Integrate perlio:
[p5sagit/p5-mst-13.2.git] / pod / perlopentut.pod
CommitLineData
f8284313 1=head1 NAME
2
3perlopentut - tutorial on opening things in Perl
4
5=head1 DESCRIPTION
6
7Perl has two simple, built-in ways to open files: the shell way for
8convenience, and the C way for precision. The choice is yours.
9
10=head1 Open E<agrave> la shell
11
12Perl's C<open> function was designed to mimic the way command-line
13redirection in the shell works. Here are some basic examples
14from the shell:
15
16 $ myprogram file1 file2 file3
17 $ myprogram < inputfile
18 $ myprogram > outputfile
19 $ myprogram >> outputfile
20 $ myprogram | otherprogram
21 $ otherprogram | myprogram
22
23And here are some more advanced examples:
24
25 $ otherprogram | myprogram f1 - f2
26 $ otherprogram 2>&1 | myprogram -
27 $ myprogram <&3
28 $ myprogram >&4
29
30Programmers accustomed to constructs like those above can take comfort
31in learning that Perl directly supports these familiar constructs using
32virtually the same syntax as the shell.
33
34=head2 Simple Opens
35
36The C<open> function takes two arguments: the first is a filehandle,
37and the second is a single string comprising both what to open and how
38to open it. C<open> returns true when it works, and when it fails,
39returns a false value and sets the special variable $! to reflect
40the system error. If the filehandle was previously opened, it will
41be implicitly closed first.
42
43For example:
44
45 open(INFO, "datafile") || die("can't open datafile: $!");
46 open(INFO, "< datafile") || die("can't open datafile: $!");
47 open(RESULTS,"> runstats") || die("can't open runstats: $!");
48 open(LOG, ">> logfile ") || die("can't open logfile: $!");
49
50If you prefer the low-punctuation version, you could write that this way:
51
52 open INFO, "< datafile" or die "can't open datafile: $!";
53 open RESULTS,"> runstats" or die "can't open runstats: $!";
54 open LOG, ">> logfile " or die "can't open logfile: $!";
55
56A few things to notice. First, the leading less-than is optional.
57If omitted, Perl assumes that you want to open the file for reading.
58
59The other important thing to notice is that, just as in the shell,
60any white space before or after the filename is ignored. This is good,
61because you wouldn't want these to do different things:
62
63 open INFO, "<datafile"
64 open INFO, "< datafile"
65 open INFO, "< datafile"
66
67Ignoring surround whitespace also helps for when you read a filename in
68from a different file, and forget to trim it before opening:
69
70 $filename = <INFO>; # oops, \n still there
71 open(EXTRA, "< $filename") || die "can't open $filename: $!";
72
73This is not a bug, but a feature. Because C<open> mimics the shell in
74its style of using redirection arrows to specify how to open the file, it
75also does so with respect to extra white space around the filename itself
13a2d996 76as well. For accessing files with naughty names, see
77L<"Dispelling the Dweomer">.
f8284313 78
79=head2 Pipe Opens
80
81In C, when you want to open a file using the standard I/O library,
82you use the C<fopen> function, but when opening a pipe, you use the
83C<popen> function. But in the shell, you just use a different redirection
84character. That's also the case for Perl. The C<open> call
85remains the same--just its argument differs.
86
f5daac4a 87If the leading character is a pipe symbol, C<open> starts up a new
f8284313 88command and open a write-only filehandle leading into that command.
89This lets you write into that handle and have what you write show up on
90that command's standard input. For example:
91
92 open(PRINTER, "| lpr -Plp1") || die "cannot fork: $!";
93 print PRINTER "stuff\n";
94 close(PRINTER) || die "can't close lpr: $!";
95
96If the trailing character is a pipe, you start up a new command and open a
97read-only filehandle leading out of that command. This lets whatever that
98command writes to its standard output show up on your handle for reading.
99For example:
100
101 open(NET, "netstat -i -n |") || die "cannot fork: $!";
102 while (<NET>) { } # do something with input
103 close(NET) || die "can't close netstat: $!";
104
105What happens if you try to open a pipe to or from a non-existent command?
106In most systems, such an C<open> will not return an error. That's
107because in the traditional C<fork>/C<exec> model, running the other
108program happens only in the forked child process, which means that
109the failed C<exec> can't be reflected in the return value of C<open>.
13a2d996 110Only a failed C<fork> shows up there. See
111L<perlfaq8/"Why doesn't open() return an error when a pipe open fails?">
112to see how to cope with this. There's also an explanation in L<perlipc>.
f8284313 113
114If you would like to open a bidirectional pipe, the IPC::Open2
13a2d996 115library will handle this for you. Check out
116L<perlipc/"Bidirectional Communication with Another Process">
f8284313 117
118=head2 The Minus File
119
120Again following the lead of the standard shell utilities, Perl's
121C<open> function treats a file whose name is a single minus, "-", in a
122special way. If you open minus for reading, it really means to access
123the standard input. If you open minus for writing, it really means to
124access the standard output.
125
40b7eeef 126If minus can be used as the default input or default output, what happens
f8284313 127if you open a pipe into or out of minus? What's the default command it
40b7eeef 128would run? The same script as you're currently running! This is actually
13a2d996 129a stealth C<fork> hidden inside an C<open> call. See
130L<perlipc/"Safe Pipe Opens"> for details.
f8284313 131
132=head2 Mixing Reads and Writes
133
134It is possible to specify both read and write access. All you do is
135add a "+" symbol in front of the redirection. But as in the shell,
136using a less-than on a file never creates a new file; it only opens an
137existing one. On the other hand, using a greater-than always clobbers
138(truncates to zero length) an existing file, or creates a brand-new one
139if there isn't an old one. Adding a "+" for read-write doesn't affect
140whether it only works on existing files or always clobbers existing ones.
141
142 open(WTMP, "+< /usr/adm/wtmp")
143 || die "can't open /usr/adm/wtmp: $!";
144
145 open(SCREEN, "+> /tmp/lkscreen")
146 || die "can't open /tmp/lkscreen: $!";
147
148 open(LOGFILE, "+>> /tmp/applog"
149 || die "can't open /tmp/applog: $!";
150
151The first one won't create a new file, and the second one will always
152clobber an old one. The third one will create a new file if necessary
153and not clobber an old one, and it will allow you to read at any point
154in the file, but all writes will always go to the end. In short,
155the first case is substantially more common than the second and third
156cases, which are almost always wrong. (If you know C, the plus in
157Perl's C<open> is historically derived from the one in C's fopen(3S),
158which it ultimately calls.)
159
160In fact, when it comes to updating a file, unless you're working on
161a binary file as in the WTMP case above, you probably don't want to
162use this approach for updating. Instead, Perl's B<-i> flag comes to
163the rescue. The following command takes all the C, C++, or yacc source
164or header files and changes all their foo's to bar's, leaving
165the old version in the original file name with a ".orig" tacked
166on the end:
167
168 $ perl -i.orig -pe 's/\bfoo\b/bar/g' *.[Cchy]
169
170This is a short cut for some renaming games that are really
171the best way to update textfiles. See the second question in
172L<perlfaq5> for more details.
173
174=head2 Filters
175
176One of the most common uses for C<open> is one you never
177even notice. When you process the ARGV filehandle using
c47ff5f1 178C<< <ARGV> >>, Perl actually does an implicit open
f8284313 179on each file in @ARGV. Thus a program called like this:
180
181 $ myprogram file1 file2 file3
182
183Can have all its files opened and processed one at a time
184using a construct no more complex than:
185
186 while (<>) {
187 # do something with $_
188 }
189
190If @ARGV is empty when the loop first begins, Perl pretends you've opened
191up minus, that is, the standard input. In fact, $ARGV, the currently
c47ff5f1 192open file during C<< <ARGV> >> processing, is even set to "-"
f8284313 193in these circumstances.
194
195You are welcome to pre-process your @ARGV before starting the loop to
196make sure it's to your liking. One reason to do this might be to remove
197command options beginning with a minus. While you can always roll the
198simple ones by hand, the Getopts modules are good for this.
199
200 use Getopt::Std;
201
202 # -v, -D, -o ARG, sets $opt_v, $opt_D, $opt_o
203 getopts("vDo:");
204
205 # -v, -D, -o ARG, sets $args{v}, $args{D}, $args{o}
206 getopts("vDo:", \%args);
207
208Or the standard Getopt::Long module to permit named arguments:
209
210 use Getopt::Long;
211 GetOptions( "verbose" => \$verbose, # --verbose
212 "Debug" => \$debug, # --Debug
213 "output=s" => \$output );
214 # --output=somestring or --output somestring
215
216Another reason for preprocessing arguments is to make an empty
217argument list default to all files:
218
219 @ARGV = glob("*") unless @ARGV;
220
221You could even filter out all but plain, text files. This is a bit
222silent, of course, and you might prefer to mention them on the way.
223
224 @ARGV = grep { -f && -T } @ARGV;
225
226If you're using the B<-n> or B<-p> command-line options, you
227should put changes to @ARGV in a C<BEGIN{}> block.
228
229Remember that a normal C<open> has special properties, in that it might
230call fopen(3S) or it might called popen(3S), depending on what its
231argument looks like; that's why it's sometimes called "magic open".
232Here's an example:
233
234 $pwdinfo = `domainname` =~ /^(\(none\))?$/
235 ? '< /etc/passwd'
236 : 'ypcat passwd |';
237
238 open(PWD, $pwdinfo)
239 or die "can't open $pwdinfo: $!";
240
241This sort of thing also comes into play in filter processing. Because
c47ff5f1 242C<< <ARGV> >> processing employs the normal, shell-style Perl C<open>,
f8284313 243it respects all the special things we've already seen:
244
245 $ myprogram f1 "cmd1|" - f2 "cmd2|" f3 < tmpfile
246
247That program will read from the file F<f1>, the process F<cmd1>, standard
248input (F<tmpfile> in this case), the F<f2> file, the F<cmd2> command,
249and finally the F<f3> file.
250
251Yes, this also means that if you have a file named "-" (and so on) in
252your directory, that they won't be processed as literal files by C<open>.
253You'll need to pass them as "./-" much as you would for the I<rm> program.
254Or you could use C<sysopen> as described below.
255
256One of the more interesting applications is to change files of a certain
257name into pipes. For example, to autoprocess gzipped or compressed
258files by decompressing them with I<gzip>:
259
260 @ARGV = map { /^\.(gz|Z)$/ ? "gzip -dc $_ |" : $_ } @ARGV;
261
262Or, if you have the I<GET> program installed from LWP,
263you can fetch URLs before processing them:
264
265 @ARGV = map { m#^\w+://# ? "GET $_ |" : $_ } @ARGV;
266
c47ff5f1 267It's not for nothing that this is called magic C<< <ARGV> >>.
f8284313 268Pretty nifty, eh?
269
270=head1 Open E<agrave> la C
271
272If you want the convenience of the shell, then Perl's C<open> is
273definitely the way to go. On the other hand, if you want finer precision
274than C's simplistic fopen(3S) provides, then you should look to Perl's
275C<sysopen>, which is a direct hook into the open(2) system call.
276That does mean it's a bit more involved, but that's the price of
277precision.
278
279C<sysopen> takes 3 (or 4) arguments.
280
281 sysopen HANDLE, PATH, FLAGS, [MASK]
282
283The HANDLE argument is a filehandle just as with C<open>. The PATH is
284a literal path, one that doesn't pay attention to any greater-thans or
285less-thans or pipes or minuses, nor ignore white space. If it's there,
286it's part of the path. The FLAGS argument contains one or more values
287derived from the Fcntl module that have been or'd together using the
288bitwise "|" operator. The final argument, the MASK, is optional; if
289present, it is combined with the user's current umask for the creation
290mode of the file. You should usually omit this.
291
292Although the traditional values of read-only, write-only, and read-write
293are 0, 1, and 2 respectively, this is known not to hold true on some
294systems. Instead, it's best to load in the appropriate constants first
295from the Fcntl module, which supplies the following standard flags:
296
297 O_RDONLY Read only
298 O_WRONLY Write only
299 O_RDWR Read and write
300 O_CREAT Create the file if it doesn't exist
301 O_EXCL Fail if the file already exists
302 O_APPEND Append to the file
303 O_TRUNC Truncate the file
304 O_NONBLOCK Non-blocking access
305
ca6e1c26 306Less common flags that are sometimes available on some operating
307systems include C<O_BINARY>, C<O_TEXT>, C<O_SHLOCK>, C<O_EXLOCK>,
308C<O_DEFER>, C<O_SYNC>, C<O_ASYNC>, C<O_DSYNC>, C<O_RSYNC>,
309C<O_NOCTTY>, C<O_NDELAY> and C<O_LARGEFILE>. Consult your open(2)
310manpage or its local equivalent for details. (Note: starting from
311Perl release 5.6 the O_LARGEFILE flag, if available, is automatically
106325ad 312added to the sysopen() flags because large files are the default.)
f8284313 313
314Here's how to use C<sysopen> to emulate the simple C<open> calls we had
315before. We'll omit the C<|| die $!> checks for clarity, but make sure
316you always check the return values in real code. These aren't quite
317the same, since C<open> will trim leading and trailing white space,
318but you'll get the idea:
319
320To open a file for reading:
321
322 open(FH, "< $path");
323 sysopen(FH, $path, O_RDONLY);
324
325To open a file for writing, creating a new file if needed or else truncating
326an old file:
327
328 open(FH, "> $path");
329 sysopen(FH, $path, O_WRONLY | O_TRUNC | O_CREAT);
330
331To open a file for appending, creating one if necessary:
332
333 open(FH, ">> $path");
334 sysopen(FH, $path, O_WRONLY | O_APPEND | O_CREAT);
335
336To open a file for update, where the file must already exist:
337
338 open(FH, "+< $path");
339 sysopen(FH, $path, O_RDWR);
340
341And here are things you can do with C<sysopen> that you cannot do with
342a regular C<open>. As you see, it's just a matter of controlling the
343flags in the third argument.
344
345To open a file for writing, creating a new file which must not previously
346exist:
347
348 sysopen(FH, $path, O_WRONLY | O_EXCL | O_CREAT);
349
350To open a file for appending, where that file must already exist:
351
352 sysopen(FH, $path, O_WRONLY | O_APPEND);
353
354To open a file for update, creating a new file if necessary:
355
356 sysopen(FH, $path, O_RDWR | O_CREAT);
357
358To open a file for update, where that file must not already exist:
359
360 sysopen(FH, $path, O_RDWR | O_EXCL | O_CREAT);
361
362To open a file without blocking, creating one if necessary:
363
364 sysopen(FH, $path, O_WRONLY | O_NONBLOCK | O_CREAT);
365
366=head2 Permissions E<agrave> la mode
367
368If you omit the MASK argument to C<sysopen>, Perl uses the octal value
3690666. The normal MASK to use for executables and directories should
370be 0777, and for anything else, 0666.
371
372Why so permissive? Well, it isn't really. The MASK will be modified
373by your process's current C<umask>. A umask is a number representing
374I<disabled> permissions bits; that is, bits that will not be turned on
375in the created files' permissions field.
376
377For example, if your C<umask> were 027, then the 020 part would
378disable the group from writing, and the 007 part would disable others
379from reading, writing, or executing. Under these conditions, passing
380C<sysopen> 0666 would create a file with mode 0640, since C<0666 &~ 027>
381is 0640.
382
383You should seldom use the MASK argument to C<sysopen()>. That takes
384away the user's freedom to choose what permission new files will have.
385Denying choice is almost always a bad thing. One exception would be for
386cases where sensitive or private data is being stored, such as with mail
387folders, cookie files, and internal temporary files.
388
389=head1 Obscure Open Tricks
390
391=head2 Re-Opening Files (dups)
392
393Sometimes you already have a filehandle open, and want to make another
394handle that's a duplicate of the first one. In the shell, we place an
395ampersand in front of a file descriptor number when doing redirections.
c47ff5f1 396For example, C<< 2>&1 >> makes descriptor 2 (that's STDERR in Perl)
f8284313 397be redirected into descriptor 1 (which is usually Perl's STDOUT).
398The same is essentially true in Perl: a filename that begins with an
399ampersand is treated instead as a file descriptor if a number, or as a
400filehandle if a string.
401
402 open(SAVEOUT, ">&SAVEERR") || die "couldn't dup SAVEERR: $!";
403 open(MHCONTEXT, "<&4") || die "couldn't dup fd4: $!";
404
405That means that if a function is expecting a filename, but you don't
406want to give it a filename because you already have the file open, you
407can just pass the filehandle with a leading ampersand. It's best to
408use a fully qualified handle though, just in case the function happens
409to be in a different package:
410
411 somefunction("&main::LOGFILE");
412
413This way if somefunction() is planning on opening its argument, it can
414just use the already opened handle. This differs from passing a handle,
415because with a handle, you don't open the file. Here you have something
416you can pass to open.
417
418If you have one of those tricky, newfangled I/O objects that the C++
419folks are raving about, then this doesn't work because those aren't a
420proper filehandle in the native Perl sense. You'll have to use fileno()
421to pull out the proper descriptor number, assuming you can:
422
423 use IO::Socket;
424 $handle = IO::Socket::INET->new("www.perl.com:80");
425 $fd = $handle->fileno;
426 somefunction("&$fd"); # not an indirect function call
427
428It can be easier (and certainly will be faster) just to use real
429filehandles though:
430
431 use IO::Socket;
432 local *REMOTE = IO::Socket::INET->new("www.perl.com:80");
433 die "can't connect" unless defined(fileno(REMOTE));
434 somefunction("&main::REMOTE");
435
436If the filehandle or descriptor number is preceded not just with a simple
437"&" but rather with a "&=" combination, then Perl will not create a
438completely new descriptor opened to the same place using the dup(2)
439system call. Instead, it will just make something of an alias to the
440existing one using the fdopen(3S) library call This is slightly more
441parsimonious of systems resources, although this is less a concern
442these days. Here's an example of that:
443
444 $fd = $ENV{"MHCONTEXTFD"};
445 open(MHCONTEXT, "<&=$fd") or die "couldn't fdopen $fd: $!";
446
c47ff5f1 447If you're using magic C<< <ARGV> >>, you could even pass in as a
448command line argument in @ARGV something like C<"<&=$MHCONTEXTFD">,
f8284313 449but we've never seen anyone actually do this.
450
451=head2 Dispelling the Dweomer
452
453Perl is more of a DWIMmer language than something like Java--where DWIM
454is an acronym for "do what I mean". But this principle sometimes leads
455to more hidden magic than one knows what to do with. In this way, Perl
456is also filled with I<dweomer>, an obscure word meaning an enchantment.
457Sometimes, Perl's DWIMmer is just too much like dweomer for comfort.
458
459If magic C<open> is a bit too magical for you, you don't have to turn
460to C<sysopen>. To open a file with arbitrary weird characters in
461it, it's necessary to protect any leading and trailing whitespace.
462Leading whitespace is protected by inserting a C<"./"> in front of a
463filename that starts with whitespace. Trailing whitespace is protected
464by appending an ASCII NUL byte (C<"\0">) at the end off the string.
465
466 $file =~ s#^(\s)#./$1#;
467 open(FH, "< $file\0") || die "can't open $file: $!";
468
469This assumes, of course, that your system considers dot the current
470working directory, slash the directory separator, and disallows ASCII
471NULs within a valid filename. Most systems follow these conventions,
472including all POSIX systems as well as proprietary Microsoft systems.
473The only vaguely popular system that doesn't work this way is the
474proprietary Macintosh system, which uses a colon where the rest of us
475use a slash. Maybe C<sysopen> isn't such a bad idea after all.
476
c47ff5f1 477If you want to use C<< <ARGV> >> processing in a totally boring
f8284313 478and non-magical way, you could do this first:
479
480 # "Sam sat on the ground and put his head in his hands.
481 # 'I wish I had never come here, and I don't want to see
482 # no more magic,' he said, and fell silent."
483 for (@ARGV) {
484 s#^([^./])#./$1#;
485 $_ .= "\0";
486 }
487 while (<>) {
488 # now process $_
489 }
490
491But be warned that users will not appreciate being unable to use "-"
492to mean standard input, per the standard convention.
493
494=head2 Paths as Opens
495
496You've probably noticed how Perl's C<warn> and C<die> functions can
497produce messages like:
498
1761cee5 499 Some warning at scriptname line 29, <FH> line 7.
f8284313 500
501That's because you opened a filehandle FH, and had read in seven records
502from it. But what was the name of the file, not the handle?
503
504If you aren't running with C<strict refs>, or if you've turn them off
505temporarily, then all you have to do is this:
506
507 open($path, "< $path") || die "can't open $path: $!";
508 while (<$path>) {
509 # whatever
510 }
511
512Since you're using the pathname of the file as its handle,
513you'll get warnings more like
514
1761cee5 515 Some warning at scriptname line 29, </etc/motd> line 7.
f8284313 516
517=head2 Single Argument Open
518
519Remember how we said that Perl's open took two arguments? That was a
520passive prevarication. You see, it can also take just one argument.
521If and only if the variable is a global variable, not a lexical, you
522can pass C<open> just one argument, the filehandle, and it will
523get the path from the global scalar variable of the same name.
524
525 $FILE = "/etc/motd";
526 open FILE or die "can't open $FILE: $!";
527 while (<FILE>) {
528 # whatever
529 }
530
531Why is this here? Someone has to cater to the hysterical porpoises.
532It's something that's been in Perl since the very beginning, if not
533before.
534
535=head2 Playing with STDIN and STDOUT
536
537One clever move with STDOUT is to explicitly close it when you're done
538with the program.
539
540 END { close(STDOUT) || die "can't close stdout: $!" }
541
542If you don't do this, and your program fills up the disk partition due
543to a command line redirection, it won't report the error exit with a
544failure status.
545
546You don't have to accept the STDIN and STDOUT you were given. You are
547welcome to reopen them if you'd like.
548
549 open(STDIN, "< datafile")
550 || die "can't open datafile: $!";
551
552 open(STDOUT, "> output")
553 || die "can't open output: $!";
554
555And then these can be read directly or passed on to subprocesses.
556This makes it look as though the program were initially invoked
557with those redirections from the command line.
558
559It's probably more interesting to connect these to pipes. For example:
560
561 $pager = $ENV{PAGER} || "(less || more)";
562 open(STDOUT, "| $pager")
563 || die "can't fork a pager: $!";
564
565This makes it appear as though your program were called with its stdout
566already piped into your pager. You can also use this kind of thing
567in conjunction with an implicit fork to yourself. You might do this
568if you would rather handle the post processing in your own program,
569just in a different process:
570
571 head(100);
572 while (<>) {
573 print;
574 }
575
576 sub head {
577 my $lines = shift || 20;
578 return unless $pid = open(STDOUT, "|-");
579 die "cannot fork: $!" unless defined $pid;
580 while (<STDIN>) {
581 print;
582 last if --$lines < 0;
583 }
584 exit;
585 }
586
587This technique can be applied to repeatedly push as many filters on your
588output stream as you wish.
589
590=head1 Other I/O Issues
591
592These topics aren't really arguments related to C<open> or C<sysopen>,
593but they do affect what you do with your open files.
594
595=head2 Opening Non-File Files
596
597When is a file not a file? Well, you could say when it exists but
598isn't a plain file. We'll check whether it's a symbolic link first,
599just in case.
600
601 if (-l $file || ! -f _) {
602 print "$file is not a plain file\n";
603 }
604
605What other kinds of files are there than, well, files? Directories,
606symbolic links, named pipes, Unix-domain sockets, and block and character
607devices. Those are all files, too--just not I<plain> files. This isn't
608the same issue as being a text file. Not all text files are plain files.
609Not all plain files are textfiles. That's why there are separate C<-f>
610and C<-T> file tests.
611
612To open a directory, you should use the C<opendir> function, then
613process it with C<readdir>, carefully restoring the directory
614name if necessary:
615
616 opendir(DIR, $dirname) or die "can't opendir $dirname: $!";
617 while (defined($file = readdir(DIR))) {
618 # do something with "$dirname/$file"
619 }
620 closedir(DIR);
621
622If you want to process directories recursively, it's better to use the
623File::Find module. For example, this prints out all files recursively,
624add adds a slash to their names if the file is a directory.
625
626 @ARGV = qw(.) unless @ARGV;
627 use File::Find;
628 find sub { print $File::Find::name, -d && '/', "\n" }, @ARGV;
629
630This finds all bogus symbolic links beneath a particular directory:
631
632 find sub { print "$File::Find::name\n" if -l && !-e }, $dir;
633
634As you see, with symbolic links, you can just pretend that it is
635what it points to. Or, if you want to know I<what> it points to, then
636C<readlink> is called for:
637
638 if (-l $file) {
639 if (defined($whither = readlink($file))) {
640 print "$file points to $whither\n";
641 } else {
642 print "$file points nowhere: $!\n";
643 }
644 }
645
646Named pipes are a different matter. You pretend they're regular files,
647but their opens will normally block until there is both a reader and
648a writer. You can read more about them in L<perlipc/"Named Pipes">.
649Unix-domain sockets are rather different beasts as well; they're
650described in L<perlipc/"Unix-Domain TCP Clients and Servers">.
651
652When it comes to opening devices, it can be easy and it can tricky.
653We'll assume that if you're opening up a block device, you know what
654you're doing. The character devices are more interesting. These are
655typically used for modems, mice, and some kinds of printers. This is
656described in L<perlfaq8/"How do I read and write the serial port?">
657It's often enough to open them carefully:
658
659 sysopen(TTYIN, "/dev/ttyS1", O_RDWR | O_NDELAY | O_NOCTTY)
660 # (O_NOCTTY no longer needed on POSIX systems)
661 or die "can't open /dev/ttyS1: $!";
662 open(TTYOUT, "+>&TTYIN")
663 or die "can't dup TTYIN: $!";
664
665 $ofh = select(TTYOUT); $| = 1; select($ofh);
666
667 print TTYOUT "+++at\015";
668 $answer = <TTYIN>;
669
670With descriptors that you haven't opened using C<sysopen>, such as a
671socket, you can set them to be non-blocking using C<fcntl>:
672
673 use Fcntl;
674 fcntl(Connection, F_SETFL, O_NONBLOCK)
675 or die "can't set non blocking: $!";
676
677Rather than losing yourself in a morass of twisting, turning C<ioctl>s,
678all dissimilar, if you're going to manipulate ttys, it's best to
679make calls out to the stty(1) program if you have it, or else use the
680portable POSIX interface. To figure this all out, you'll need to read the
681termios(3) manpage, which describes the POSIX interface to tty devices,
682and then L<POSIX>, which describes Perl's interface to POSIX. There are
683also some high-level modules on CPAN that can help you with these games.
684Check out Term::ReadKey and Term::ReadLine.
685
686What else can you open? To open a connection using sockets, you won't use
13a2d996 687one of Perl's two open functions. See
688L<perlipc/"Sockets: Client/Server Communication"> for that. Here's an
689example. Once you have it, you can use FH as a bidirectional filehandle.
f8284313 690
691 use IO::Socket;
692 local *FH = IO::Socket::INET->new("www.perl.com:80");
693
694For opening up a URL, the LWP modules from CPAN are just what
695the doctor ordered. There's no filehandle interface, but
696it's still easy to get the contents of a document:
697
698 use LWP::Simple;
6cecdcac 699 $doc = get('http://www.linpro.no/lwp/');
f8284313 700
701=head2 Binary Files
702
703On certain legacy systems with what could charitably be called terminally
704convoluted (some would say broken) I/O models, a file isn't a file--at
705least, not with respect to the C standard I/O library. On these old
706systems whose libraries (but not kernels) distinguish between text and
707binary streams, to get files to behave properly you'll have to bend over
708backwards to avoid nasty problems. On such infelicitous systems, sockets
709and pipes are already opened in binary mode, and there is currently no
710way to turn that off. With files, you have more options.
711
712Another option is to use the C<binmode> function on the appropriate
713handles before doing regular I/O on them:
714
715 binmode(STDIN);
716 binmode(STDOUT);
717 while (<STDIN>) { print }
718
719Passing C<sysopen> a non-standard flag option will also open the file in
720binary mode on those systems that support it. This is the equivalent of
721opening the file normally, then calling C<binmode>ing on the handle.
722
723 sysopen(BINDAT, "records.data", O_RDWR | O_BINARY)
724 || die "can't open records.data: $!";
725
726Now you can use C<read> and C<print> on that handle without worrying
727about the system non-standard I/O library breaking your data. It's not
728a pretty picture, but then, legacy systems seldom are. CP/M will be
729with us until the end of days, and after.
730
731On systems with exotic I/O systems, it turns out that, astonishingly
732enough, even unbuffered I/O using C<sysread> and C<syswrite> might do
733sneaky data mutilation behind your back.
734
735 while (sysread(WHENCE, $buf, 1024)) {
736 syswrite(WHITHER, $buf, length($buf));
737 }
738
739Depending on the vicissitudes of your runtime system, even these calls
740may need C<binmode> or C<O_BINARY> first. Systems known to be free of
741such difficulties include Unix, the Mac OS, Plan9, and Inferno.
742
743=head2 File Locking
744
745In a multitasking environment, you may need to be careful not to collide
746with other processes who want to do I/O on the same files as others
747are working on. You'll often need shared or exclusive locks
748on files for reading and writing respectively. You might just
749pretend that only exclusive locks exist.
750
751Never use the existence of a file C<-e $file> as a locking indication,
752because there is a race condition between the test for the existence of
753the file and its creation. Atomicity is critical.
754
755Perl's most portable locking interface is via the C<flock> function,
756whose simplicity is emulated on systems that don't directly support it,
757such as SysV or WindowsNT. The underlying semantics may affect how
758it all works, so you should learn how C<flock> is implemented on your
759system's port of Perl.
760
761File locking I<does not> lock out another process that would like to
762do I/O. A file lock only locks out others trying to get a lock, not
763processes trying to do I/O. Because locks are advisory, if one process
764uses locking and another doesn't, all bets are off.
765
766By default, the C<flock> call will block until a lock is granted.
767A request for a shared lock will be granted as soon as there is no
d1be9408 768exclusive locker. A request for an exclusive lock will be granted as
f8284313 769soon as there is no locker of any kind. Locks are on file descriptors,
770not file names. You can't lock a file until you open it, and you can't
771hold on to a lock once the file has been closed.
772
773Here's how to get a blocking shared lock on a file, typically used
774for reading:
775
776 use 5.004;
777 use Fcntl qw(:DEFAULT :flock);
778 open(FH, "< filename") or die "can't open filename: $!";
779 flock(FH, LOCK_SH) or die "can't lock filename: $!";
780 # now read from FH
781
782You can get a non-blocking lock by using C<LOCK_NB>.
783
784 flock(FH, LOCK_SH | LOCK_NB)
785 or die "can't lock filename: $!";
786
787This can be useful for producing more user-friendly behaviour by warning
788if you're going to be blocking:
789
790 use 5.004;
791 use Fcntl qw(:DEFAULT :flock);
792 open(FH, "< filename") or die "can't open filename: $!";
793 unless (flock(FH, LOCK_SH | LOCK_NB)) {
794 $| = 1;
795 print "Waiting for lock...";
796 flock(FH, LOCK_SH) or die "can't lock filename: $!";
797 print "got it.\n"
798 }
799 # now read from FH
800
801To get an exclusive lock, typically used for writing, you have to be
802careful. We C<sysopen> the file so it can be locked before it gets
803emptied. You can get a nonblocking version using C<LOCK_EX | LOCK_NB>.
804
805 use 5.004;
806 use Fcntl qw(:DEFAULT :flock);
807 sysopen(FH, "filename", O_WRONLY | O_CREAT)
808 or die "can't open filename: $!";
809 flock(FH, LOCK_EX)
810 or die "can't lock filename: $!";
811 truncate(FH, 0)
812 or die "can't truncate filename: $!";
813 # now write to FH
814
815Finally, due to the uncounted millions who cannot be dissuaded from
816wasting cycles on useless vanity devices called hit counters, here's
817how to increment a number in a file safely:
818
819 use Fcntl qw(:DEFAULT :flock);
820
821 sysopen(FH, "numfile", O_RDWR | O_CREAT)
822 or die "can't open numfile: $!";
823 # autoflush FH
824 $ofh = select(FH); $| = 1; select ($ofh);
825 flock(FH, LOCK_EX)
826 or die "can't write-lock numfile: $!";
827
828 $num = <FH> || 0;
829 seek(FH, 0, 0)
830 or die "can't rewind numfile : $!";
831 print FH $num+1, "\n"
832 or die "can't write numfile: $!";
833
834 truncate(FH, tell(FH))
835 or die "can't truncate numfile: $!";
836 close(FH)
837 or die "can't close numfile: $!";
838
839=head1 SEE ALSO
840
841The C<open> and C<sysopen> function in perlfunc(1);
842the standard open(2), dup(2), fopen(3), and fdopen(3) manpages;
843the POSIX documentation.
844
845=head1 AUTHOR and COPYRIGHT
846
847Copyright 1998 Tom Christiansen.
848
5a7beb56 849This documentation is free; you can redistribute it and/or modify it
850under the same terms as Perl itself.
f8284313 851
852Irrespective of its distribution, all code examples in these files are
853hereby placed into the public domain. You are permitted and
854encouraged to use this code in your own programs for fun or for profit
855as you see fit. A simple comment in the code giving credit would be
856courteous but is not required.
857
858=head1 HISTORY
859
860First release: Sat Jan 9 08:09:11 MST 1999