sem.pod: toilet analogy added to explain semaphore.

src/parallel

@@ -1061,7 +1061,7 @@ sub parse_options {
 
 sub init_globals {
     # Defaults:
-    $Global::version = 20150503;
+    $Global::version = 20150509;
     $Global::progname = 'parallel';
     $Global::infinity = 2**31;
     $Global::debug = 0;
@@ -1243,8 +1243,12 @@ sub parse_semaphore {
     if(defined $opt::bg) { $Global::semaphore = 1; }
     if(defined $opt::wait) { $Global::semaphore = 1; }
     if($Global::semaphore) {
-        # A semaphore does not take input from neither stdin nor file
-        @opt::a = ("/dev/null");
+	if(@opt::a) {
+	    # A semaphore does not take input from neither stdin nor file
+	    ::error("A semaphore does not take input from neither stdin nor a file\n");
+	    ::wait_and_exit(255);
+	}
+	@opt::a = ("/dev/null");
         push(@Global::unget_argv, [Arg->new("")]);
         $Semaphore::timeout = $opt::semaphoretimeout || 0;
         if(defined $opt::semaphorename) {

src/parallel.pod

@@ -861,7 +861,7 @@ Number of jobslots on each machine. Run up to N jobs in parallel.  0
 means as many as possible. Default is 100% which will run one job per
 CPU core on each machine.
 
-If B<--semaphore> is set default is 1 thus making a mutex.
+If B<--semaphore> is set, the default is 1 thus making a mutex.
 
 
 =item B<--jobs> I<+N>

src/sem.pod

@@ -12,11 +12,14 @@ B<sem> [--fg] [--id <id>] [--semaphoretimeout <secs>] [-j <num>] [--wait] comman
 
 GNU B<sem> is an alias for GNU B<parallel --semaphore>.
 
-It works as a tool for executing shell commands in parallel. GNU
-B<sem> acts as a counting semaphore. When GNU B<sem> is called with
-command it will start the command in the background. When I<num>
-number of commands are running in the background, GNU B<sem> will wait
-for one of these to complete before starting another command.
+GNU B<sem> acts as a counting semaphore. When GNU B<sem> is called
+with command it starts the command in the background. When I<num>
+number of commands are running in the background, GNU B<sem> waits for
+one of these to complete before starting the command.
+
+GNU B<sem> does not read any arguments to build the command (no -a,
+:::, and ::::). It simply waits for a semaphore to become available
+and then runs the command given.
 
 Before looking at the options you may want to check out the examples
 after the list of options. That will give you an idea of what GNU
@@ -28,23 +31,21 @@ B<sem> is capable of.
 
 =item I<command>
 
-Command to execute. The command may be followed by arguments for the command.
+Command to execute. The command may be followed by arguments for the
+command.
 
 
 =item B<--bg>
 
-Run command in background thus GNU B<parallel> will not wait for
+Run command in background thus GNU B<sem> will not wait for
 completion of the command before exiting. This is the default.
 
+In toilet analogy: GNU B<sem> waits for a toilet to be available,
+gives the toilet to a person, and exits immediately.
+
 See also: B<--fg>
 
 
-=item B<-j> I<N>
-
-Run up to N commands in parallel. Default is 1 thus acting like a
-mutex.
-
-
 =item B<--jobs> I<N>
 
 =item B<-j> I<N>
@@ -56,6 +57,8 @@ mutex.
 Run up to N commands in parallel. Default is 1 thus acting like a
 mutex.
 
+In toilet analogy: B<-j> is the number of toilets.
+
 
 =item B<--jobs> I<+N>
 
@@ -122,29 +125,70 @@ are often a good value.
 
 The semaphore is stored in ~/.parallel/semaphores/
 
+In toilet analogy the name corresponds to different types of toilets:
+e.g. male, female, customer, staff.
+
 
 =item B<--fg>
 
 Do not put command in background.
 
+In toilet analogy: GNU B<sem> waits for a toilet to be available,
+takes a person to the toilet, waits for the person to finish, and
+exits.
+
 
 =item B<--semaphoretimeout> I<secs> (alpha testing)
 
 =item B<--st> I<secs> (alpha testing)
 
-If I<secs> > 0: If the semaphore is not released within I<secs> seconds, take it anyway.
+If I<secs> > 0: If the semaphore is not released within I<secs>
+seconds, take it anyway.
 
-If I<secs> < 0: If the semaphore is not released within I<secs> seconds, exit.
+If I<secs> < 0: If the semaphore is not released within I<secs>
+seconds, exit.
+
+In toilet analogy: I<secs> > 0: If no toilet becomes available within
+I<secs> seconds, pee on the floor. I<secs> < 0: If no toilet becomes
+available within I<secs> seconds, exit without doing anything.
 
 
 =item B<--wait>
 
-=item B<-w>
-
 Wait for all commands to complete.
 
+In toilet analogy: Wait until all toilets are empty, then exit.
+
+
 =back
 
+=head1 UNDERSTANDING A SEMAPHORE
+
+Try the following example:
+
+  sem -j 2 'sleep 1;echo 1 finished';   echo sem 1 exited
+  sem -j 2 'sleep 2;echo 2 finished';   echo sem 2 exited
+  sem -j 2 'sleep 3;echo 3 finished';   echo sem 3 exited
+  sem -j 2 'sleep 4;echo 4 finished';   echo sem 4 exited
+  sem --wait; echo sem --wait done
+
+In toilet analogy this uses 2 toilets (B<-j 2>). GNU B<sem> takes '1'
+to a toilet, and exits immediately. While '1' is sleeping, another GNU
+B<sem> takes '2' to a toilet, and exits immediately.
+
+While '1' and '2' are sleeping, another GNU B<sem> waits for a free
+toilet. When '1' finishes, a toilet becomes available, and this GNU
+B<sem> stops waiting, and takes '3' to a toilet, and exits
+immediately.
+
+While '2' and '3' are sleeping, another GNU B<sem> waits for a free
+toilet.  When '2' finishes, a toilet becomes available, and this GNU
+B<sem> stops waiting, and takes '4' to a toilet, and exits
+immediately.
+
+Finally another GNU B<sem> waits for all toilets to become free.
+
+
 =head1 EXAMPLE: Gzipping *.log
 
 Run one gzip process per CPU core. Block until a CPU core becomes
@@ -165,15 +209,28 @@ you run multiple pod2html in parallel (e.g. in a Makefile with make
 write to the files at the same time:
 
   # This may fail due to shared pod2htmd.tmp/pod2htmi.tmp files
-  pod2html foo.pod --outfile foo.html & pod2html bar.pod --outfile bar.html
+  foo.html:
+          pod2html foo.pod --outfile foo.html
+
+  bar.html:
+          pod2html bar.pod --outfile bar.html
+
+  $ make -j foo.html bar.html
 
 You need to protect pod2html from running twice at the same time.
 B<sem> running as a mutex will make sure only one runs:
 
-  sem --id pod2html pod2html foo.pod --outfile foo.html
-  sem --id pod2html pod2html bar.pod --outfile bar.html
-  sem --fg --id pod2html rm -f pod2htmd.tmp pod2htmi.tmp
+  foo.html:
+          sem --id pod2html pod2html foo.pod --outfile foo.html
 
+  bar.html:
+          sem --id pod2html pod2html bar.pod --outfile bar.html
+
+  clean: foo.html bar.html
+          sem --id pod2html --wait
+          rm -f pod2htmd.tmp pod2htmi.tmp
+
+  $ make -j foo.html bar.html clean
 
 =head1 BUGS