| 1 |
NAME |
| 2 |
|
| 3 |
CPS - manage flow of control in Continuation-Passing Style |
| 4 |
|
| 5 |
OVERVIEW |
| 6 |
|
| 7 |
Note: This module is entirely deprecated now. It is maintained for |
| 8 |
compatibility for any code still using it, but please consider |
| 9 |
rewriting to use Future instead, which offers a far neater method of |
| 10 |
representing asynchronous program and data flow. In addition, |
| 11 |
Future::AsyncAwait can further improve readability of Future-based |
| 12 |
code by letting it use the familiar kinds of Perl control structure |
| 13 |
while still being asynchronous. |
| 14 |
|
| 15 |
At some later date this entire CPS module distribution may be |
| 16 |
deleted. |
| 17 |
|
| 18 |
The functions in this module implement or assist the writing of |
| 19 |
programs, or parts of them, in Continuation Passing Style (CPS). |
| 20 |
Briefly, CPS is a style of writing code where the normal call/return |
| 21 |
mechanism is replaced by explicit "continuations", values passed in to |
| 22 |
functions which they should invoke, to implement return behaviour. For |
| 23 |
more detail on CPS, see the SEE ALSO section. |
| 24 |
|
| 25 |
What this module implements is not in fact true CPS, as Perl does not |
| 26 |
natively support the idea of a real continuation (such as is created by |
| 27 |
a co-routine). Furthermore, for CPS to be efficient in languages that |
| 28 |
natively support it, their runtimes typically implement a lot of |
| 29 |
optimisation of CPS code, which the Perl interpreter would be unable to |
| 30 |
perform. Instead, CODE references are passed around to stand in their |
| 31 |
place. While not particularly useful for most regular cases, this |
| 32 |
becomes very useful whenever some form of asynchronous or event-based |
| 33 |
programming is being used. Continuations passed in to the body function |
| 34 |
of a control structure can be stored in the event handlers of the |
| 35 |
asynchronous or event-driven framework, so that when they are invoked |
| 36 |
later, the code continues, eventually arriving at its final answer at |
| 37 |
some point in the future. |
| 38 |
|
| 39 |
In order for these examples to make sense, a fictional and simple |
| 40 |
asynchronisation framework has been invented. The exact details of |
| 41 |
operation should not be important, as it simply stands to illustrate |
| 42 |
the point. I hope its general intention should be obvious. :) |
| 43 |
|
| 44 |
read_stdin_line( \&on_line ); # wait on a line from STDIN, then pass it |
| 45 |
# to the handler function |
| 46 |
|
| 47 |
This module itself provides functions that manage the flow of control |
| 48 |
through a continuation passing program. They do not directly facilitate |
| 49 |
the flow of data through a program. That can be managed by lexical |
| 50 |
variables captured by the closures passed around. See the EXAMPLES |
| 51 |
section. |
| 52 |
|
| 53 |
For CPS versions of data-flow functionals, such as map and grep, see |
| 54 |
also CPS::Functional. |
| 55 |
|
| 56 |
SYNOPSIS |
| 57 |
|
| 58 |
use CPS qw( kloop ); |
| 59 |
|
| 60 |
kloop( sub { |
| 61 |
my ( $knext, $klast ) = @_; |
| 62 |
|
| 63 |
print "Enter a number, or q to quit: "; |
| 64 |
|
| 65 |
read_stdin_line( sub { |
| 66 |
my ( $first ) = @_; |
| 67 |
chomp $first; |
| 68 |
|
| 69 |
return $klast->() if $first eq "q"; |
| 70 |
|
| 71 |
print "Enter a second number: "; |
| 72 |
|
| 73 |
read_stdin_line( sub { |
| 74 |
my ( $second ) = @_; |
| 75 |
|
| 76 |
print "The sum is " . ( $first + $second ) . "\n"; |
| 77 |
|
| 78 |
$knext->(); |
| 79 |
} ); |
| 80 |
} ); |
| 81 |
}, |
| 82 |
sub { exit } |
| 83 |
); |
| 84 |
|
| 85 |
FUNCTIONS |
| 86 |
|
| 87 |
In all of the following functions, the \&body function can provide |
| 88 |
results by invoking its continuation / one of its continuations, either |
| 89 |
synchronously or asynchronously at some point later (via some event |
| 90 |
handling or other mechanism); the next invocation of \&body will not |
| 91 |
take place until the previous one exits if it is done synchronously. |
| 92 |
|
| 93 |
They all take the prefix k before the name of the regular perl keyword |
| 94 |
or function they aim to replace. It is common in CPS code in other |
| 95 |
languages, such as Scheme or Haskell, to store a continuation in a |
| 96 |
variable called k. This convention is followed here. |
| 97 |
|
| 98 |
kloop( \&body, $k ) |
| 99 |
|
| 100 |
CPS version of perl's while(true) loop. Repeatedly calls the body code |
| 101 |
until it indicates the end of the loop, then invoke $k. |
| 102 |
|
| 103 |
$body->( $knext, $klast ) |
| 104 |
$knext->() |
| 105 |
$klast->() |
| 106 |
|
| 107 |
$k->() |
| 108 |
|
| 109 |
If $knext is invoked, the body will be called again. If $klast is |
| 110 |
invoked, the continuation $k is invoked. |
| 111 |
|
| 112 |
kwhile( \&body, $k ) |
| 113 |
|
| 114 |
Compatibility synonym for kloop; it was renamed after version 0.10. New |
| 115 |
code should use kloop instead. |
| 116 |
|
| 117 |
kforeach( \@items, \&body, $k ) |
| 118 |
|
| 119 |
CPS version of perl's foreach loop. Calls the body code once for each |
| 120 |
element in @items, until either the items are exhausted or the body |
| 121 |
invokes its $klast continuation, then invoke $k. |
| 122 |
|
| 123 |
$body->( $item, $knext, $klast ) |
| 124 |
$knext->() |
| 125 |
$klast->() |
| 126 |
|
| 127 |
$k->() |
| 128 |
|
| 129 |
kdescendd( $root, \&body, $k ) |
| 130 |
|
| 131 |
CPS version of recursive descent on a tree-like structure, defined by a |
| 132 |
function, body, which when given a node in the tree, yields a list of |
| 133 |
child nodes. |
| 134 |
|
| 135 |
$body->( $node, $kmore ) |
| 136 |
$kmore->( @child_nodes ) |
| 137 |
|
| 138 |
$k->() |
| 139 |
|
| 140 |
The first value to be passed into body is $root. |
| 141 |
|
| 142 |
At each iteration, a node is given to the body function, and it is |
| 143 |
expected to pass a list of child nodes into its $kmore continuation. |
| 144 |
These will then be iterated over, in the order given. The tree-like |
| 145 |
structure is visited depth-first, descending fully into one subtree of |
| 146 |
a node before moving on to the next. |
| 147 |
|
| 148 |
This function does not provide a way for the body to accumulate a |
| 149 |
resultant data structure to pass into its own continuation. The body is |
| 150 |
executed simply for its side-effects and its continuation is invoked |
| 151 |
with no arguments. A variable of some sort should be shared between the |
| 152 |
body and the continuation if this is required. |
| 153 |
|
| 154 |
kdescendb( $root, \&body, $k ) |
| 155 |
|
| 156 |
A breadth-first variation of kdescendd. This function visits each child |
| 157 |
node of the parent, before iterating over all of these nodes's |
| 158 |
children, recursively until the bottom of the tree. |
| 159 |
|
| 160 |
kpar( @bodies, $k ) |
| 161 |
|
| 162 |
This CPS function takes a list of function bodies and calls them all |
| 163 |
immediately. Each is given its own continuation. Once every body has |
| 164 |
invoked its continuation, the main continuation $k is invoked. |
| 165 |
|
| 166 |
$body->( $kdone ) |
| 167 |
$kdone->() |
| 168 |
|
| 169 |
$k->() |
| 170 |
|
| 171 |
This allows running multiple operations in parallel, and waiting for |
| 172 |
them all to complete before continuing. It provides in a CPS form |
| 173 |
functionality similar to that provided in a more object-oriented |
| 174 |
fashion by modules such as Async::MergePoint or Event::Join. |
| 175 |
|
| 176 |
kpareach( \@items, \&body, $k ) |
| 177 |
|
| 178 |
This CPS function takes a list of items and a function body, and calls |
| 179 |
the body immediately once for each item in the list. Each invocation is |
| 180 |
given its own continuation. Once every body has invoked its |
| 181 |
continuation, the main continuation $k is invoked. |
| 182 |
|
| 183 |
$body->( $item, $kdone ) |
| 184 |
$kdone->() |
| 185 |
|
| 186 |
$k->() |
| 187 |
|
| 188 |
This is similar to kforeach, except that the body is started |
| 189 |
concurrently for all items in the list list, rather than each item |
| 190 |
waiting for the previous to finish. |
| 191 |
|
| 192 |
kseq( @bodies, $k ) |
| 193 |
|
| 194 |
This CPS function takes a list of function bodies and calls them each, |
| 195 |
one at a time in sequence. Each is given a continuation to invoke, |
| 196 |
which will cause the next body to be invoked. When the last body has |
| 197 |
invoked its continuation, the main continuation $k is invoked. |
| 198 |
|
| 199 |
$body->( $kdone ) |
| 200 |
$kdone->() |
| 201 |
|
| 202 |
$k->() |
| 203 |
|
| 204 |
A benefit of this is that it allows a long operation that uses many |
| 205 |
continuation "pauses", to be written without code indenting further and |
| 206 |
further to the right. Another is that it allows easy skipping of |
| 207 |
conditional parts of a computation, which would otherwise be tricky to |
| 208 |
write in a CPS form. See the EXAMPLES section. |
| 209 |
|
| 210 |
GOVERNORS |
| 211 |
|
| 212 |
All of the above functions are implemented using a loop which |
| 213 |
repeatedly calls the body function until some terminating condition. By |
| 214 |
controlling the way this loop re-invokes itself, a program can control |
| 215 |
the behaviour of the functions. |
| 216 |
|
| 217 |
For every one of the above functions, there also exists a variant which |
| 218 |
takes a CPS::Governor object as its first argument. These functions use |
| 219 |
the governor object to control their iteration. |
| 220 |
|
| 221 |
kloop( \&body, $k ) |
| 222 |
gkloop( $gov, \&body, $k ) |
| 223 |
|
| 224 |
kforeach( \@items, \&body, $k ) |
| 225 |
gkforeach( $gov, \@items, \&body, $k ) |
| 226 |
|
| 227 |
etc... |
| 228 |
|
| 229 |
In this way, other governor objects can be constructed which have |
| 230 |
different running properties; such as interleaving iterations of their |
| 231 |
loop with other IO activity in an event-driven framework, or giving |
| 232 |
rate-limitation control on the speed of iteration of the loop. |
| 233 |
|
| 234 |
CPS UTILITIES |
| 235 |
|
| 236 |
These function names do not begin with k because they are not |
| 237 |
themselves CPS primatives, but may be useful in CPS-oriented code. |
| 238 |
|
| 239 |
$kfunc = liftk { BLOCK } |
| 240 |
|
| 241 |
$kfunc = liftk( \&func ) |
| 242 |
|
| 243 |
Returns a new CODE reference to a CPS-wrapped version of the code block |
| 244 |
or passed CODE reference. When $kfunc is invoked, the function &func is |
| 245 |
called in list context, being passed all the arguments given to $kfunc |
| 246 |
apart from the last, expected to be its continuation. When &func |
| 247 |
returns, the result is passed into the continuation. |
| 248 |
|
| 249 |
$kfunc->( @func_args, $k ) |
| 250 |
$k->( @func_ret ) |
| 251 |
|
| 252 |
The following are equivalent |
| 253 |
|
| 254 |
print func( 1, 2, 3 ); |
| 255 |
|
| 256 |
my $kfunc = liftk( \&func ); |
| 257 |
$kfunc->( 1, 2, 3, sub { print @_ } ); |
| 258 |
|
| 259 |
Note that the returned wrapper function only has one continuation slot |
| 260 |
in its arguments. It therefore cannot be used as the body for kloop(), |
| 261 |
kforeach() or kgenerate(), because these pass two continuations. There |
| 262 |
does not exist a "natural" way to lift a normal call/return function |
| 263 |
into a CPS function which requires more than one continuation, because |
| 264 |
there is no way to distinguish the different named returns. |
| 265 |
|
| 266 |
$func = dropk { BLOCK } $kfunc |
| 267 |
|
| 268 |
$func = dropk $waitfunc, $kfunc |
| 269 |
|
| 270 |
Returns a new CODE reference to a plain call/return version of the |
| 271 |
passed CPS-style CODE reference. When the returned ("dropped") function |
| 272 |
is called, it invokes the passed CPS function, then waits for it to |
| 273 |
invoke its continuation. When it does, the list that was passed to the |
| 274 |
continuation is returned by the dropped function. If called in scalar |
| 275 |
context, only the first value in the list is returned. |
| 276 |
|
| 277 |
$kfunc->( @func_args, $k ) |
| 278 |
$k->( @func_ret ) |
| 279 |
|
| 280 |
$waitfunc->() |
| 281 |
|
| 282 |
@func_ret = $func->( @func_args ) |
| 283 |
|
| 284 |
Given the following trivial CPS function: |
| 285 |
|
| 286 |
$kadd = sub { $_[2]->( $_[0] + $_[1] ) }; |
| 287 |
|
| 288 |
The following are equivalent |
| 289 |
|
| 290 |
$kadd->( 10, 20, sub { print "The total is $_[0]\n" } ); |
| 291 |
|
| 292 |
$add = dropk { } $kadd; |
| 293 |
print "The total is ".$add->( 10, 20 )."\n"; |
| 294 |
|
| 295 |
In the general case the CPS function hasn't yet invoked its |
| 296 |
continuation by the time it returns (such as would be the case when |
| 297 |
using any sort of asynchronisation or event-driven framework). For |
| 298 |
dropk to actually work in this situation, it requires a way to run the |
| 299 |
event framework, to cause it to process events until the continuation |
| 300 |
has been invoked. |
| 301 |
|
| 302 |
This is provided by the block, or the first passed CODE reference. When |
| 303 |
the returned function is invoked, it repeatedly calls the block or wait |
| 304 |
function, until the CPS function has invoked its continuation. |
| 305 |
|
| 306 |
EXAMPLES |
| 307 |
|
| 308 |
Returning Data From Functions |
| 309 |
|
| 310 |
No facilities are provided directly to return data from CPS body |
| 311 |
functions in kloop, kpar and kseq. Instead, normal lexical variable |
| 312 |
capture may be used here. |
| 313 |
|
| 314 |
my $bat; |
| 315 |
my $ball; |
| 316 |
|
| 317 |
kpar( |
| 318 |
sub { |
| 319 |
my ( $k ) = @_; |
| 320 |
get_bat( on_bat => sub { $bat = shift; goto &$k } ); |
| 321 |
}, |
| 322 |
sub { |
| 323 |
my ( $k ) = @_; |
| 324 |
serve_ball( on_ball => sub { $ball = shift; goto &$k } ); |
| 325 |
}, |
| 326 |
|
| 327 |
sub { |
| 328 |
$bat->hit( $ball ); |
| 329 |
}, |
| 330 |
); |
| 331 |
|
| 332 |
The body function can set the value of a variable that it and its final |
| 333 |
continuation both capture. |
| 334 |
|
| 335 |
Using kseq For Conditionals |
| 336 |
|
| 337 |
Consider the call/return style of code |
| 338 |
|
| 339 |
A(); |
| 340 |
if( $maybe ) { |
| 341 |
B(); |
| 342 |
} |
| 343 |
C(); |
| 344 |
|
| 345 |
We cannot easily write this in CPS form without naming C twice |
| 346 |
|
| 347 |
kA( sub { |
| 348 |
$maybe ? |
| 349 |
kB( sub { kC() } ) : |
| 350 |
kC(); |
| 351 |
} ); |
| 352 |
|
| 353 |
While not so problematic here, it could get awkward if C were in fact a |
| 354 |
large code block, or if more than a single conditional were employed in |
| 355 |
the logic; a likely scenario. A further issue is that the logical |
| 356 |
structure becomes much harder to read. |
| 357 |
|
| 358 |
Using kseq allows us to name the continuation so each arm of kmaybe can |
| 359 |
invoke it indirectly. |
| 360 |
|
| 361 |
kseq( |
| 362 |
\&kA, |
| 363 |
sub { my $k = shift; $maybe ? kB( $k ) : goto &$k; }, |
| 364 |
\&kC |
| 365 |
); |
| 366 |
|
| 367 |
SEE ALSO |
| 368 |
|
| 369 |
* Future - represent an operation awaiting completion |
| 370 |
|
| 371 |
* Future::AsyncAwait - deferred subroutine syntax for futures |
| 372 |
|
| 373 |
* CPS::Functional - functional utilities in Continuation-Passing |
| 374 |
Style |
| 375 |
|
| 376 |
* http://en.wikipedia.org/wiki/Continuation-passing_style on |
| 377 |
wikipedia |
| 378 |
|
| 379 |
ACKNOWLEDGEMENTS |
| 380 |
|
| 381 |
Matt S. Trout (mst) <[email protected]> - for the inspiration of |
| 382 |
kpareach and with apologies to for naming of the said. ;) |
| 383 |
|
| 384 |
AUTHOR |
| 385 |
|
| 386 |
Paul Evans <[email protected]> |
| 387 |
|