perlfaq8.pod

MSYS在windows下模拟了一个类unix的终端

You'll have to catch the SIGCHLD signal, and possibly SIGPIPE too.
SIGCHLD is sent when the backgrounded process finishes.  SIGPIPE is
sent when you write to a filehandle whose child process has closed (an
untrapped SIGPIPE can cause your program to silently die).  This is
not an issue with C<system("cmd&")>.

=item Zombies

You have to be prepared to "reap" the child process when it finishes

    $SIG{CHLD} = sub { wait };

See L<perlipc/"Signals"> for other examples of code to do this.
Zombies are not an issue with C<system("prog &")>.

=back

=head2 How do I trap control characters/signals?

You don't actually "trap" a control character.  Instead, that character
generates a signal which is sent to your terminal's currently
foregrounded process group, which you then trap in your process.
Signals are documented in L<perlipc/"Signals"> and the
section on ``Signals'' in the Camel.

Be warned that very few C libraries are re-entrant.  Therefore, if you
attempt to print() in a handler that got invoked during another stdio
operation your internal structures will likely be in an
inconsistent state, and your program will dump core.  You can
sometimes avoid this by using syswrite() instead of print().

Unless you're exceedingly careful, the only safe things to do inside a
signal handler are (1) set a variable and (2) exit.  In the first case,
you should only set a variable in such a way that malloc() is not
called (eg, by setting a variable that already has a value).

For example:

    $Interrupted = 0;	# to ensure it has a value
    $SIG{INT} = sub {
        $Interrupted++;
	syswrite(STDERR, "ouch\n", 5);
    }

However, because syscalls restart by default, you'll find that if
you're in a "slow" call, such as <FH>, read(), connect(), or
wait(), that the only way to terminate them is by "longjumping" out;
that is, by raising an exception.  See the time-out handler for a
blocking flock() in L<perlipc/"Signals"> or the section on ``Signals''
in the Camel book.

=head2 How do I modify the shadow password file on a Unix system?

If perl was installed correctly and your shadow library was written
properly, the getpw*() functions described in L<perlfunc> should in
theory provide (read-only) access to entries in the shadow password
file.  To change the file, make a new shadow password file (the format
varies from system to system--see L<passwd(5)> for specifics) and use
pwd_mkdb(8) to install it (see L<pwd_mkdb(8)> for more details).

=head2 How do I set the time and date?

Assuming you're running under sufficient permissions, you should be
able to set the system-wide date and time by running the date(1)
program.  (There is no way to set the time and date on a per-process
basis.)  This mechanism will work for Unix, MS-DOS, Windows, and NT;
the VMS equivalent is C<set time>.

However, if all you want to do is change your timezone, you can
probably get away with setting an environment variable:

    $ENV{TZ} = "MST7MDT";		   # unixish
    $ENV{'SYS$TIMEZONE_DIFFERENTIAL'}="-5" # vms
    system "trn comp.lang.perl.misc";

=head2 How can I sleep() or alarm() for under a second?

If you want finer granularity than the 1 second that the sleep()
function provides, the easiest way is to use the select() function as
documented in L<perlfunc/"select">.  Try the Time::HiRes and
the BSD::Itimer modules (available from CPAN).

=head2 How can I measure time under a second?

In general, you may not be able to.  The Time::HiRes module (available
from CPAN) provides this functionality for some systems.

If your system supports both the syscall() function in Perl as well as
a system call like gettimeofday(2), then you may be able to do
something like this:

    require 'sys/syscall.ph';

    $TIMEVAL_T = "LL";

    $done = $start = pack($TIMEVAL_T, ());

    syscall(&SYS_gettimeofday, $start, 0) != -1
               or die "gettimeofday: $!";

       ##########################
       # DO YOUR OPERATION HERE #
       ##########################

    syscall( &SYS_gettimeofday, $done, 0) != -1
           or die "gettimeofday: $!";

    @start = unpack($TIMEVAL_T, $start);
    @done  = unpack($TIMEVAL_T, $done);

    # fix microseconds
    for ($done[1], $start[1]) { $_ /= 1_000_000 }

    $delta_time = sprintf "%.4f", ($done[0]  + $done[1]  )
                                            -
                                 ($start[0] + $start[1] );

=head2 How can I do an atexit() or setjmp()/longjmp()? (Exception handling)

Release 5 of Perl added the END block, which can be used to simulate
atexit().  Each package's END block is called when the program or
thread ends (see L<perlmod> manpage for more details).  

For example, you can use this to make sure your filter program
managed to finish its output without filling up the disk:

    END {
	close(STDOUT) || die "stdout close failed: $!";
    } 

The END block isn't called when untrapped signals kill the program,
though, so if you use END blocks you should also use

	use sigtrap qw(die normal-signals);

Perl's exception-handling mechanism is its eval() operator.  You can
use eval() as setjmp and die() as longjmp.  For details of this, see
the section on signals, especially the time-out handler for a blocking
flock() in L<perlipc/"Signals"> or the section on ``Signals'' in
the Camel Book.

If exception handling is all you're interested in, try the
exceptions.pl library (part of the standard perl distribution).

If you want the atexit() syntax (and an rmexit() as well), try the
AtExit module available from CPAN.

=head2 Why doesn't my sockets program work under System V (Solaris)?  What does the error message "Protocol not supported" mean?

Some Sys-V based systems, notably Solaris 2.X, redefined some of the
standard socket constants.  Since these were constant across all
architectures, they were often hardwired into perl code.  The proper
way to deal with this is to "use Socket" to get the correct values.

Note that even though SunOS and Solaris are binary compatible, these
values are different.  Go figure.

=head2 How can I call my system's unique C functions from Perl?

In most cases, you write an external module to do it--see the answer
to "Where can I learn about linking C with Perl? [h2xs, xsubpp]".
However, if the function is a system call, and your system supports
syscall(), you can use the syscall function (documented in
L<perlfunc>).

Remember to check the modules that came with your distribution, and
CPAN as well--someone may already have written a module to do it.

=head2 Where do I get the include files to do ioctl() or syscall()?

Historically, these would be generated by the h2ph tool, part of the
standard perl distribution.  This program converts cpp(1) directives
in C header files to files containing subroutine definitions, like
&SYS_getitimer, which you can use as arguments to your functions.
It doesn't work perfectly, but it usually gets most of the job done.
Simple files like F<errno.h>, F<syscall.h>, and F<socket.h> were fine,
but the hard ones like F<ioctl.h> nearly always need to hand-edited.
Here's how to install the *.ph files:

    1.  become super-user
    2.  cd /usr/include
    3.  h2ph *.h */*.h

If your system supports dynamic loading, for reasons of portability and
sanity you probably ought to use h2xs (also part of the standard perl
distribution).  This tool converts C header files to Perl extensions.
See L<perlxstut> for how to get started with h2xs.

If your system doesn't support dynamic loading, you still probably
ought to use h2xs.  See L<perlxstut> and L<ExtUtils::MakeMaker> for
more information (in brief, just use B<make perl> instead of a plain
B<make> to rebuild perl with a new static extension).

=head2 Why do setuid perl scripts complain about kernel problems?

Some operating systems have bugs in the kernel that make setuid
scripts inherently insecure.  Perl gives you a number of options
(described in L<perlsec>) to work around such systems.

=head2 How can I open a pipe both to and from a command?

The IPC::Open2 module (part of the standard perl distribution) is an
easy-to-use approach that internally uses pipe(), fork(), and exec() to do
the job.  Make sure you read the deadlock warnings in its documentation,
though (see L<IPC::Open2>).  See 
L<perlipc/"Bidirectional Communication with Another Process"> and 
L<perlipc/"Bidirectional Communication with Yourself">

You may also use the IPC::Open3 module (part of the standard perl
distribution), but be warned that it has a different order of
arguments from IPC::Open2 (see L<IPC::Open3>).

=head2 Why can't I get the output of a command with system()?

You're confusing the purpose of system() and backticks (``).  system()
runs a command and returns exit status information (as a 16 bit value:
the low 7 bits are the signal the process died from, if any, and
the high 8 bits are the actual exit value).  Backticks (``) run a
command and return what it sent to STDOUT.

    $exit_status   = system("mail-users");
    $output_string = `ls`;

=head2 How can I capture STDERR from an external command?

There are three basic ways of running external commands:

    system $cmd;		# using system()
    $output = `$cmd`;		# using backticks (``)
    open (PIPE, "cmd |");	# using open()

With system(), both STDOUT and STDERR will go the same place as the
script's STDOUT and STDERR, unless the system() command redirects them.
Backticks and open() read B<only> the STDOUT of your command.

With any of these, you can change file descriptors before the call:

    open(STDOUT, ">logfile");
    system("ls");

or you can use Bourne shell file-descriptor redirection:

    $output = `$cmd 2>some_file`;
    open (PIPE, "cmd 2>some_file |");

You can also use file-descriptor redirection to make STDERR a
duplicate of STDOUT:

    $output = `$cmd 2>&1`;
    open (PIPE, "cmd 2>&1 |");

Note that you I<cannot> simply open STDERR to be a dup of STDOUT
in your Perl program and avoid calling the shell to do the redirection.
This doesn't work:

    open(STDERR, ">&STDOUT");
    $alloutput = `cmd args`;  # stderr still escapes

This fails because the open() makes STDERR go to where STDOUT was
going at the time of the open().  The backticks then make STDOUT go to
a string, but don't change STDERR (which still goes to the old
STDOUT).

Note that you I<must> use Bourne shell (sh(1)) redirection syntax in
backticks, not csh(1)!  Details on why Perl's system() and backtick
and pipe opens all use the Bourne shell are in
http://www.perl.com/CPAN/doc/FMTEYEWTK/versus/csh.whynot .
To capture a command's STDERR and STDOUT together:

    $output = `cmd 2>&1`;                       # either with backticks
    $pid = open(PH, "cmd 2>&1 |");              # or with an open pipe
    while (<PH>) { }                            #    plus a read

To capture a command's STDOUT but discard its STDERR:

    $output = `cmd 2>/dev/null`;                # either with backticks
    $pid = open(PH, "cmd 2>/dev/null |");       # or with an open pipe
    while (<PH>) { }                            #    plus a read

To capture a command's STDERR but discard its STDOUT:

    $output = `cmd 2>&1 1>/dev/null`;           # either with backticks
    $pid = open(PH, "cmd 2>&1 1>/dev/null |");  # or with an open pipe
    while (<PH>) { }                            #    plus a read

To exchange a command's STDOUT and STDERR in order to capture the STDERR
but leave its STDOUT to come out our old STDERR:

    $output = `cmd 3>&1 1>&2 2>&3 3>&-`;        # either with backticks
    $pid = open(PH, "cmd 3>&1 1>&2 2>&3 3>&-|");# or with an open pipe
    while (<PH>) { }                            #    plus a read

To read both a command's STDOUT and its STDERR separately, it's easiest
and safest to redirect them separately to files, and then read from those
files when the program is done:

    system("program args 1>/tmp/program.stdout 2>/tmp/program.stderr");

Ordering is important in all these examples.  That's because the shell
processes file descriptor redirections in strictly left to right order.

    system("prog args 1>tmpfile 2>&1");
    system("prog args 2>&1 1>tmpfile");

The first command sends both standard out and standard error to the
temporary file.  The second command sends only the old standard output
there, and the old standard error shows up on the old standard out.

=head2 Why doesn't open() return an error when a pipe open fails?

Because the pipe open takes place in two steps: first Perl calls
fork() to start a new process, then this new process calls exec() to
run the program you really wanted to open.  The first step reports
success or failure to your process, so open() can only tell you
whether the fork() succeeded or not.

To find out if the exec() step succeeded, you have to catch SIGCHLD
and wait() to get the exit status.  You should also catch SIGPIPE if
you're writing to the child--you may not have found out the exec()
failed by the time you write.  This is documented in L<perlipc>.

In some cases, even this won't work.  If the second argument to a
piped open() contains shell metacharacters, perl fork()s, then exec()s
a shell to decode the metacharacters and eventually run the desired
program.  Now when you call wait(), you only learn whether or not the
I<shell> could be successfully started...it's best to avoid shell
metacharacters.

On systems that follow the spawn() paradigm, open() I<might> do what
you expect--unless perl uses a shell to start your command. In this
case the fork()/exec() description still applies.

=head2 What's wrong with using backticks in a void context?

Strictly speaking, nothing.  Stylistically speaking, it's not a good
way to write maintainable code because backticks have a (potentially
humongous) return value, and you're ignoring it.  It's may also not be very
efficient, because you have to read in all the lines of output, allocate
memory for them, and then throw it away.  Too often people are lulled
to writing:

    `cp file file.bak`;

And now they think "Hey, I'll just always use backticks to run programs."
Bad idea: backticks are for capturing a program's output; the system()
function is for running programs.

Consider this line:

    `cat /etc/termcap`;

You haven't assigned the output anywhere, so it just wastes memory
(for a little while).  You forgot to check C<$?> to see whether
the program even ran correctly, too.  Even if you wrote

    print `cat /etc/termcap`;

this code could and probably should be written as

    system("cat /etc/termcap") == 0
	or die "cat program failed!";

which will get the output quickly (as it is generated, instead of only
at the end) and also check the return value.

system() also provides direct control over whether shell wildcard
processing may take place, whereas backticks do not.