perlfaq5 - Files and Formats ($Revision: 1.22 $,
$Date: 1997/04/24 22:44:02 $)
This section deals with I/O
and the ``f'' issues: filehandles, flushing, formats, and footers.
The C standard I/O library (stdio) normally buffers characters sent to devices.
This is done for efficiency reasons, so that there isn't a system call for each byte. Any
time you use print() or write() in Perl, you go though this
buffering. syswrite() circumvents stdio and buffering.
In most stdio implementations, the type of buffering
and the size of the buffer varies according to the type of device. Disk files are block
buffered, often with a buffer size of more than 2k. Pipes and sockets are often buffered
with a buffer size between 1/2 and 2k. Serial devices (e.g. modems, terminals) are
normally line-buffered, and stdio sends the entire line when it gets the newline.
Perl does not support truly unbuffered output (except
insofar as you can syswrite(OUT, $char, 1)). What it does instead support is
``command buffering'', in which a physical write is performed after every output command.
This isn't as hard on your system as unbuffering, but does get the output where you want
it when you want it.
If you expect characters to get to your device when
you print them there, you'll want to autoflush its handle, as in the older:
use FileHandle;
open(DEV, "<+/dev/tty"); # ceci n'est pas une pipe
DEV->autoflush(1);
or the newer IO::*
modules:
use IO::Handle;
open(DEV, ">/dev/printer"); # but is this?
DEV->autoflush(1);
or even this:
use IO::Socket; # this one is kinda a pipe?
$sock = IO::Socket::INET->new(PeerAddr => 'www.perl.com',
PeerPort => 'http(80)',
Proto => 'tcp');
die "$!" unless $sock;
$sock->autoflush();
$sock->print("GET /\015\012");
$document = join('', $sock->getlines());
print "DOC IS: $document\n";
Note the hardcoded carriage return and newline in
their octal equivalents. This is the ONLY way
(currently) to assure a proper flush on all platforms, including Macintosh.
You can use select() and the $|
variable to control autoflushing (see $| and select):
$oldh = select(DEV);
$| = 1;
select($oldh);
You'll also see code that does this without a
temporary variable, as in
select((select(DEV), $| = 1)[0]);
Although humans have an easy time thinking of a text
file as being a sequence of lines that operates much like a stack of playing cards -- or
punch cards -- computers usually see the text file as a sequence of bytes. In general,
there's no direct way for Perl to seek to a particular line of a file, insert text into a
file, or remove text from a file.
(There are exceptions in special circumstances.
Replacing a sequence of bytes with another sequence of the same length is one. Another is
using the $DB_RECNO array bindings as documented in the DB_File manpage.
Yet another is manipulating files with all lines the same length.)
The general solution is to create a temporary copy of
the text file with the changes you want, then copy that over the original.
$old = $file;
$new = "$file.tmp.$$";
$bak = "$file.bak";
open(OLD, "< $old") or die "can't open $old: $!";
open(NEW, "> $new") or die "can't open $new: $!";
# Correct typos, preserving case
while (<OLD>) {
s/\b(p)earl\b/${1}erl/i;
(print NEW $_) or die "can't write to $new: $!";
}
close(OLD) or die "can't close $old: $!";
close(NEW) or die "can't close $new: $!";
rename($old, $bak) or die "can't rename $old to $bak: $!";
rename($new, $old) or die "can't rename $new to $old: $!";
Perl can do this sort of thing for you automatically
with the -i command-line switch or the closely-related $^I
variable (see the perlrun manpage
for more details). Note that -i may require a suffix on some non-Unix
systems; see the platform-specific documentation that came with your port.
# Renumber a series of tests from the command line
perl -pi -e 's/(^\s+test\s+)\d+/ $1 . ++$count /e' t/op/taint.t
# form a script
local($^I, @ARGV) = ('.bak', glob("*.c"));
while (<>) {
if ($. == 1) {
print "This line should appear at the top of each file\n";
}
s/\b(p)earl\b/${1}erl/i; # Correct typos, preserving case
print;
close ARGV if eof; # Reset $.
}
If you need to seek to an arbitrary line of a file
that changes infrequently, you could build up an index of byte positions of where the line
ends are in the file. If the file is large, an index of every tenth or hundredth line end
would allow you to seek and read fairly efficiently. If the file is sorted, try the
look.pl library (part of the standard perl distribution).
In the unique case of deleting lines at the end of a
file, you can use tell() and truncate(). The following code
snippet deletes the last line of a file without making a copy or reading the whole file
into memory:
open (FH, "+< $file");
while ( <FH> ) { $addr = tell(FH) unless eof(FH) }
truncate(FH, $addr);
Error checking is left as an exercise for the reader.
One fairly efficient way is to count newlines in the
file. The following program uses a feature of tr///, as documented in the perlop manpage.
If your text file doesn't end with a newline, then it's not really a proper text file, so
this may report one fewer line than you expect.
$lines = 0;
open(FILE, $filename) or die "Can't open `$filename': $!";
while (sysread FILE, $buffer, 4096) {
$lines += ($buffer =~ tr/\n//);
}
close FILE;
Use the process ID
and/or the current time-value. If you need to have many temporary files in one process,
use a counter:
BEGIN {
use IO::File;
use Fcntl;
my $temp_dir = -d '/tmp' ? '/tmp' : $ENV{TMP} || $ENV{TEMP};
my $base_name = sprintf("%s/%d-%d-0000", $temp_dir, $$, time());
sub temp_file {
my $fh = undef;
my $count = 0;
until (defined($fh) || $count > 100) {
$base_name =~ s/-(\d+)$/"-" . (1 + $1)/e;
$fh = IO::File->new($base_name, O_WRONLY|O_EXCL|O_CREAT, 0644)
}
if (defined($fh)) {
return ($fh, $base_name);
} else {
return ();
}
}
}
Or you could simply use IO::Handle::new_tmpfile.
The most efficient way is using pack()
and unpack(). This is faster than using substr(). Here is a
sample chunk of code to break up and put back together again some fixed-format input
lines, in this case from the output of a normal, Berkeley-style ps:
# sample input line:
# 15158 p5 T 0:00 perl /home/tchrist/scripts/now-what
$PS_T = 'A6 A4 A7 A5 A*';
open(PS, "ps|");
$_ = <PS>; print;
while (<PS>) {
($pid, $tt, $stat, $time, $command) = unpack($PS_T, $_);
for $var (qw!pid tt stat time command!) {
print "$var: <$$var>\n";
}
print 'line=', pack($PS_T, $pid, $tt, $stat, $time, $command),
"\n";
}
You may have some success with typeglobs, as we
always had to use in days of old:
local(*FH);
But while still supported, that isn't the best to go
about getting local filehandles. Typeglobs have their drawbacks. You may well want to use
the FileHandle module, which creates new filehandles for you (see the FileHandle
manpage):
use FileHandle;
sub findme {
my $fh = FileHandle->new();
open($fh, "</etc/hosts") or die "no /etc/hosts: $!";
while (<$fh>) {
print if /\b127\.(0\.0\.)?1\b/;
}
# $fh automatically closes/disappears here
}
Internally, Perl believes filehandles to be of class
IO::Handle. You may use that module directly if you'd like (see Handle), or one
of its more specific derived classes.
Once you have IO::File or FileHandle objects, you can
pass them between subroutines or store them in hashes as you would any other scalar
values:
use FileHandle;
# Storing filehandles in a hash and array
foreach $filename (@names) {
my $fh = new FileHandle($filename) or die;
$file{$filename} = $fh;
push(@files, $fh);
}
# Using the filehandles in the array
foreach $file (@files) {
print $file "Testing\n";
}
# You have to do the { } ugliness when you're specifying the
# filehandle by anything other than a simple scalar variable.
print { $files[2] } "Testing\n";
# Passing filehandles to subroutines
sub debug {
my $filehandle = shift;
printf $filehandle "DEBUG: ", @_;
}
debug($fh, "Testing\n");
There's no builtin way to do this, but the perlform manpage
has a couple of techniques to make it possible for the intrepid hacker.
See the perlform manpage
for an swrite() function.
This one will do it for you:
sub commify {
local $_ = shift;
1 while s/^(-?\d+)(\d{3})/$1,$2/;
return $_;
}
$n = 23659019423.2331;
print "GOT: ", commify($n), "\n";
GOT: 23,659,019,423.2331
You can't just:
s/^(-?\d+)(\d{3})/$1,$2/g;
because you have to put the comma in and then
recalculate your position.
Alternatively, this commifies all numbers in a line
regardless of whether they have decimal portions, are preceded by + or -, or whatever:
# from Andrew Johnson <ajohnson@gpu.srv.ualberta.ca>
sub commify {
my $input = shift;
$input = reverse $input;
$input =~ s<(\d\d\d)(?=\d)(?!\d*\.)><$1,>g;
return reverse $input;
}
Use the <> (glob()) operator, documented in the perlfunc manpage.
This requires that you have a shell installed that groks tildes, meaning csh or tcsh or
(some versions of) ksh, and thus may have portability problems. The Glob::KGlob module
(available from CPAN) gives more portable glob
functionality.
Within Perl, you may use this directly:
$filename =~ s{
^ ~ # find a leading tilde
( # save this in $1
[^/] # a non-slash character
* # repeated 0 or more times (0 means me)
)
}{
$1
? (getpwnam($1))[7]
: ( $ENV{HOME} || $ENV{LOGDIR} )
}ex;
Because you're using something like this, which
truncates the file and then gives you read-write access:
open(FH, "+> /path/name"); # WRONG
Whoops. You should instead use this, which will fail
if the file doesn't exist.
open(FH, "+< /path/name"); # open for update
If this is an issue, try:
sysopen(FH, "/path/name", O_RDWR|O_CREAT, 0644);
Error checking is left as an exercise for the reader.
The <> operator performs a
globbing operation (see above). By default glob() forks csh(1)
to do the actual glob expansion, but csh can't handle more than 127 items and so gives the
error message Argument list too long. People who installed tcsh as csh won't
have this problem, but their users may be surprised by it.
To get around this, either do the glob yourself with Dirhandles
and patterns, or use a module like Glob::KGlob, one that doesn't use the shell to do
globbing.
Due to the current implementation on some operating
systems, when you use the glob() function or its angle-bracket alias in a
scalar context, you may cause a leak and/or unpredictable behavior. It's best therefore to
use glob() only in list context.
Normally perl ignores trailing blanks in filenames,
and interprets certain leading characters (or a trailing ``|'') to mean something special.
To avoid this, you might want to use a routine like this. It makes incomplete pathnames
into explicit relative ones, and tacks a trailing null byte on the name to make perl leave
it alone:
sub safe_filename {
local $_ = shift;
return m#^/#
? "$_\0"
: "./$_\0";
}
$fn = safe_filename("<<<something really wicked ");
open(FH, "> $fn") or "couldn't open $fn: $!";
You could also use the sysopen()
function (see sysopen).
Well, usually you just use Perl's rename()
function. But that may not work everywhere, in particular, renaming files across file
systems. If your operating system supports a mv(1) program or its moral
equivalent, this works:
rename($old, $new) or system("mv", $old, $new);
It may be more compelling to use the File::Copy
module instead. You just copy to the new file to the new name (checking return values),
then delete the old one. This isn't really the same semantics as a real rename(),
though, which preserves metainformation like permissions, timestamps, inode info, etc.
Perl's builtin flock() function (see the perlfunc manpage
for details) will call flock(2) if that exists, fcntl(2) if it
doesn't (on perl version 5.004 and later), and lockf(3) if neither of the two
previous system calls exists. On some systems, it may even use a different form of native
locking. Here are some gotchas with Perl's flock():
- Produces a fatal error if none of the three system
calls (or their close equivalent) exists.
lockf(3) does not provide shared locking,
and requires that the filehandle be open for writing (or appending, or read/writing). - Some versions of
flock() can't lock files
over a network (e.g. on NFS file systems), so you'd
need to force the use of fcntl(2) when you build Perl. See the flock entry of
the perlfunc manpage,
and the INSTALL file in the source distribution for information on building Perl
to do this.
The CPAN module
File::Lock offers similar functionality and (if you have dynamic loading) won't require
you to rebuild perl if your flock() can't lock network files.
A
common bit of code NOT TO USE is this:
sleep(3) while -e "file.lock"; # PLEASE DO NOT USE
open(LCK, "> file.lock"); # THIS BROKEN CODE
This is a classic race condition: you take two steps
to do something which must be done in one. That's why computer hardware provides an atomic
test-and-set instruction. In theory, this ``ought'' to work:
sysopen(FH, "file.lock", O_WRONLY|O_EXCL|O_CREAT, 0644)
or die "can't open file.lock: $!":
except that lamentably, file creation (and deletion)
is not atomic over NFS, so this won't work (at least,
not every time) over the net. Various schemes involving involving link() have
been suggested, but these tend to involve busy-wait, which is also subdesirable.
Didn't anyone ever tell you web-page hit counters
were useless?
Anyway, this is what to do:
use Fcntl;
sysopen(FH, "numfile", O_RDWR|O_CREAT, 0644) or die "can't open numfile: $!";
flock(FH, 2) or die "can't flock numfile: $!";
$num = <FH> || 0;
seek(FH, 0, 0) or die "can't rewind numfile: $!";
truncate(FH, 0) or die "can't truncate numfile: $!";
(print FH $num+1, "\n") or die "can't write numfile: $!";
# DO NOT UNLOCK THIS UNTIL YOU CLOSE
close FH or die "can't close numfile: $!";
Here's a much better web-page hit counter:
$hits = int( (time() - 850_000_000) / rand(1_000) );
If the count doesn't impress your friends, then the
code might. :-)
If you're just trying to patch a binary, in many
cases something as simple as this works:
perl -i -pe 's{window manager}{window mangler}g' /usr/bin/emacs
However, if you have fixed sized records, then you
might do something more like this:
$RECSIZE = 220; # size of record, in bytes
$recno = 37; # which record to update
open(FH, "+<somewhere") || die "can't update somewhere: $!";
seek(FH, $recno * $RECSIZE, 0);
read(FH, $record, $RECSIZE) == $RECSIZE || die "can't read record $recno: $!";
# munge the record
seek(FH, $recno * $RECSIZE, 0);
print FH $record;
close FH;
Locking and error checking are left as an exercise
for the reader. Don't forget them, or you'll be quite sorry.
Don't forget to set binmode() under
DOS-like platforms when operating on files that have anything other than straight text in
them. See the docs on open() and on binmode() for more details.
If you want to retrieve the time at which the file
was last read, written, or had its meta-data (owner, etc) changed, you use the -M,
-A, or -C filetest operations as documented in the perlfunc manpage.
These retrieve the age of the file (measured against the start-time of your program) in
days as a floating point number. To retrieve the ``raw'' time in seconds since the epoch,
you would call the stat function, then use localtime(), gmtime(),
or POSIX::strftime() to convert this into human-readable form.
Here's an example:
$write_secs = (stat($file))[9];
print "file $file updated at ", scalar(localtime($file)), "\n";
If you prefer something more legible, use the
File::stat module (part of the standard distribution in version 5.004 and later):
use File::stat;
use Time::localtime;
$date_string = ctime(stat($file)->mtime);
print "file $file updated at $date_string\n";
Error checking is left as an exercise for the reader.
You use the utime() function documented
in utime. By
way of example, here's a little program that copies the read and write times from its
first argument to all the rest of them.
if (@ARGV < 2) {
die "usage: cptimes timestamp_file other_files ...\n";
}
$timestamp = shift;
($atime, $mtime) = (stat($timestamp))[8,9];
utime $atime, $mtime, @ARGV;
Error checking is left as an exercise for the reader.
Note that utime() currently doesn't work
correctly with Win95/NT ports. A bug has been
reported. Check it carefully before using it on those platforms.
If you only have to do this once, you can do this:
for $fh (FH1, FH2, FH3) { print $fh "whatever\n" }
To connect up to one filehandle to several output
filehandles, it's easiest to use the tee(1) program if you have it, and let
it take care of the multiplexing:
open (FH, "| tee file1 file2 file3");
Otherwise you'll have to write your own multiplexing
print function -- or your own tee program -- or use Tom Christiansen's, at http://www.perl.com/CPAN/authors/id/TOMC/scripts/tct.gz
, which is written in Perl.
In theory a IO::Tee class could be written, but to
date we haven't seen such.
Use the $\ variable (see the perlvar manpage
for details). You can either set it to "" to eliminate empty
paragraphs ("abc\n\n\n\ndef", for instance, gets treated as two
paragraphs and not three), or "\n\n" to accept empty paragraphs.
You can use the builtin getc() function
for most filehandles, but it won't (easily) work on a terminal device. For STDIN, either use the Term::ReadKey module from CPAN, or use the sample code in getc.
If your system supports POSIX,
you can use the following code, which you'll note turns off echo processing as well.
#!/usr/bin/perl -w
use strict;
$| = 1;
for (1..4) {
my $got;
print "gimme: ";
$got = getone();
print "--> $got\n";
}
exit;
BEGIN {
use POSIX qw(:termios_h);
my ($term, $oterm, $echo, $noecho, $fd_stdin);
$fd_stdin = fileno(STDIN);
$term = POSIX::Termios->new();
$term->getattr($fd_stdin);
$oterm = $term->getlflag();
$echo = ECHO | ECHOK | ICANON;
$noecho = $oterm & ~$echo;
sub cbreak {
$term->setlflag($noecho);
$term->setcc(VTIME, 1);
$term->setattr($fd_stdin, TCSANOW);
}
sub cooked {
$term->setlflag($oterm);
$term->setcc(VTIME, 0);
$term->setattr($fd_stdin, TCSANOW);
}
sub getone {
my $key = '';
cbreak();
sysread(STDIN, $key, 1);
cooked();
return $key;
}
}
END { cooked() }
The Term::ReadKey module from CPAN
may be easier to use:
use Term::ReadKey;
open(TTY, "</dev/tty");
print "Gimme a char: ";
ReadMode "raw";
$key = ReadKey 0, *TTY;
ReadMode "normal";
printf "\nYou said %s, char number %03d\n",
$key, ord $key;
For DOS systems,
Dan Carson <dbc@tc.fluke.COM> reports the
following:
To put the PC in
``raw'' mode, use ioctl with some magic numbers gleaned from msdos.c (Perl source file)
and Ralf Brown's interrupt list (comes across the net every so often):
$old_ioctl = ioctl(STDIN,0,0); # Gets device info
$old_ioctl &= 0xff;
ioctl(STDIN,1,$old_ioctl | 32); # Writes it back, setting bit 5
Then to read a single character:
sysread(STDIN,$c,1); # Read a single character
And to put the PC
back to ``cooked'' mode:
ioctl(STDIN,1,$old_ioctl); # Sets it back to cooked mode.
So now you have $c. If ord($c) == 0, you
have a two byte code, which means you hit a special key. Read another byte with sysread(STDIN,$c,1),
and that value tells you what combination it was according to this table:
# PC 2-byte keycodes = ^@ + the following:
# HEX KEYS
# --- ----
# 0F SHF TAB
# 10-19 ALT QWERTYUIOP
# 1E-26 ALT ASDFGHJKL
# 2C-32 ALT ZXCVBNM
# 3B-44 F1-F10
# 47-49 HOME,UP,PgUp
# 4B LEFT
# 4D RIGHT
# 4F-53 END,DOWN,PgDn,Ins,Del
# 54-5D SHF F1-F10
# 5E-67 CTR F1-F10
# 68-71 ALT F1-F10
# 73-77 CTR LEFT,RIGHT,END,PgDn,HOME
# 78-83 ALT 1234567890-=
# 84 CTR PgUp
This is all trial and error I
did a long time ago, I hope I'm reading the file that
worked.
You should check out the Frequently Asked Questions
list in comp.unix.* for things like this: the answer is essentially the same. It's very
system dependent. Here's one solution that works on BSD
systems:
sub key_ready {
my($rin, $nfd);
vec($rin, fileno(STDIN), 1) = 1;
return $nfd = select($rin,undef,undef,0);
}
You should look into getting the Term::ReadKey
extension from CPAN.
You need to use the O_NDELAY
or O_NONBLOCK flag from the Fcntl module in
conjunction with sysopen():
use Fcntl;
sysopen(FH, "/tmp/somefile", O_WRONLY|O_NDELAY|O_CREAT, 0644)
or die "can't open /tmp/somefile: $!":
You need to use the O_CREAT
and O_EXCL flags from the Fcntl module in conjunction
with sysopen():
use Fcntl;
sysopen(FH, "/tmp/somefile", O_WRONLY|O_EXCL|O_CREAT, 0644)
or die "can't open /tmp/somefile: $!":
Be warned that neither creation nor deletion of files
is guaranteed to be an atomic operation over NFS.
That is, two processes might both successful create or unlink the same file!
First try
seek(GWFILE, 0, 1);
The statement seek(GWFILE, 0, 1) doesn't
change the current position, but it does clear the end-of-file condition on the handle, so
that the next <GWFILE> makes Perl try again to
read something.
If that doesn't work (it relies on features of your
stdio implementation), then you need something more like this:
for (;;) {
for ($curpos = tell(GWFILE); <GWFILE>; $curpos = tell(GWFILE)) {
# search for some stuff and put it into files
}
# sleep for a while
seek(GWFILE, $curpos, 0); # seek to where we had been
}
If this still doesn't work, look into the POSIX module. POSIX defines
the clearerr() method, which can remove the end of file condition on a
filehandle. The method: read until end of file, clearerr(), read some more.
Lather, rinse, repeat.
If you check open, you'll
see that several of the ways to call open() should do the trick. For example:
open(LOG, ">>/tmp/logfile");
open(STDERR, ">&LOG");
Or even with a literal numeric descriptor:
$fd = $ENV{MHCONTEXTFD};
open(MHCONTEXT, "<&=$fd"); # like fdopen(3S)
Error checking has been left as an exercise for the
reader.
This should rarely be necessary, as the Perl close()
function is to be used for things that Perl opened itself, even if it was a dup of a
numeric descriptor, as with MHCONTEXT above. But if
you really have to, you may be able to do this:
require 'sys/syscall.ph';
$rc = syscall(&SYS_close, $fd + 0); # must force numeric
die "can't sysclose $fd: $!" unless $rc == -1;
Whoops! You just put a tab and a formfeed into that
filename! Remember that within double quoted strings (``like\this''), the backslash is an
escape character. The full list of these is in Quote
and Quote-like Operators. Unsurprisingly, you don't have a file called
``c:(tab)emp(formfeed)oo'' or ``c:(tab)emp(formfeed)oo.exe'' on your DOS filesystem.
Either single-quote your strings, or (preferably) use
forward slashes. Since all DOS and Windows versions
since something like MS-DOS 2.0 or so have treated /
and \ the same in a path, you might as well use the one that doesn't clash
with Perl -- or the POSIX shell, ANSI C and C++, awk, Tcl, Java, or Python, just to mention a few.
Because even on non-Unix ports, Perl's glob function
follows standard Unix globbing semantics. You'll need glob("*") to
get all (non-hidden) files.
This is elaborately and painstakingly described in
the ``Far More Than You Every Wanted To Know'' in http://www.perl.com/CPAN/doc/FMTEYEWTK/file-dir-perms
.
The executive summary: learn how your filesystem
works. The permissions on a file say what can happen to the data in that file. The
permissions on a directory say what can happen to the list of files in that directory. If
you delete a file, you're removing its name from the directory (so the operation depends
on the permissions of the directory, not of the file). If you try to write to the file,
the permissions of the file govern whether you're allowed to.
Here's an algorithm from the Camel Book:
srand;
rand($.) < 1 && ($line = $_) while <>;
This has a significant advantage in space over
reading the whole file in.
|
