion of a catch clause clears $@
and @@
.
Tracking @@
information is going to become more important if Perl
starts using exceptions for error handling. For example, if an open
throws and then a calling DB module throws, and then your UI catches it,
you are (in many cases) going to want to know why the open threw.
eval
exception 'Exception::Foo::Bar';
Makes Exception::Foo::Bar
into a class that inherits from the
built-in Exception
class, something like:
@Exception::Foo::Bar::ISA = 'Exception::Foo';
exception 'MyError::App::DB::Foo';
Makes MyError::App::DB::Foo
into a class that inherits from the
built-in Exception
class.
If the given name matches /::/
, something like this happens:
@MyError::App::DB::Foo::ISA = 'MyError::App::DB';
and all non-existent parent classes are automatically created as inheriting from their parent, or Exception in the tail case. If a parent class is found to exist and not inherit from Exception, a run-time error exception is raised.
If the given name does not match /::/
(say it's just Alarm
),
this happens instead:
@Alarm::ISA = 'Exception';
This means that every exception class isa Exception, even if
Exception::
is not used at the beginning of the class name.
The exception function can also take optional arguments, along the lines of
exception 'Error_DB', isa => "MyError::App";
which results (after error checking) in something like
@Error_DB::ISA = 'MyError::App';
Other options may possibly be given to exception
to control things
like the raise-time stack traceback.
The first three of these examples cover the most common uses of try, catch, and finally.
try { my $f = open "foo"; ... } finally { $f and close $f; }
If the open completed successfully (so defining $f
), close $f
is attempted no matter what else happens.
If anything in the try or finally clauses raises an exception, said exception is propagated after attempting the finally block. Otherwise execution continues locally.
This sort of functionality is commonly required, not just with files and close, but (for example) with objects that don't want to rely on garbage collection for the invocation of cleanup code.
try { fragile(); } catch { print "$@\n"; }
If the try block raises an exception, the catch block is entered. No exception is propagated, unless the catch block raises an exception (because if it doesn't the exception raised by try is considered "cleanly caught").
In this example, if fragile()
raises an exception it's shown on
STDOUT
, and then execution continues normally (unless print
raises
an exception). The stringification method of the exception object is
used to generate the output. Note that print $@->show;
can be
used to dump the entire exception unwind stack, as described elsewhere
herein.
try { ... } catch Exception::Foo => { ... } finally { ... }
If the try block raises an Exception::Foo
exception then the catch
block is invoked. Whether or not try or catch raised an exception, the
finally block is invoked.
If try and catch both raised exceptions, or finally raised an exception, then it is propagated, but if try raised an exception and catch and finally didn't, then no exception is propagated (because the exception was cleanly caught).
Here's another simple yet quite robust example based on the first two examples.
sub attempt_closure_after_successful_candidate_file_open { my ($closure, @fileList) = @_; local (*F); foreach my $file (@fileList) { try { open F, $file; } catch { next; } try { &$closure(*F); } finally { close F; } return; } throw Exception "Can't open any file.", debug => @fileList . " tried."; }
Most developers will usually only find need for cases like those shown above. The following examples can be used when circumstances merit, but they should be avoided by those looking for maximal simplicity. Anything can be done with state variables and nested simple trys, at least until you run off the right of the page or forget a re-throw. The following examples are provided for your convenience, not as a requirement to understanding this RFC or using its mechanism.
try { ... } catch Exception::Foo => { ... } catch Exception::Bar => { ... } catch { ... } finally { ... }
If the try block raises an exception then: if the first catch matches it is invoked, if the second catch matches it is invoked, otherwise the third catch is invoked.
The finally block is entered no matter what happens.
No exception is propagated unless one of the catch blocks raises an exception, or the finally block does (because otherwise any exception raised by try is considered to have been "cleanly caught").
try { ... } catch $@->{message} =~ /.../ => { ... }
Any exception object instance variable, such as message, tag, or
severity, can be used to test whether or not the exception should be
caught. The result of the stringification of $@
can be similarly
tested.
This also allows one to define a convenience subroutine to make, for example, the following work:
catch not &TooSevere => { ... }
try { ... } catch ref $@ =~ /.../ => { ... }
try { ... } catch grep { $_->isa("Foo") } @@ => { ... }
The catch clause is invoked only if any exception on the exception
unwind stack, aka @@
, isa Foo. Note that developers of applications
that need to search the stack a lot can define a simple convenience
subroutine to allow
catch AnyException("Error::IO") => { ... }
which is an excellent example of the utility of the
catch <expr>
form.
try { ... } catch grep { $@->isa($_) } @list => { ... }
Rather that repeating a long list of exceptions that you need to catch in multiple places, this form allows you to put them in a list. Again, in practice, a convenience subroutine would probably be defined to hold the list and compute the predicate.
try { ... } catch $@->isa("Foo") && $@->CanBar => { ... }
Derived exception classes can add new instance variables and methods to do things like test new predicates. The above form allows these predicates to be used with try. If polymorphism is desired, use
catch $@->can("CanBar") && $@->CanBar => { ... }
try { my $p = P->new; my $q = Q->new; ... } finally { $p and $p->Done; } finally { $q and $q->Done; }
This construct makes sure that if $q
is successfully constructed,
then $q->Done
is invoked even if $p->Done
raises an
exception. Works for opening and closing a pair of files too.
try { TryToFoo; } catch { TryToHandle; } finally { TryToCleanUp; } catch { throw Exception "Can't cleanly Foo."; }
Unshifts a new exception onto @@
if any of the first three blocks
throws, unless successfully handled. Use of this technique at major API
entry points can result in the availability of better information when
unwinding is eventually caught, like this:
UIM.1234: Can't add a new person to the database. APP.2345: Can't update Company relationship. DBM.3456: Trouble processing SQL UPDATE clause. DBM.4567: Unable to write to Locations table. IOM.5678: Can't open file ".../locations.ndx". IOM.6789: File ".../locations.ndx" not found.
The following table shows the details of this construct. The key is 1 = succeeds, 0 = fails, x = skipped.
TryTo TryTo TryTo | Throw After Finally Foo Handle CleanUp | Description ------------------------+-------------------------------------- 1 x 1 | No. We did Foo and cleaned up. 1 x 0 | Yes. We did Foo but didn't clean up. 0 1 1 | No. We didn't Foo but did Handle it, | and then we cleaned up. 0 1 0 | Yes. We didn't Foo but did Handle it, | but then we didn't clean up. 0 0 1 | Yes. We didn't Foo and didn't Handle | it, even though we did clean up. 0 0 0 | Yes. We didn't do anything right!
try { ... $avoidCatches_JustUnwind = predicate(); ... } catch $avoidCatches_JustUnwind => { throw $@ } catch { ... } finally { ... }
This construction allows a try to dynamically decide to avoid its catch clauses and just propagate any error that occurs in the try block (after invoking the finally block).
<exception> := exception <string> <options> ; <throw> := throw <class> <message> <options> ; | throw <string> <options> ; | throw ; <message> := # empty | <string> | <string> <options> <try> := try <block> <clauses> <clauses> := # empty | <clause> <clauses> <clause> := <catch> | <finally> <catch> := catch <block> | catch <classes> <comma> <block> | catch <expr> <comma> <block> <finally> := finally <block> <classes> := <class> | <class> , <classes> <class> # Unquoted name of a class that inherits from the # built-in Exception class, such as Exception::Foo. <options> := # empty | , <string> => <value> | , <string> => <value> <options> <block> := { ... } # A Perl code block. <comma> := , | => <expr> # A Perl expression which is evaluated when # the relevant clause is processed. <string> # A Perl expression which is evaluated and # stringified.
"Computer Architecture, Hardware and Software" [RYK-1989] says (p.347):
Three basic design issues concerning exceptions are (1) whether or not the exception is permitted to return to the point where the exception is taken, (2) how the execution context to be used during the the handler's execution is found, and (3) how the association of an exception handler with and exception event is established.
Perl's behaviour after a die
starts call-stack unwinding, as
envisioned by this RFC, is as described by the following rules.
The "current exception" is the value of $@
.
Whenever an exception is raised Perl looks for the first enclosing try/catch/finally block, or eval. If none is found program shutdown is initiated.
If a try/catch/finally clause is found Perl traps the exception and proceeds as per rule 2.
If the exception is trapped by an eval the eval returns locally,
preserving the value of $@
, according to the usual Perl 5 rules
(regardless of any outer try/catch/finally clauses).
The try block's "next" associated trap/catch or finally clause is processed according to rules 3 and 4. When there are no more clauses rule 5 is used.
If a catch <test> returns true (without itself raising an exception), its associated catch block is entered.
If the catch block is entered and it completes without itself raising an exception, the current exception and stack are cleared. But if a catch <test> or a block raises an exception, it becomes the current exception, but it does not propagate out of the try statement (at this point).
If a catch <test> raises an exception or returns true, then whether or not the catch block raises an exception, any succeeding try/catch clauses up to the next finally clause are skipped (for the purpose of the "next" iterator in rule 2). This makes sequential catches work like switch/case.
Processing then continues with rule 2.
When a finally clause is encountered its block is entered.
If the finally block raises an exception it becomes the current exception, but it does not propagate out of the statement (at this point).
Processing continues with rule 2.
After the catch and finally blocks are processed, if there is a current exception then it is re-raised and propagated as per Rule 1 (beginning above the current try statement in the call stack).
Otherwise the current exception and stack are cleared, the try statement completes normally, and Perl continues with the statement after the try statement.
Exceptions raised by the guts of Perl are envisioned by this RFC to all be instances of derived classes that inherit from Exception (so the class hierarchy can be used for exception classification). This is discussed below under IMPACT + RFC 80.
The built-in Exception class and unwinding functionality described in this RFC can be used whether or not Perl 6 goes to internal exceptions.
Instances of the actual (non-subclassed) Exception class itself are
used for simple exceptions, for those cases in which one more or
less just wants to say throw Exception "My message."
, without a
lot of extra tokens, and without getting into higher levels of the
taxonomy of exceptions.
The built-in Exception class reserves all instance variable and
method names matching /^[_a-z]/
. The following instance
variables are defined.
message
This is a description of the exception in language intended for the "end user". Potentially sensitive information should not be included here.
This instance variable is also used to record the string given when
die "Can't foo."
is used. Because of stringification (described
below), such a die
will result in
"$@" =~ /Can't foo\./
tag
This is a string which package developers can use to assign a unique
"identifier" to each exception object constructor invocation in their
package. A package-based namespace control mechanism that helps ensure
the uniqueness of tags is described below, under the tag
and
settag
methods. The tag instance variable always contains the fully
qualified tag value.
debug
This is a place for additional description that is not intended for the end user (because it is "too technical" or "sensitive").
For example, in a web application, internal file names might be considered sensitive information, but you would still like to get the name of the file that couldn't be opened into the server log.
data
If the exception is related to some particular object, or the developer wants to associate other information with this exception, said $data can be specified via:
throw Exception "...", data => $data;
trace
This is a listref containing a snapshot of the Perl call-stack as at the time the exception is first raised. The array contains hashes (one per call stack level), each containing one key-value pair for each snapshot value at that level. Here are some examples:
$e->{trace}->[0]->{file} $e->{trace}->[0]->{line} $e->{trace}->[0]->{package}
Alternatively, $e->{trace}
could be some sort of snapshot object
thingy. Similar stuff has been done by the Perl 5 Devel
bundle;
perhaps there should be a Perl 6 RFC for it.
The functionality of caller()
is used to populate this snapshot data,
but said snapshot has to be taken early and held for possible historic
debugging use, because usually we want to know where we were when things
went wrong, not where we were when we caught that (by which time, the
raise-time stack snapshot data is irrecoverable).
This snapshot data is set up by the snapshot
method described below,
so that derived classes that don't want this overhead can override that
method to do nothing. This can be important to applications that want
to use a large number of light-weight exceptions to implement non-local
success-based flow-control gotos, where the cost of taking and
maintaining the snapshots could prove to be prohibitive, especially
since they would normally never be used.
The snapshot method is an overrideable built-in rather than a stub though, because in fact in most cases one does want to pay the price for being able to debug exceptions, because said price is small (in most cases).
sysmsg
This a place for the internal exceptions raised by Perl to record
system information, along the lines of $!
, $?
, and $^E
.
Note that the sysmsg instance variable is properly the province
of the various "exceptions for built-ins" proposals, such as RFC 80.
However, if the %@
concept described under ISSUES becomes a
part of Perl 6, then sysmsg
could just be a reference to a copy
of that hash.
The built-in Exception base class defines the following methods.
new
Constructs a new object instance, using its arguments as a hash to initialize the instance variables by name.
The tag
instance variable is treated specially in order to control
the namespace for tags, as follows:
$self->{tag} = $self->settag($arg{tag});
throw
As a class method a new instance is constructed, passing any arguments
to the constructor. This object becomes $self
, and we now have an
instance method.
As an instance method, any arguments are used to update the instance
variables (unless just constructed), and this method effects
die $self
.
This method does not return to local flow control (modulo the
on_raise
mechanism described below).
overload '""'
Stringification produces a concatenation of various Exception object instance variables and certain delimiters. The message instance variable is always to be included. The details are to be worked out, but an example would be:
ABC.1234 [Warning]: Confirmed but not acknowledged.
settag
This method effects whatever operations are performed on the tag short form given in a throw like this
throw Exception "a message", tag => "My/Tag/Scheme"
in order to come up with fully qualified value to be stored in the tag instance variable.
These operations are properly the province of RFC 80, but the default could be something like this:
$self->{tag} = (ref $self) .".". $_[0];
or equivalent, as described below.
This would mean that the tags provided to exception constructors need only be unique across a class, file, or package, which is often constrained enough to make uniqueness relatively easy in practice.
The example of exception class constructor overriding given later in this RFC needs to set the tag, while maintaining namespace control, without knowing the namespace control details, so this method is provided for that.
Perhaps all this instance variable accessor method stuff can be cleaned up with other changes to the Perl 6 OO mechanism.
tag
This method returns true if the object's tag instance variable matches its argument. This allows us to easily trap by namespace-controlled tag, using a form like this:
catch $@->tag("Foo") => { ... }
The definition of "matches" can be changed, for an exception class hierarchy, by overriding this method. The default rule properly is the province of RFC 80, but it could be something like this (for unique tags across all throws of the same class):
$self->{tag} eq (ref $self) .".". $_[0]
or for unique tags across all throws in the same package:
$self->{tag} eq (caller(0))[0] ."\t".$_[0]
or for unique tags across all throws in the same file:
$self->{tag} eq (caller(0))[1] ."\t".$_[0]
show
This class method and instance method generates a string formatted
version of the exception unwinding stack based on the contents of
@@
, stringifying each exception object as it goes. For example,
print $@->show
produces something like this:
UIM.1234: Can't add a new person to database. APP.2345: Can't update Company relationship. DBM.4567: Unable to write to Company table. IOM.5678: Can't open file ".../company.db".
show
takes the following optional parameters.
label => 1
If set, the formatted exception stack is annotated with the classes of the objects therein, as in:
Exception::UI: UIM.1234: Can't add a new person to database. Exception::DB: APP.2345: Can't update Company relationship. Exception::DB: DBM.4567: Unable to write to Company table. Exception::IO: IOM.5678: Can't open file ".../company.db".
trace => 1
If set, the value returned by show
includes the Perl stack traceback
using the information from the last exception in the stack, that is
$@[-1]
. Something like this:
UIM.1234: Can't add a new person to database. APP.2345: Can't update Company relationship. DBM.4567: Unable to write to Company table. IOM.5678: Can't open file ".../company.db". Try::throw called from test-403.pl[7]. Try::try called from test-403.pl[9]. Try::try called from test-403.pl[11]. Try::try called from test-403.pl[14].
debug => 1
Annotates entries in the unwind stack with the values from their
debug
instance variables, if any. For example:
UIM.1234: Can't add a new person to database. Debug: Fred Flintstone APP.2345: Can't update Company relationship. DBM.4567: Unable to write to Company table. IOM.5678: Can't open Company file. Debug: /foo/bar/company.dat
Developers who need alternative output can override show
.
snapshot
Used internally to generate the trace
instance variable. Designed
to be overridden in derived classes for performance or extension
purposes. See the description of the trace instance variable, above.
on_raise
Derived classes may override this method to attempt to "handle" an
exception or otherwise manipulate it, just before it is raised. If
on_raise
throws or returns true the exception is raised, otherwise it
is not. An exception can be manipulated or replaced and then propagated
in modified form simply by re-raising it in on_raise
.
In the Exception
base class this method returns true, so by default
exceptions are always raised.
Unraised throws make no sense for "error" exceptions and are abhorrent
to users of exception handling for error handling, but that's ok because
Exception's on_raise
method returns true by default, you have to
actually override a method to change that behaviour, and, well, error
classes could use on_raise
for instance variable munging at
propagate-time, as long as they return true.
Nevertheless, other uses for Exception objects may want this functionality, at least based on the reference and text books which say that classically, an exception has such a handler hook.
If we do provide it, it need not be used, but if we don't provide it the functionality cannot otherwise be provided, should it be deemed desirable for some application.
on_catch
In the name of symmetric hooks, an on_catch
method is called just
before an exception object is cleared from @@
. on_catch may be
overridden by a derived exception class to perform additional debug
logging, or (in fact), to refuse the completion of the catch (by
raising a new exception).
In addition to the exception 'MyException'
mechanism described
above, custom exception and/or error classes can be created along
the following lines:
File main.pl: use Error::App; exception 'Error_DB', isa => 'Error::App'; throw Error_DB "ABC.1234: Can't foo."; File Error/App.pm: package Error::App; @Error::App::ISA = 'Error'; use Error; # It inherits from Exception. sub new { my ($C, $msg, %opt) = @_; $C = ref $C || $C; my $self = $C->SUPER::new($msg, %opt); $self->{message} =~ s/^([A-Z]+\/\d+):\s+// and $self->settag($1); return bless($self, $C); }
Note that the scope of classes like Error::App
is is limited to
packages that use it, which presumably want such functionality.
Over the last ten years, the author has come to rely on exception handling as a relatively robust mechanism for error handling, and has used exception handling to implement other, non-error related, non-local flow-control algorithms, in Scheme, C++, Visual Objects, Delphi, and Perl. He has developed a relatively complete implementation of the functionality described herein, in Perl 5, in the Try.pm module and its associated regression tests [TRY-2000]. A similar version of Try.pm is in use in production applications.
The authors' motivation is to help Perl 6 achieve a relatively robust exception handling mechanism that is suitable for error handling via exceptions, is still capable enough to handle the needs of production programs, and is still suitable for light-weight exceptions that may not involve errors at all.
"The Encyclopedia of Software Engineering" [ESE-1994] says (p.847):
Among the features offered by programming languages to support exception handling are the following.
The ability to distinguish the normal control flow from the exception handling flow to make the structure of the program clear.
The ability to detect faults as they occur in the program, not only by explicitly raising an exception but also by implicitly raising it on account of the run-time environment. [...] Both kinds of faults should be handled uniformly.
The ability to transfer control to a programmer-definable exception handler when the fault is detected. The language should specify the rules by which this detected fault is bound to its corresponding exception-handling routine.
The ability to specify how control flows after the exception handler is executed, i.e., whether one can resume execution from the point at which it left off, or whether the program should fail.
Most early programming languages do not provide specific features for exception handling, but rather use the normal constructs to implement it. [...] Obviously this and other ad hoc methods do not satisfy the requirements listed above.
To this end, new keywords have been deliberately chosen to represent the new mechanism, in order to make it clear to the developer when the code is expecting to deal with unwind semantics (rather than with local flow control).
In addition, the exception handling mechanism propagates exceptions that
are not cleanly caught, which minimizes the chances for the developer to
forget to re-raise uncaught exceptions. How many of us check for IO
failures after print
s? And if you're writing a simple program you
wouldn't want to have to, but you would want the program to shut down
after a failure even if you don't check.
Remembering to always check all subroutine and functions for failure return codes can be difficult, since nothing about the form of the call, in the source code, indicates whether or not a failure return code should be expected. And, the exception handling technique not only works with subroutines and functions, it works with operators too (which, you will note, is why divide dies on zero denominator: it has no other way to return an error code).
Although the following code using the new mechanism:
try { may_throw_1 } catch may_throw_2 => { may_throw_3 } finally { may_throw_4 }
can be written in Perl 5 like this:
eval { may_throw_1 }; my $exception = $@; if ($exception) { my $test = eval { may_throw_2 }; $@ and $exception = $@; if ( ! $@ and $test ) { eval { may_throw_3 }; $exception = $@; } } eval { may_throw_4 }; ($exception ||= $@) and die $exception;
the opportunity for flow-control errors increases.
Without the functionality provided by the mechanisms described in this RFC, instead of having a way to be able to write
throw Error_DB "ABC.1234: Can't write to table $table.";
a developer would be required to write something like
throw Exception::Error::App::DB tag => "ABC.1234", message => "Can't write to table $table.";
The latter has a much lower signal to noise ratio than the former, which is of significant importance to regular users of exception handling mechanisms.
Although the technique of using exception handing for error handling often seems foreign at first to developers who are not used to it, many find that it becomes quite natural when four concepts are kept in mind.
Wherever you previously would have return undef
or some other
special return code (or a pass-by-reference value), to indicate the
failure of a subroutine or function, instead use throw Exception
(or some fancier form thereof).
Wherever you previously would have written
$x = foo(); defined $x or return undef;
to propagate an error from a callee back to your caller, you can just
write $x = foo();
because unhandled exceptions automatically
propagate.
Wherever you previously would have written the equivalent of the following to do something about an error, and then ignore it:
$x = foo(); unless (defined $x) { # do something about error here }
you can now write
try { $x = foo(); } catch { # do something about error here }
Wherever you previously would have ignored an error in order to allow
you to restore invariants or enforce postconditions, and then used
return undef
to propagate the error, like this:
open F, ...; $x = foo(); close F; defined $x or return undef;
you can now write
open F, ...; try { $x = foo(); } finally { close F; }
because unhandled exceptions automatically propagate.
And, of course, if you don't want to use exception objects just
don't use throw
, instead just keep on using die "string";
as in Perl 5 (because $@
stringifies reasonably).
Perl should be Perl
Given that the functionality described in this RFC can be provided by a Perl 5 module, as implemented in [TRY-2000], the author would prefer that these capabilities not be made impossible in Perl 6.
Core Functionality
If the guts of Perl 6 are to use exceptions for errors, some
of the stuff in this RFC has to go into the guts of Perl 6 (but
see "Use %@
for errors from builtins", below, for a minimalist
way to do so).
The authors are of the opinion that the public interface to to the
core Perl API should support but not require the use of exceptions
for errors that would set $!
, under the control of a
use fatal;
or use strict 'fatal';
pragma.
New Keywords
RFC 88 only introduces the try
, throw
, catch
, finally
,
and exception
"keywords", which are all traditionally related to
exception handling. And they don't even need to all actually be
keywords, some can be functions (depending on the Perl 6 parser).
New keywords were chosen so that this can be written:
try { try { try { ... } finally { ... } } catch Exception1 => { ... } catch Exception2 => { ... } } catch { print $@, "\n"; }
instead of cognitively overloading existing keywords and concepts in a manner like this:
eval { eval { eval { ... } continue { ... } } else { switch ($@) { case /Exception1/ { ... } case /Exception2/ { ... } else { die $@; } } } } else { print $@, "\n"; }
because the authors are of the opinion that overloading else
and
continue
with unwind semantics not traditionally associated with
else
and continue
can be confusing, especially when intermixed
with local flow-control forms of else
and continue
(which may
be present in any { ... }
block), or when an else die $@
is forgotten on a switch
that needs to re-throw.
Some perl6-language-error discussions have suggested leaving out
the try altogether, as in simply writing { } else { }
to indicate
non-local flow-control at work. Yikes!
The try
is not for Perl's sake. It's for the developer's sake.
It says, watch out, some sort of non-local flow control is going on
here. It signals intent to deal with action at a distance
(unwinding semantics). It satisfies the first requirement listed
under MOTIVATION.
The throw
verb was chosen for raising an exception, because
of its neutral connotation (unlike fail
for example), because
exceptions do not necessarily encapsulate a negative. Similarly,
catch
is neutral, unlike try { ... } error { ... }
.
Keyword Names
Much discussion was generated on the perl6-language lists about
the selection of keywords for this mechanism. The minimalist
approach of cognitively overloading existing keywords such as
eval
, else
, and continue
has been discussed above.
The try
/ throw
/ catch
/ finally
combination works
well in consideration of the other requirements discussed herein.
For the record, here are some of other words that have either
been used by other programming languages, or suggested on the
mailing lists: raise, always, onException, when, signal, fail,
handle, otherwise, unwind, trap, quit, trip, deal, freak, panic,
cope, punt, ascend, cough, sustain, overrule, and (of course)
longjmp
.
Syntax
The comma or =>
in a conditional catch clause is required so
the expression can be parsed from the block, in the fashion of Perl
5's parsing of: map <expression> , <list>
;
Without the comma, the form catch $foo { ... }
could be
a test for $foo
or a test for $foo{...}
(the hash element).
Object Model
This RFC is written using the basic Perl 5 concept of an object as a reference to a blessed hash containing instance variable name-value pairs. It may need to be modified to account for any new basic Perl 6 object model. On the other hand, it makes minimal demands of any object model.
Exception Base Class
If Exception
objects are not to be used for signaling assertion
failures by the guts of Perl 6, this RFC stands alone. Otherwise,
this RFC delegates the details of the Exception class to RFC 80,
modulo the required functionality described herein (more or less,
that internal exception objects are instances of classes that
inherit from the built-in Exception class).
Either way, the following questions about Exception should be addressed.
Exception::Foo
, where does one
connect non-core Exceptions into the taxonomy? Possibly the core
exceptions can derive from Exception::CORE
, and everyone else can
use the Exception::MyPackage
convention.tag
could default to file + line number.Lexical Scope
The authors would prefer that try
, catch
, and finally
blocks
share the same lexical scope.
If it is not possible to have try, catch, and finally blocks share lexical scope (due, perhaps, to the vagaries of stack unwinding), this feature can simply be deleted, and the outer scope can be shared.
One possible shared-scope problem is illustrated by this case:
try { fragile(); } catch { my $caught = 1; } finally { $caught and ... }
If fragile()
doesn't raise an exception then finally
is going
to test $caught
even though the my
statement was never executed.
Will $caught
already be allocated and undef
?
These matters will have to be referred to the internals experts.
Mixed Flow Control
Some of the reference texts, when discussing exception handling, refer
to the matter that it may be difficult to implement a go to
across
an unwinding semantics block, as in:
try { open F, $f } catch { next; }
This matter will have to be referred to the internals experts. It's ok if this functionality is not possible, it can always be simulated with lexical state variables instead.
However, the authors would very much prefer that goto
s across
unwinding boundries would DWIM. If that is not possible, hopefully
some sort of compile-time warning could be produced.
Use %@
for Errors from Builtins
RFC 151 proposes a mechanism for consolidating the information provided
by of $@
, $!
, $?
, and $^E
. In the opinion of the author of
RFC 88, merging $@
and $!
should not be undertaken, because $@
should only be set if an exception is raised.
%@
should be used to hold this fault-hash, based on the following
arguments for symmetry.
$@ current exception @@ current exception stack %@ current core fault information $@[0] same as $@ $@{type} "IO::File::NotFound" $@{message} "can't find file" $@{param} "/foo/bar/baz.dat" $@{child} $? $@{errno} $! $@{os_err} $^E $@{chunk} That chunk thingy in some msgs. $@{file} Source file name of caller. $@{line} Source line number of caller.
%@
should not contain a severity or fatality classification.
Every call to a core API function should clear %@
if it returns
successfully.
Internally, Perl can use a simple structured data type to hold the
whole canonical %@
. The code that handles reading from %@
will
construct it out of the internal data on the fly.
If use fatal;
is in scope, then just before returning, each core
API function should do something like: %@ and internal_die %@;
The internal_die
becomes the one place where a canonical Exception
can be generated to encapsulate %@
just before raising an exception,
whether or not the use of such canonical Exceptions is controlled by a
pragma such as use exceptions;
.
retry / resume
There has been some discussion on perl6-language-error about the
concept of re-entering try
blocks on catch
(including the
possibility of using such a mechanism to replace AUTOLOAD
).
The author is of the opinion that in order to do this sort of thing properly one should use continuations, which are being discussed elsewhere to this RFC.
The intent of this RFC is to provide a simple yet robust exception
handling mechanism that is suitable for error handling, not for
replacing AUTOLOAD
or continuations in general.
always
{ ... <statement> always <block>; ... }
With the semantics that the <block> would be invoked when the
lexical scope in which <statement> occurs is exited. The author
of this RFC thinks that's a fine idea, but that it is orthogonal
to structured exception handling. The behaviour of always
would
be useful whether or not the lexical scope is defined by a simple
local-flow-control block or by a set of try/catch/finally clauses.
eval
The semantics of eval
are, "clear $@
and don't unwind unless
the user re-dies after the eval
". The semantics of try
are
"unwind after try
, unless any raised exception was cleanly and
completely handled, in which case clear $@
".
In the author's opinion, both eval
and try
should exist in
Perl 6. This would also mean that the legacy of examples of how to
use eval
in Perl will still work.
And, of course, we still need eval $string
.
Discussions on perl6-language-errors have shown that some would
prefer the eval { ... }
form to be removed from Perl 6,
because having two exception handling methods in Perl could be
confusing to developers. This would in fact be possible, since
the same effect can be achieved with:
try { } catch { } # Clears $@. my $e; try { ... } catch { $e = $@; } # now process $e instead of $@
On the other hand, eval
is a convenient synonym for all that,
given that it already works that way.
catch v/s else + switch
Some participants in discussions on perl6-language-errors have
expressed the opinion that not only should eval
be used instead
of try
, but else
should be used instead of multiple catch
blocks. They are of the opinion that an else { switch ... }
should be used to handle multiple catch clauses, as in:
eval { ... } else { switch ($@) { case $@->isa("Exception::IO") { ... } case $@->my_method { ... } } }
This problem with else { switch ... }
is: how should the code
implicitly rethrow uncaught exceptions? Many proponents of this
model think that uncaught exceptions should not be implicitly
rethrown; one suggests that the programmer should undef $@
at the
end of every successful <case> block, so that Perl re-raises any
$@
still extant at the end of the else
.
This RFC allows a switch
to be used in a catch { ... }
clause, for cases where that approach would minimize redundant code
in catch <expr
{ ... }> clauses, but with the mechanism
proposed in this RFC, the switch functionality shown above can be
written like this, while still maintaining the automatic exception
propagation when no cases match:
try { ... } catch Exception::IO => { ... } catch $@->my_method => { ... }
Fatal Exceptions
Discussions on perl6-language-errors have indicated a desire by some to be able to easily separate "fatal" exceptions (like divide by zero) from "non-fatal" exceptions (such as can't open file) if Perl 6 uses exceptions for error signalling.
The authors are of the opinion that this mechanism and Perl core
exceptions (if any) should not define any sort of severity
instance
variable or is_fatal
method for the Exception
base class, because
no a priori classification of fatality can in general be provided
by the thrower of an exception--that's for the catcher to determine.
Derived Exception class hierarchies can, on the other hand, add instance data and methods that support a severity concept in their domain, at which point catching can be made conditional on such a concept.
on_raise & on_catch
Some participants in discussions on perl6-language-errors currently
prefer that the functionality represented by the on_raise
method of
the Exception class not be include at all. Should that be deemed to be
the case, the on_raise
method of the Exception
class can be left
out.
It has also been mentioned that we should consider making overriding
of throw
impossible, so that throw
must die
.
The author is of the opinion that these constraints should not be enforced, because (1) they may provide useful functionality for some applications, (2) you have to override a method, which is fairly hard to do by accident, and (3) if this functionality is not built in, there is no way to simulate it.
Independent of the matter of throw
and the hook methods, the
authors agree that there should be no way to disable the raising
of exceptions on some sort of "global" level. They get through
unless they, or their handling, explicitly say otherwise.
Stack Snapshot Object
This RFC mentions that an exception's trace
instance variable could
point to a "stack snapshot" object, rather than to a data structure. A
mechanism like this is available in the Perl 5 Devel
bundle. If an
RFC for such a snapshot class is added to Perl 6, this RFC should be
reconciled with it.
Mechanism Hooks
In the name of extensibility and debugging, there should be hooks for
callbacks to be invoked when a try
, catch
, or finally
block is
entered or exited, and when a conditional catch
is evaluated. The
callbacks would be passed information about what is happening in the
context they are being called from.
In order to scope the effect of the callbacks (rather than making them global), it is proposed that the callbacks be specified as options to the try statement, something like this:
try on_catch_enter => sub { ... }, on_catch_exit => sub { ... }, { ... }
The (dynamic, not lexical) scope of these callbacks is from their try
down through all trys nested under it (until overridden at a lower
level). Nested callbacks should have a way of chaining to callbacks
that were in scope when they come into scope, perhaps by including a
reference to the outer-scope callback as a parameter to the callback.
Basically, they could be kept in "global" variables overridden with
local
.
Normally, this mechanism would only be used by an application's
top-level try
statement, typically for the purpose of debugging
non-local flow control, or extending the semantics of try.
For example, to refuse to allow any code under the scope of a try to catch a given exception, one could just say:
on_catch_enter => sub { $@->isa("Alarm") and throw $@ }
The author would find this functionality very useful in an application he works with that uses non-error exceptions to implement non-local success-based flow control in a model- view- controller architecture.
In addition, a snapshots
option could be used by the Exception
class snapshot
method to control whether or not to take raise-time
call-stack snapshots.
Mixed-Mode Modules
Authors of modules who wish to provide a public API that respects the
current state of use fatal;
if such a mechanism is available, can
do so as follows.
Internal to their modules, authors can use lexically scoped
use fatal;
to explicitly control whether or not they want
builtins to raise exceptions to signal errors.
Then, if and only if they want to support the other style, and only for public API subroutines, they do something like one of these:
sub Foo { my $err_code = ... ; # real code goes here # Replace the old return $err_code with this: return $err_code unless $FATAL_MODE && $error_code != $ok; throw Error::Code "Couldn't Foo.", code => $err_code; }
sub Foo { try { # real code goes here, may execute: throw Exception "Couldn't foo.", code => $err_code; } catch !$FATAL_MODE => { return $@->{code}; } return $ok; }
$SIG{__DIE__}
The try, catch, and finally clauses localize and undef $SIG{__DIE__}
before entering their blocks. This behaviour can be removed if
$SIG{__DIE__}
is removed.
If $SIG{__DIE__}
is not removed, it should be invoked as at the
phrase "program shutdown is initiated" in Rule 1, not at the time an
exception is raised by die
.
If the old functionality is removed, do we want to introduce such functionality to Rule 1 anyway?
Legacy
The only changes in respect of Perl 5 behaviour implied by this RFC are
that (1) $@
is now always an Exception
object (which stringifies
reasonably), it is now read-only, and it can only be set via die
, and
(2) the @@
array is now special, and it is now read-only too.
RFC 63: Exception handling syntax proposal.
RFC 70: Allow exception-based error-reporting.
RFC 88 is proposed whether or not faults in the Perl 6 guts are signaled
by exceptions, and whether or not such behaviour is under the control of
a use fatal;
.
RFC 80: Exception objects and classes for built-ins.
The author of RFC 80 is the co-author of RFC 88, and so is aware of the impact of RFC 88 on RFC 80.
RFC 88 proposes that it's Exception and RFC 80's Exception be the same thing. RFC 88 talks about the participation of an Exception in exception handling. RFC 80 talks about how to use an Exception object to signal an exception in Perl's built-in functions and other guts.
RFC 96: A Base Class for Exception Objects
RFC 119: Object neutral error handling via exceptions.
RFC 140: One Should Not Get Away With Ignoring System Call Errors
The functionality in RFC 140 is envisioned for use in cases that don't
want to be running use fatal;
. It is orthogonal to RFC 88 whether
or not running use fatal;
.
RFC-151: Merge $!
, $^E
, and $@
See IMPACT + Use %@
for Errors from Builtins for an alternative
the proposal in RFC 151.
This RFC is based on invaluable support on a number of fronts.
This RFC has been refined with the help of (via the perl6 mailing lists) contributions from Graham Barr, Chaim Frenkel, Jonathan Scott Duff, Glenn Lindermann, Dave Rolsky, and Corwin Brust. It has also benefited from conversations with Jim Hoover.
A slightly different version of the Perl 5 implementation of Try.pm [TRY-2000] has been used by the staff of Avra Software Lab Inc. for production code, it has been debated in presentations at the Software Engineering Research Lab at the University of Alberta, and it has been discussed on the perl-friends mailing list. At one point Try.pm was refined based on the study of Graham Barr's Error.pm module [GBE-1999].
Tony Olekshy <olekshy@avrasoft.com> is the principal author of this RFC, and is responsible for any errors contained herein.
Peter Scott <peter@psdt.com> is co-author of this RFC, by virtue of the major contribution he has made hereto.
Where this document refers to the singular author or author's, it refers to Tony. Where it refers to co-author or co-author's, it refers to Peter. Where it refers to the plural authors or authors', it refers to both authors.
ESE-1994: The Encyclopedia of Software Engineering, J.J. Marciniak (editor), John Wiley & Sons Inc, 1994. ISBN 0-471-54002-1
GBE-1999: Graham Barr's Error.pm
module.
search.cpan.org
RFC 63: Exception handling syntax
RFC 70: Allow exception-based error-reporting.
RFC 80: Exception objects and classes for builtins
RFC 96: A Base Class for Exception Objects
RFC 119: Object neutral error handling via exceptions
RFC 140: One Should Not Get Away With Ignoring System Call Errors
RFC 151: Merge $!
, $^E
, $@
and $?
RYK-1989: Computer Architecture, Hardware and Software, R.Y.Kain, volume 1, Prentice-Hall Inc., 1989. ISBN 0-13-166752-1
TRY-2000: Try.pm, a Perl 5 reference implementation of the functionality described herein, is available at: www.avrasoft.com
Version 1, 2000-08-08
Version 2, 2000-08-23
except
keyword is gone, catch
takes optional
arguments instead.Error
class is no longer defined.Version 3, 2000-09-30
Exception
object.End of RFC 88