source: doc/theses/andrew_beach_MMath/features.tex @ f6106a6

Last change on this file since f6106a6 was f6106a6, checked in by Andrew Beach <ajbeach@…>, 3 years ago

Updated features to incorperate Peter's feedback.

  • Property mode set to 100644
File size: 32.0 KB
1\chapter{Exception Features}
3This chapter covers the design and user interface of the \CFA
4exception-handling mechanism (EHM). % or exception system.
6We will begin with an overview of EHMs in general. It is not a strict
7definition of all EHMs nor an exaustive list of all possible features.
8However it does cover the most common structure and features found in them.
10% We should cover what is an exception handling mechanism and what is an
11% exception before this. Probably in the introduction. Some of this could
12% move there.
13\paragraph{Raise / Handle}
14An exception operation has two main parts: raise and handle.
15These terms are sometimes also known as throw and catch but this work uses
16throw/catch as a particular kind of raise/handle.
17These are the two parts that the user will write themselves and may
18be the only two pieces of the EHM that have any syntax in the language.
21The raise is the starting point for exception handling. It marks the beginning
22of exception handling by \newterm{raising} an excepion, which passes it to
23the EHM.
25Some well known examples include the @throw@ statements of \Cpp and Java and
26the \codePy{raise} statement from Python. In real systems a raise may preform
27some other work (such as memory management) but for the purposes of this
28overview that can be ignored.
31The purpose of most exception operations is to run some user code to handle
32that exception. This code is given, with some other information, in a handler.
34A handler has three common features: the previously mentioned user code, a
35region of code they cover and an exception label/condition that matches
36certain exceptions.
37Only raises inside the covered region and raising exceptions that match the
38label can be handled by a given handler.
39Different EHMs will have different rules to pick a handler
40if multipe handlers could be used such as ``best match" or ``first found".
42The @try@ statements of \Cpp, Java and Python are common examples. All three
43also show another common feature of handlers, they are grouped by the covered
47After an exception is raised comes what is usually the biggest step for the
48EHM: finding and setting up the handler. The propogation from raise to
49handler can be broken up into three different tasks: searching for a handler,
50matching against the handler and installing the handler.
53The EHM begins by searching for handlers that might be used to handle
54the exception. Searching is usually independent of the exception that was
55thrown as it looks for handlers that have the raise site in their covered
57This includes handlers in the current function, as well as any in callers
58on the stack that have the function call in their covered region.
61Each handler found has to be matched with the raised exception. The exception
62label defines a condition that be use used with exception and decides if
63there is a match or not.
65In languages where the first match is used this step is intertwined with
66searching, a match check is preformed immediately after the search finds
67a possible handler.
70After a handler is chosen it must be made ready to run.
71The implementation can vary widely to fit with the rest of the
72design of the EHM. The installation step might be trivial or it could be
73the most expensive step in handling an exception. The latter tends to be the
74case when stack unwinding is involved.
76If a matching handler is not guarantied to be found the EHM will need a
77different course of action here in the cases where no handler matches.
78This is only required with unchecked exceptions as checked exceptions
79(such as in Java) can make than guaranty.
80This different action can also be installing a handler but it is usually an
81implicat and much more general one.
84A common way to organize exceptions is in a hierarchical structure.
85This is especially true in object-orientated languages where the
86exception hierarchy is a natural extension of the object hierarchy.
88Consider the following hierarchy of exceptions:
92\put(2100,-1411){\vector(1, 0){225}}
93\put(3450,-1411){\vector(1, 0){225}}
95\put(3550,-1636){\vector(1, 0){150}}
97\put(3550,-1861){\vector(1, 0){150}}
106A handler labelled with any given exception can handle exceptions of that
107type or any child type of that exception. The root of the exception hierarchy
108(here \codeC{exception}) acts as a catch-all, leaf types catch single types
109and the exceptions in the middle can be used to catch different groups of
110related exceptions.
112This system has some notable advantages, such as multiple levels of grouping,
113the ability for libraries to add new exception types and the isolation
114between different sub-hierarchies.
115This design is used in \CFA even though it is not a object-orientated
116language using different tools to create the hierarchy.
118% Could I cite the rational for the Python IO exception rework?
121After the handler has finished the entire exception operation has to complete
122and continue executing somewhere else. This step is usually simple,
123both logically and in its implementation, as the installation of the handler
124is usually set up to do most of the work.
126The EHM can return control to many different places,
127the most common are after the handler definition and after the raise.
130For effective exception handling, additional information is usually passed
131from the raise to the handler.
132So far only communication of the exceptions' identity has been covered.
133A common method is putting fields into the exception instance and giving the
134handler access to them.
137Virtual types and casts are not part of \CFA's EHM nor are they required for
138any EHM. But \CFA uses a hierarchial system of exceptions and this feature
139is leveraged to create that.
141% Maybe talk about why the virtual system is so minimal.
142% Created for but not a part of the exception system.
144The virtual system supports multiple ``trees" of types. Each tree is
145a simple hierarchy with a single root type. Each type in a tree has exactly
146one parent -- except for the root type which has zero parents -- and any
147number of children.
148Any type that belongs to any of these trees is called a virtual type.
150% A type's ancestors are its parent and its parent's ancestors.
151% The root type has no ancestors.
152% A type's decendents are its children and its children's decendents.
154Every virtual type also has a list of virtual members. Children inherit
155their parent's list of virtual members but may add new members to it.
156It is important to note that these are virtual members, not virtual methods
157of object-orientated programming, and can be of any type.
158However, since \CFA has function pointers and they are allowed, virtual
159members can be used to mimic virtual methods.
161Each virtual type has a unique id.
162This unique id and all the virtual members are combined
163into a virtual table type. Each virtual type has a pointer to a virtual table
164as a hidden field.
166Up until this point the virtual system is similar to ones found in
167object-orientated languages but this where \CFA diverges. Objects encapsulate a
168single set of behaviours in each type, universally across the entire program,
169and indeed all programs that use that type definition. In this sense the
170types are ``closed" and cannot be altered.
172In \CFA types do not encapsulate any behaviour. Traits are local and
173types can begin to statify a trait, stop satifying a trait or satify the same
174trait in a different way at any lexical location in the program.
175In this sense they are ``open" as they can change at any time. This means it
176is implossible to pick a single set of functions that repersent the type's
177implementation across the program.
179\CFA side-steps this issue by not having a single virtual table for each
180type. A user can define virtual tables which are filled in at their
181declaration and given a name. Anywhere that name is visible, even if it was
182defined locally inside a function (although that means it will not have a
183static lifetime), it can be used.
184Specifically, a virtual type is ``bound" to a virtual table which
185sets the virtual members for that object. The virtual members can be accessed
186through the object.
188While much of the virtual infrastructure is created, it is currently only used
189internally for exception handling. The only user-level feature is the virtual
190cast, which is the same as the \Cpp \lstinline[language=C++]|dynamic_cast|.
195Note, the syntax and semantics matches a C-cast, rather than the function-like
196\Cpp syntax for special casts. Both the type of @EXPRESSION@ and @TYPE@ must be
197a pointer to a virtual type.
198The cast dynamically checks if the @EXPRESSION@ type is the same or a sub-type
199of @TYPE@, and if true, returns a pointer to the
200@EXPRESSION@ object, otherwise it returns @0p@ (null pointer).
203% Leaving until later, hopefully it can talk about actual syntax instead
204% of my many strange macros. Syntax aside I will also have to talk about the
205% features all exceptions support.
207Exceptions are defined by the trait system; there are a series of traits, and
208if a type satisfies them, then it can be used as an exception. The following
209is the base trait all exceptions need to match.
211trait is_exception(exceptT &, virtualT &) {
212        virtualT const & get_exception_vtable(exceptT *);
215The trait is defined over two types, the exception type and the virtual table
216type. This should be one-to-one: each exception type has only one virtual
217table type and vice versa. The only assertion in the trait is
218@get_exception_vtable@, which takes a pointer of the exception type and
219returns a reference to the virtual table type instance.
221% TODO: This section, and all references to get_exception_vtable, are
222% out-of-data. Perhaps wait until the update is finished before rewriting it.
223The function @get_exception_vtable@ is actually a constant function.
224Regardless of the value passed in (including the null pointer) it should
225return a reference to the virtual table instance for that type.
226The reason it is a function instead of a constant is that it make type
227annotations easier to write as you can use the exception type instead of the
228virtual table type; which usually has a mangled name.
229% Also \CFA's trait system handles functions better than constants and doing
230% it this way reduce the amount of boiler plate we need.
232% I did have a note about how it is the programmer's responsibility to make
233% sure the function is implemented correctly. But this is true of every
234% similar system I know of (except Agda's I guess) so I took it out.
236There are two more traits for exceptions defined as follows:
238trait is_termination_exception(
239                exceptT &, virtualT & | is_exception(exceptT, virtualT)) {
240        void defaultTerminationHandler(exceptT &);
243trait is_resumption_exception(
244                exceptT &, virtualT & | is_exception(exceptT, virtualT)) {
245        void defaultResumptionHandler(exceptT &);
248Both traits ensure a pair of types are an exception type and its virtual table
249and defines one of the two default handlers. The default handlers are used
250as fallbacks and are discussed in detail in \VRef{s:ExceptionHandling}.
252However, all three of these traits can be tricky to use directly.
253While there is a bit of repetition required,
254the largest issue is that the virtual table type is mangled and not in a user
255facing way. So these three macros are provided to wrap these traits to
256simplify referring to the names:
259All three take one or two arguments. The first argument is the name of the
260exception type. The macro passes its unmangled and mangled form to the trait.
261The second (optional) argument is a parenthesized list of polymorphic
262arguments. This argument is only used with polymorphic exceptions and the
263list is be passed to both types.
264In the current set-up, the two types always have the same polymorphic
265arguments so these macros can be used without losing flexibility.
267For example consider a function that is polymorphic over types that have a
268defined arithmetic exception:
270forall(Num | IS_EXCEPTION(Arithmetic, (Num)))
271void some_math_function(Num & left, Num & right);
274\section{Exception Handling}
276\CFA provides two kinds of exception handling: termination and resumption.
277These twin operations are the core of \CFA's exception handling mechanism.
278This section will cover the general patterns shared by the two operations and
279then go on to cover the details each individual operation.
281Both operations follow the same set of steps.
282Both start with the user preforming a raise on an exception.
283Then the exception propogates up the stack.
284If a handler is found the exception is caught and the handler is run.
285After that control returns to normal execution.
286If the search fails a default handler is run and then control
287returns to normal execution after the raise.
289This general description covers what the two kinds have in common.
290Differences include how propogation is preformed, where exception continues
291after an exception is caught and handled and which default handler is run.
295Termination handling is the familiar kind and used in most programming
296languages with exception handling.
297It is dynamic, non-local goto. If the raised exception is matched and
298handled the stack is unwound and control will (usually) continue the function
299on the call stack that defined the handler.
300Termination is commonly used when an error has occurred and recovery is
301impossible locally.
303% (usually) Control can continue in the current function but then a different
304% control flow construct should be used.
306A termination raise is started with the @throw@ statement:
308throw EXPRESSION;
310The expression must return a reference to a termination exception, where the
311termination exception is any type that satisfies the trait
312@is_termination_exception@ at the call site.
313Through \CFA's trait system the trait functions are implicity passed into the
314throw code and the EHM.
315A new @defaultTerminationHandler@ can be defined in any scope to
316change the throw's behavior (see below).
318The throw will copy the provided exception into managed memory to ensure
319the exception is not destroyed if the stack is unwound.
320It is the user's responsibility to ensure the original exception is cleaned
321up wheither the stack is unwound or not. Allocating it on the stack is
322usually sufficient.
324Then propogation starts with the search. \CFA uses a ``first match" rule so
325matching is preformed with the copied exception as the search continues.
326It starts from the throwing function and proceeds to the base of the stack,
327from callee to caller.
328At each stack frame, a check is made for resumption handlers defined by the
329@catch@ clauses of a @try@ statement.
331try {
332        GUARDED_BLOCK
333} catch (EXCEPTION_TYPE$\(_1\)$ * [NAME$\(_1\)$]) {
334        HANDLER_BLOCK$\(_1\)$
335} catch (EXCEPTION_TYPE$\(_2\)$ * [NAME$\(_2\)$]) {
336        HANDLER_BLOCK$\(_2\)$
339When viewed on its own, a try statement will simply execute the statements
340in @GUARDED_BLOCK@ and when those are finished the try statement finishes.
342However, while the guarded statements are being executed, including any
343invoked functions, all the handlers in the statement are now on the search
344path. If a termination exception is thrown and not handled further up the
345stack they will be matched against the exception.
347Exception matching checks the handler in each catch clause in the order
348they appear, top to bottom. If the representation of the thrown exception type
349is the same or a descendant of @EXCEPTION_TYPE@$_i$ then @NAME@$_i$
350(if provided) is
351bound to a pointer to the exception and the statements in @HANDLER_BLOCK@$_i$
352are executed. If control reaches the end of the handler, the exception is
353freed and control continues after the try statement.
355If no termination handler is found during the search then the default handler
356(@defaultTerminationHandler@) is run.
357Through \CFA's trait system the best match at the throw sight will be used.
358This function is run and is passed the copied exception. After the default
359handler is run control continues after the throw statement.
361There is a global @defaultTerminationHandler@ that is polymorphic over all
362exception types. Since it is so general a more specific handler can be
363defined and will be used for those types, effectively overriding the handler
364for particular exception type.
365The global default termination handler performs a cancellation
366\see{\VRef{s:Cancellation}} on the current stack with the copied exception.
371Resumption exception handling is less common than termination but is
372just as old~\cite{Goodenough75} and is simpler in many ways.
373It is a dynamic, non-local function call. If the raised exception is
374matched a closure will be taken from up the stack and executed,
375after which the raising function will continue executing.
376These are most often used when an error occurred and if the error is repaired
377then the function can continue.
379A resumption raise is started with the @throwResume@ statement:
381throwResume EXPRESSION;
383It works much the same way as the termination throw.
384The expression must return a reference to a resumption exception,
385where the resumption exception is any type that satisfies the trait
386@is_resumption_exception@ at the call site.
387The assertions from this trait are available to
388the exception system while handling the exception.
390At run-time, no exception copy is made.
391As the stack is not unwound the exception and
392any values on the stack will remain in scope while the resumption is handled.
394The EHM then begins propogation. The search starts from the raise in the
395resuming function and proceeds to the base of the stack, from callee to caller.
396At each stack frame, a check is made for resumption handlers defined by the
397@catchResume@ clauses of a @try@ statement.
399try {
400        GUARDED_BLOCK
401} catchResume (EXCEPTION_TYPE$\(_1\)$ * [NAME$\(_1\)$]) {
402        HANDLER_BLOCK$\(_1\)$
403} catchResume (EXCEPTION_TYPE$\(_2\)$ * [NAME$\(_2\)$]) {
404        HANDLER_BLOCK$\(_2\)$
407% I wonder if there would be some good central place for this.
408Note that termination handlers and resumption handlers may be used together
409in a single try statement, intermixing @catch@ and @catchResume@ freely.
410Each type of handler will only interact with exceptions from the matching
411type of raise.
412When a try statement is executed it simply executes the statements in the
413@GUARDED_BLOCK@ and then finishes.
415However, while the guarded statements are being executed, including any
416invoked functions, all the handlers in the statement are now on the search
417path. If a resumption exception is reported and not handled further up the
418stack they will be matched against the exception.
420Exception matching checks the handler in each catch clause in the order
421they appear, top to bottom. If the representation of the thrown exception type
422is the same or a descendant of @EXCEPTION_TYPE@$_i$ then @NAME@$_i$
423(if provided) is bound to a pointer to the exception and the statements in
424@HANDLER_BLOCK@$_i$ are executed.
425If control reaches the end of the handler, execution continues after the
426the raise statement that raised the handled exception.
428Like termination, if no resumption handler is found, the default handler
429visible at the throw statement is called. It will use the best match at the
430call sight according to \CFA's overloading rules. The default handler is
431passed the exception given to the throw. When the default handler finishes
432execution continues after the raise statement.
434There is a global @defaultResumptionHandler@ is polymorphic over all
435termination exceptions and preforms a termination throw on the exception.
436The @defaultTerminationHandler@ for that raise is matched at the original
437raise statement (the resumption @throwResume@) and it can be customized by
438introducing a new or better match as well.
440\subsubsection{Resumption Marking}
441A key difference between resumption and termination is that resumption does
442not unwind the stack. A side effect that is that when a handler is matched
443and run it's try block (the guarded statements) and every try statement
444searched before it are still on the stack. This can lead to the recursive
445resumption problem.
447The recursive resumption problem is any situation where a resumption handler
448ends up being called while it is running.
449Consider a trivial case:
451try {
452        throwResume (E &){};
453} catchResume(E *) {
454        throwResume (E &){};
457When this code is executed the guarded @throwResume@ will throw, start a
458search and match the handler in the @catchResume@ clause. This will be
459call and placed on the stack on top of the try-block. The second throw then
460throws and will search the same try block and put call another instance of the
461same handler leading to an infinite loop.
463This situation is trivial and easy to avoid, but much more complex cycles
464can form with multiple handlers and different exception types.
466To prevent all of these cases we mark try statements on the stack.
467A try statement is marked when a match check is preformed with it and an
468exception. The statement will be unmarked when the handling of that exception
469is completed or the search completes without finding a handler.
470While a try statement is marked its handlers are never matched, effectify
471skipping over it to the next try statement.
473% This might need a diagram. But it is an important part of the justification
474% of the design of the traversal order.
476       throwResume2 ----------.
477            |                 |
478 generated from handler       |
479            |                 |
480         handler              |
481            |                 |
482        throwResume1 -----.   :
483            |             |   :
484           try            |   : search skip
485            |             |   :
486        catchResume  <----'   :
487            |                 |
490These rules mirror what happens with termination.
491When a termination throw happens in a handler the search will not look at
492any handlers from the original throw to the original catch because that
493part of the stack has been unwound.
494A resumption raise in the same situation wants to search the entire stack,
495but it will not try to match the exception with try statements in the section
496that would have been unwound as they are marked.
498The symmetry between resumption termination is why this pattern was picked.
499Other patterns, such as marking just the handlers that caught, also work but
500lack the symmetry means there are less rules to remember.
502\section{Conditional Catch}
503Both termination and resumption handler clauses can be given an additional
504condition to further control which exceptions they handle:
508First, the same semantics is used to match the exception type. Second, if the
509exception matches, @CONDITION@ is executed. The condition expression may
510reference all names in scope at the beginning of the try block and @NAME@
511introduced in the handler clause. If the condition is true, then the handler
512matches. Otherwise, the exception search continues as if the exception type
513did not match.
515The condition matching allows finer matching by allowing the match to check
516more kinds of information than just the exception type.
518try {
519        handle1 = open( f1, ... );
520        handle2 = open( f2, ... );
521        handle3 = open( f3, ... );
522        ...
523} catch( IOFailure * f ; fd( f ) == f1 ) {
524        // Only handle IO failure for f1.
525} catch( IOFailure * f ; fd( f ) == f3 ) {
526        // Only handle IO failure for f3.
528// Can't handle a failure relating to f2 here.
530In this example the file that experianced the IO error is used to decide
531which handler should be run, if any at all.
534% I know I actually haven't got rid of them yet, but I'm going to try
535% to write it as if I had and see if that makes sense:
538Within the handler block or functions called from the handler block, it is
539possible to reraise the most recently caught exception with @throw@ or
540@throwResume@, respectively.
542try {
543        ...
544} catch( ... ) {
545        ... throw;
546} catchResume( ... ) {
547        ... throwResume;
550The only difference between a raise and a reraise is that reraise does not
551create a new exception; instead it continues using the current exception, \ie
552no allocation and copy. However the default handler is still set to the one
553visible at the raise point, and hence, for termination could refer to data that
554is part of an unwound stack frame. To prevent this problem, a new default
555handler is generated that does a program-level abort.
558\subsection{Comparison with Reraising}
559A more popular way to allow handlers to match in more detail is to reraise
560the exception after it has been caught if it could not be handled here.
561On the surface these two features seem interchangable.
563If we used @throw;@ to start a termination reraise then these two statements
564would have the same behaviour:
566try {
567    do_work_may_throw();
568} catch(exception_t * exc ; can_handle(exc)) {
569    handle(exc);
574try {
575    do_work_may_throw();
576} catch(exception_t * exc) {
577    if (can_handle(exc)) {
578        handle(exc);
579    } else {
580        throw;
581    }
584If there are further handlers after this handler only the first version will
585check them. If multiple handlers on a single try block could handle the same
586exception the translations get more complex but they are equivilantly
589Until stack unwinding comes into the picture. In termination handling, a
590conditional catch happens before the stack is unwound, but a reraise happens
591afterwards. Normally this might only cause you to loose some debug
592information you could get from a stack trace (and that can be side stepped
593entirely by collecting information during the unwind). But for \CFA there is
594another issue, if the exception isn't handled the default handler should be
595run at the site of the original raise.
597There are two problems with this: the site of the original raise doesn't
598exist anymore and the default handler might not exist anymore. The site will
599always be removed as part of the unwinding, often with the entirety of the
600function it was in. The default handler could be a stack allocated nested
601function removed during the unwind.
603This means actually trying to pretend the catch didn't happening, continuing
604the original raise instead of starting a new one, is infeasible.
605That is the expected behaviour for most languages and we can't replicate
606that behaviour.
608\section{Finally Clauses}
610Finally clauses are used to preform unconditional clean-up when leaving a
611scope and are placed at the end of a try statement after any handler clauses:
613try {
614        GUARDED_BLOCK
615} ... // any number or kind of handler clauses
616... finally {
617        FINALLY_BLOCK
620The @FINALLY_BLOCK@ is executed when the try statement is removed from the
621stack, including when the @GUARDED_BLOCK@ finishes, any termination handler
622finishes or during an unwind.
623The only time the block is not executed is if the program is exited before
624the stack is unwound.
626Execution of the finally block should always finish, meaning control runs off
627the end of the block. This requirement ensures control always continues as if
628the finally clause is not present, \ie finally is for cleanup not changing
629control flow.
630Because of this requirement, local control flow out of the finally block
631is forbidden. The compiler precludes any @break@, @continue@, @fallthru@ or
632@return@ that causes control to leave the finally block. Other ways to leave
633the finally block, such as a long jump or termination are much harder to check,
634and at best requiring additional run-time overhead, and so are only
637Not all languages with unwinding have finally clauses. Notably \Cpp does
638without it as descructors serve a similar role. Although destructors and
639finally clauses can be used in many of the same areas they have their own
640use cases like top-level functions and lambda functions with closures.
641Destructors take a bit more work to set up but are much easier to reuse while
642finally clauses are good for one-off uses and
643can easily include local information.
647Cancellation is a stack-level abort, which can be thought of as as an
648uncatchable termination. It unwinds the entire current stack, and if
649possible forwards the cancellation exception to a different stack.
651Cancellation is not an exception operation like termination or resumption.
652There is no special statement for starting a cancellation; instead the standard
653library function @cancel_stack@ is called passing an exception. Unlike a
654raise, this exception is not used in matching only to pass information about
655the cause of the cancellation.
656(This also means matching cannot fail so there is no default handler.)
658After @cancel_stack@ is called the exception is copied into the EHM's memory
659and the current stack is
660unwound. After that it depends one which stack is being cancelled.
662\item[Main Stack:]
663The main stack is the one used by the program main at the start of execution,
664and is the only stack in a sequential program.
665After the main stack is unwound there is a program-level abort.
667There are two reasons for this. The first is that it obviously had to do this
668in a sequential program as there is nothing else to notify and the simplicity
669of keeping the same behaviour in sequential and concurrent programs is good.
670Also, even in concurrent programs there is no stack that an innate connection
671to, so it would have be explicitly managed.
673\item[Thread Stack:]
674A thread stack is created for a \CFA @thread@ object or object that satisfies
675the @is_thread@ trait.
676After a thread stack is unwound there exception is stored until another
677thread attempts to join with it. Then the exception @ThreadCancelled@,
678which stores a reference to the thread and to the exception passed to the
679cancellation, is reported from the join.
680There is one difference between an explicit join (with the @join@ function)
681and an implicit join (from a destructor call). The explicit join takes the
682default handler (@defaultResumptionHandler@) from its calling context while
683the implicit join provides its own which does a program abort if the
684@ThreadCancelled@ exception cannot be handled.
686Communication is done at join because a thread only has to have to points of
687communication with other threads: start and join.
688Since a thread must be running to perform a cancellation (and cannot be
689cancelled from another stack), the cancellation must be after start and
690before the join. So join is the one that we will use.
692% TODO: Find somewhere to discuss unwind collisions.
693The difference between the explicit and implicit join is for safety and
694debugging. It helps prevent unwinding collisions by avoiding throwing from
695a destructor and prevents cascading the error across multiple threads if
696the user is not equipped to deal with it.
697Also you can always add an explicit join if that is the desired behaviour.
699\item[Coroutine Stack:]
700A coroutine stack is created for a @coroutine@ object or object that
701satisfies the @is_coroutine@ trait.
702After a coroutine stack is unwound control returns to the resume function
703that most recently resumed it. The resume statement reports a
704@CoroutineCancelled@ exception, which contains a references to the cancelled
705coroutine and the exception used to cancel it.
706The resume function also takes the @defaultResumptionHandler@ from the
707caller's context and passes it to the internal report.
709A coroutine knows of two other coroutines, its starter and its last resumer.
710The starter has a much more distant connection while the last resumer just
711(in terms of coroutine state) called resume on this coroutine, so the message
712is passed to the latter.
Note: See TracBrowser for help on using the repository browser.