venerdì 14 aprile 2017

A baby Perl program from a backup in the attic!

…I was assigned one of my very first program to develop: a way to assure an user can run no more than a specific number of instances
of the same executable.


The real story was this: a company was using an Unix ncurses based ERP, and users were forced to connect via remote shell and
launch it automatically. The problem was that several users were opening more terminals (and therefore ERP clients) limiting
resources to other users.


I wanted to demonstrate my skills being able to implement the solution in a couple of hours, and instead of spending time
searching for a production-ready solution on the Internet, I decided to do it all on my own.


It was clear to me I was needing a kind of wrapper to control the execution of external programs. In fact, since users were
automatically switched from the remote shell into the ERP client, it did suffice to place the wrapper as the
user shell.


But how to implement it? At that time I knew Java, C++, C, Perl and some Bourne Shell. Making it in Java was too heavy, and
most notably required to install a JVM on the server (it was the time of the Red Hat 5 branch). C and C++ required too much time
to come with a working solution, and Perl was the most promising allowing me to do everything I was able to do with
other languages, overcoming the limitation of the shell (e.g., arithmetic).


But at that time I was not a fluent Perl developer.


A few days ago I found the above program out of an old backup, and so I spent five minutes to read it again. Not surprisingly, I found
it awful with respect to the code I write today, and moreover I see a lot of code uglyness due to baby steps. But hey, it was fun
to see how many things I've changed in my coding style, idioms, adoption of the language and of its operators, and most notably
about how I'm today toward compact code instead of long and blown code. At that time I was convinced that as long the code was, as much
it must be simpler to read, today I believe it is the opposite (with a very few exceptions I tend not to adopt until I
can remember very well).


In this article I will dissect a few parts of the above program to show you how I was writing code back in those days, and it was
a huge time ago (at least in computer timeline): 15+ years ago.


2 Baby Steps

Well, as you can imagine when I started doing a real job I was nothing more than a programmer-to-be, and my knowlegde
about Perl was really tiny. I mean, I read the Camel Book (of course!), and I was experimenting with Perl as much as I could,
but I was not an avid consumer of the Perl community (as well as other communities) and so I did not have a lot of chances
to learn from other people experience and code.


More than that, I just finished my university degree, and so my mind was literally full of theories about writing
correct and beautiful code, but that were just theories without any implementation! And last but not least, I was convinced
I could do everything by my own (that is, who needs modules?).


Today I recognize the following as baby steps in the Perl world.

2.1 Method Prototypes

Yes, I was using prototypes everywhere.


Why? I suspect they were comfortable to me, since I was used to languages where each method has a prototype (C, C++, Java), and
so I was trying to use a kind of common approach to every language, applying to Perl prototypes even when I was not needing
them at all!


Of course, by time and experience, I found that prototypes are usually not useful at all in my programs, and made refactoring
a lot harder.

2.2 Method instead of Operators

Perl allows operators to be used without braces, but that was something my eyes were not able to parse!
Again, coming from languages where methods and operators both need parentheses, I tried to bend Perl to my will
and use operators the very same way.

2.3 Untrusting the Thruth

Perl has the great ability to cast a variable depending on the context, but it was something too much complex for me.
So for instances, instead of testing a scalar against a not-true value, I was using the defined operator:


# if ( $scalar )
if ( defined $scalar ){ ... }

Worst: I was not using the ability of an array to be scalar-contextualized in tests:


#if ( ! @array )
if ( not defined @array || $#array <= 0 ){ ... }

2.4 Global Variables

my what?


I was not using scoped variables, and the reasons were:

  1. it was not clear to me the real difference between my and local, and I have to confess I thought local was what should be the
    name for scoped variables (maybe because of the way some shells declare local variables), but since local was usually a bad idea
    I decided not to use either;
  2. my scripts were quite small and self contained, so there was no risk at all about variable clash.

2.5 Backtick as a Rescue!

I knew pretty much well a kind of command line Unix tools (e.g., grep, awk, sed, find), and sometimes I needed
to do something in Perl without knowing how to do but with shell tools. So, how could I merge the both?


Easy pal: backtick the Unix tools and manage the result in Perl!

2.6 No Modules, please

I was not using modules, both by design and by fear.


With by design I mean that I was writing scripts for machine loosely connected to the Internet (just think I was working with server behind an ISDN router!),
and therefore even dowloading a module and install it could be quite a pain.


By fear means I was quite scared about importing other people's code. I was still learning about how to use correctly the CPAN, how to read
the documentation, and so on. And I did not want to install things in the wrong way on code that must run. After all, I was half-an-hour
away from implementing some kind of module functionality by my own (ah, the ego…).

2.7 Regexp Sometimes…

I tried to use regexp, but I was too less experienced. Therefore I usually did the substitution with friendly tools, like
split and alike.


I fear regexp nomore, and even if I'm not a master at all, I find interesting placing them in programs not only because
they allow me to write clear and compact code, but also because they still are some criptic stuff other developers
have hard time reading.


3 The Code

3.1 University Distortion

There are a couple of places where you can see evidence of the teachings at my university, most notably the
command line argument loop:


sub parseCommandLine()
{

    # The script must be at least two arguments.
    if ( $#ARGV < 1 or $#ARGV > 4 )
    {
        help();
        exit(0);
    }

    $POLICY_FILE = $DEFAULT_POLICY_FILE;

    foreach $argument (@ARGV)
    {
        if ( $argument =~  /(-program=)/ )
        {
            $argument =~ s/(-program=)//;
            $PROGRAM = $argument;
        }
        elsif ( $argument =~ /(-policy=)/ )
        {
            $argument =~ s/(-policy=)//;
            $POLICY_FILE = $argument;
        }
        elsif ( $argument =~ /(-username=)/ )
        {
            $argument =~ s/(-username=)//;
            $USERNAME = $argument;
        }
        elsif ( $argument =~ /(-message=)/ )
        {
            $argument =~ s/(-message=)//;
            $MESSAGE_FILE = $argument;
        }
    }

    # check main parameters
    if ( not defined $USERNAME or $USERNAME eq "")
    {
        warn("\nCannot find username !!\n\n");
        help();
        exit(1);
    }
    elsif ( not defined $PROGRAM or $PROGRAM eq "")
    {
        warn("\nCannot find the program name \n\n");
        help();
        exit(2);
    }
    elsif ( not defined $POLICY_FILE or $POLICY_FILE eq "")
    {
        warn("\nI need a policy file to run!\n\n");
        help();
        exit(3);
    }


}

As you can see I was hardwiring the arguments directly into the program, as well as I was not using any getopt-like module.
That was by culture: university never told me there was a getopt way of getting parameters!


Please note also that I was checking the argument numbers as well as exiting from the program with a different exit code for each branch.
The style reminds pretty much the C-style I was used to work with during my exams.

3.2 Ready? Go!

How can I execute another program from within the Perl wrapper?


Well, the simplest way that I was aware of is to call execute, in other words, to work as a low level C:


sub launch($)
{
    # take arguments
    my ($program) = @_;

    # execute the program
    sleep(2);
    exec($program);
}

3.3 How Many Instances can you Run?

How to define how many instances an user can run?


Here I decided to take inspiration from the sudo policy file, and so I invented my own policy file with a pretty
simple syntax:


username@executable=instances


where:

  • username was the Unix username;
  • executable was the name of the executable to run;
  • instances was the number of the maximum allowed instances of the executable for the specified username.

And to make it looking mor eprofessional, I allowed the policy file to include a star in any field
in order to mean anything or any instance.


Of course, the above was full crap. For instance, if you were able to create a link from the executable to another
with the right name you could change you allowance. But luckily, no user was able to do that, and to some
extent even the other sysadmins!


Having defined the policy file syntax, reading the file was as simple as:


sub getAllowedInstances($$$)
{
    # take params
    my ($policy_file, $username, $program) = @_;
    my $instances = -1;

    # try to open the policy file
    open (POLICY_FILE_HANDLER, "<".$policy_file) || die("\nCannot parse policy file <$policy_file> !\n");
    print "\nParsing configuration file $policy_file for program $program, username $username...";

    $instances = 0;           # by default do not allow a user to execute a program

    # get each line from the policy file, and then search for the username
    while ( $line = <POLICY_FILE_HANDLER>  )
    {

        print "\n\t Configuration line: $line ";


        # take only lines with the program name specified
        if( grep("$program", $line) ){

            @lineParts = split("@",$line);
            $configUsername = $lineParts[0];
            print "\ncontrollo se $username == $configUsername";

            if ( $username eq $configUsername )
            {
                # this is the right line
                # take the instances number
                @pieces = split("=" , $line);
                $instances = $pieces[1];
                # remove the '\n'
                chomp($instances);
                print "\n\t\t\tUser allowance: $instances";
                return $instances;
            }
            elsif ( $configUsername eq "*" )
            {
                # a valid entry for all users
                # take the instances number
                @pieces = split("=" , $line);
                $instances = $pieces[1];
                # remove the '\n'
                chomp($instances);
                print "\n\t\t\tGlobal allowance: $instances";
            }
        }
    }


    # no lines found, the user has no restrictions
    return $instances;
}

What an amazing piece of code, uh?


Instead of reading the policy file once (or seldom), I was reading it fully every time I needed to check an user; it was
a very non-optimzed code.
Moreover, I was reading and grepping it a line at time, a very keyboard effort.

3.4 How much are you Running?

How to get the number of runnable instances an user was running?


Piece of cake: let's do it in Unix, backtick in Perl and increase by one:



sub getRunningInstancesCount($$)
{
    # get parameters
    my ( $program, $username ) = @_;

    # get all processes for this program and the username
    @processes = `ps -u $username | grep $program | grep -v "grep"`;


    # get all lines count
    $running_instances = 0;
    if ( $#processes >= 0 )
    {
        $running_instances = $#processes + 1;
    }
    else
    {
        return 0;
    }

    return $running_instances;
}

Please note also the adoption of literals instead of variables: for example return 0; instead of return $running_instances where
the latter has been already initialized to zero.

3.5 All Pieces Together

Now, putting the places together made the main loop as simple as:


$RUNNING_INSTANCES = getRunningInstancesCount($BASE_NAME, $USERNAME);
print "\nYou're running $RUNNING_INSTANCES $BASE_NAME\n";

# if the user can, launch the program
if ( $RUNNING_INSTANCES < $ALLOWED_INSTANCES )
{
    print "\nExecuting $PROGRAM...";
    launch($PROGRAM);
    print "\nBye!\n";
}
else
{
    print "\nYou can't run no more instances of $PROGRAM";
    print "\nBye\n";
    displayMessage();
}


4 Lesson Learned

There is no lesson to learn here, but a lot of lessons learned in the between.


The only thing I could say is that you should never, never, throw away your code. Keep it, it is cheap, and someday you could
see how you progressed from a green programmer to the very guru developer you probably are today.

Dimensioni delle tabelle e dei dati dump di testo, qualche insignificante esperimento

Mi sono ritrovato per le mani una vecchia e obsoleta istanza PostgreSQL 8.4 piuttosto grossa, lo spazio disco del tablespace
risultava essere di circa 13 GB! Ho quindi preso spunto per fare una piccola indagine su cosa occupasse tanto spazio,
concentrandomi solo sulle relazioni (tabelle):


SELECT c.oid,nspname AS table_schema
      , relname AS TABLE_NAME
      , c.reltuples AS row_estimate
      , to_char( pg_total_relation_size(c.oid)::real/(1024*1024), '99G999D99' ) AS MB
FROM pg_class c LEFT JOIN pg_namespace n ON n.oid = c.relnamespace
WHERE
      relkind = 'r'
      AND
      nspname = 'public'
ORDER BY 4 DESC, 3 ASC;

Ebbene sono saltate subito all'occhio tre tabelle in particolari (nomi di fantasia):


  oid   | table_schema | table_name | row_estimate |     mb
---------+--------------+------------+--------------+------------
   63740 | public       | tab1      |  8.74153e+06 |   2.248,58
   66161 | public       | tab2      |   2.9728e+06 |   1.192,00
   65032 | public       | tab3      |  2.44735e+06 |   1.280,77

Come si nota queste tre tabelle superano di slancio ciascuna una occupazione di 1 GB, arrivando fino a 9 milioni di tuple circa!
Insomma, non una cosa eccezionale per PostgreSQL, ma sicuramente nemmeno una cosa di routine, e che comunque indica forse la necessità
di una riprogettazione o di un partitioning.
Comunque, le stesse tabelle su disco quando occuperebbero in formato testo?



% pg_dump -h localhost -U luca -t tab1 testdb > tab1.sql

Effettuando il dump, con molta pazienza, di quelle tre tabelle sopra indicate si ottiene che:


% ls -lh tab?.sql
-rw-r--r-- 1 luca luca 579M 2017-04-12 11:32 tab1.sql
-rw-r--r-- 1 luca luca 494M 2017-04-12 11:37 tab2.sql
-rw-r--r-- 1 luca luca 571M 2017-04-12 11:36 tab3.sql

e quindi lo spazio occupato all'interno di PostgreSQL risulta da 2 a 4 volte superiore allo spazio disco dei dati testuali.
Chiaramente questa non rappresenta una inefficienza di PostgreSQL, quanto una naturale esigenza del database di tenere i dati allineati,
avere le pagine dati (8kB) con spazio sufficiente per consentire aggiornamenti, ecc.


Se si effettua un vacuum full sulle tabelle di cui sopra si ottiene il seguente risultato:


> vacuum full verbose tab1;
...
0 index pages have been deleted, 0 are currently reusable.

ad indicare che il database era già "buono", e abbastanza compattato. Ovviamente i dati che PostgreSQL riporta sul numero di tuple e dimensione
dei dati sono rimaste invariate.

lunedì 10 aprile 2017

On learning thru Open Source Software

I read an interesting blog post on adopting open source at university as a way of teaching computer science, and I posted also a comment
on such article.
Here I would like to extend my point of view about the subject.
Having worked in an university for a few years, having done computer science at both sides of the desk, and having a quite good
experience in what became my day-to-day work, I believe I can express some opinion about.

Adopting open source as a teaching methodology should just be.
Period!
Instead of asking students to produce the same crappy software all over the time (a database layer, a GUI layer, a music store,
a calculator), put them on a real piece of software that someone in the world could use and adopt.
Why?
  1. motivation: students will be happy to work on some real stuff with the main author(s) thanking them for their help, time,
    and work;
  2. learn something useful: real programs do things in the real world, and things in the real world must work, and work fast, and work
    in accurate way. That is:
    • work on real data: nobody will ever notice your crappy homework do a full table scan each time you need to display one of your
      fake music titles out of ten, but a real program will freeze once you try to do a full table scan just to display out a detail
      out of a million records. This is something university cannot teach you, trust me!
    • deal with problems: university will teach you a lot of theory about using natural keys, algorithms to sort structures, avoid
      data replication, and so on. Not always these approaches will drive you to a manageable software: learn to deal
      with surrogate keys, duplicate data when it makes sense (e.g., network latency, historical reasons and so on).
    • learn the tools: developers around the world need to coordinate. Learn to use bug reports, stay on mailing lists, IRC channels,
      and alike. Don't ask stackoverflow for someone to do your homework, learn how to find documentation and search for answers. Become
      acquainted with revision control, compilers, linkers, and different platforms.
    • document for someone else: it is easy for you to explain what you have done to your teacher, in particular if you did it in the very
      previous period of time (typically a semester). But can you document something so that another developer, even another student like
      you, can understand one year later why and how you did a task?
  3. do not start a project from scratch: typically the university cycle during semesters is something like design-implement-compile-run-explain-throw away_
    and then you start over and over again every time you got an assignment, homework, or project. This is wrong! Real life does not work as such:
    in real life you are often assigned to maintain or refactor an existing piece of code without having to throw it away.
  4. learn idioms: developers around the globe are smarter than you. It is not they are more intelligent, it is just they are more expert
    and know better the subject. Reading someone else (smarter) code is a great way to improve your knowledge and become smarter too. Learning idioms,
    seeing patterns applied to real world code is a great start to become a professional coder.
  5. fun: developers have their habits. They meet together in so called conferences, and usually got beers while talking about code, travel
    around the world, and have a lot of fun. And even if you stay closed in your room, doing your homework, having a chat, a video call
    or an email in your in-box with a "thank you" or "good job!" is really fun.

There are other reasons, but the above are my main choices to push students to the open source world.
So far, it seems that you are going to have only advantages and no drawbacks, but that's not true.
Becoming an open source contributor you are going to become smarter than your own university teacher, and this is a drawback so far as
the teacher signs your curriculum. It is something that is really hard for a teacher to keep in mind, but it is so true.
I was always saying to my student, in the very beginning of each class, that at the end they will know better than me the subject, and the reason
for that is that "I'm not going to spend whole nights studying before the exam!".
So, if you are a teacher, simply accept that.
Accept that a student could prove a better algorithm, or an idiom you don't know that works. Ask him for the details, learn from him.
Learning is not a one-way process, with a god-like teacher and an idiot-like student; learning is a cooperation where someone expert provides the sparkle to someone else.
Would not be nice to see if both can grow during the path?

There is also another drawback: open source is not something you can teach without knowledge. You have to know the tools: revision control, IDEs,
bug tracking, issue tracking, wiki, testing and alike.
Most teachers urge teaching C pointers arhitmetic instead of basic tools, and that's not good.
Allow me to provide my very own example: I spent five years in a computer science degree, and nobody told me about revision control. While
doing my master thesis, being afraid of loosing some change or to mistakenly do a single line change (that will not blow up your project, right?),
I did my very own backup shell script that was running every hour to keep an historical copy of my work.
Shame on me, in the very same time I could have learnt rc or cvs (no, it was before git).

So my advice for students is be a part of an open source community, you will surely learn something that will make the difference in
your real job.
And my advice for teachers is accept smarter students and promote the adoption of an open source code base. There are a lot of "mini-hackers"
initiatives around the world (CPAN Pull Request, Hacktoberfest, etc.), pick one and try let your student do the rest.
You'll be happier, your student will be happier, the open source community will be happier and, who knows, your employer could also
become a partner in an open source community.

Considerazioni sul planet italiano di PostgreSQL

Ho fatto caso che ormai sul planet ufficiale dell'associazione ITPUG, ovvero www.planetpostgresql.it, sto scrivendo
sporadicamente solo io. Questo secondo me è un campanello di allarme: io non sono certo migliore o piu' bravo di altri
soci e componenti dell'associazione, ma questa assenza dell'associazione dal planet indica che forse non si crede piu'
in questo strumento. Il fatto però è che nemmeno sul planet ufficiale si leggono post di ITPUG, e quindi non è tanto
la piattaforma italiana ad essere trascurata, ma il sistema di pubblicazione di notizie e articoli in generale.


Tornando al planet italiano, è facile verificare che su un periodo abbastanza ampio gli unici post
aggregati sono miei, e spesso risultano a loro volta dei link ad altre notizie ed articoli:


  1. 6 Gennaio 2017
  2. 13 Gennaio 2017
  3. 18 Gennaio 2017
  4. 27 Gennaio 2017
  5. 3 Febbraio 2017
  6. 21 Febbraio 2017
  7. 23 Febbraio 2017
  8. 24 Marzo 2017 (a) e (b)
  9. 2 Aprile 2017
  10. 5 Aprile 2017 (a) e (b)
  11. 8 Aprile 2017

Possibile che in un periodo di circa 3 mesi ITPUG non abbia avuto nessuna notizia da pubblicare?
Perfino in questi giorni, dove si richiede la regolarizzazione della quota per l'anno 2017 in vista
dell'imminente assemblea ordinaria, non vi sono notizie a riguardo.


Il consiglio che mi sento di dare al futuro consiglio è quello di prendere una decisione in merito al planet: se non lo si
vuole aggiornare allora tanto vale "spegnerlo", ma se lo si mantiene (come è tradizione anche nell'ambiente PostgreSQL e non solo)
allora lo si deve popolare periodicamente.

domenica 9 aprile 2017

Rust, Gnome & friends

C'è un certo momentum attorno al linguaggio di programmazione Rust.
Premetto che non ho mai utilizzato Rust nemmeno per il classico "Hello World", tuttavia ho provato a leggere la documentazione
e devo essere sincero: mi pare un linguaggio abbastanza complesso e pieno di keyword e di trabocchetti che non sono propri
di un linguaggio che si propone come un C migliorato.

Ad ogni modo pare che il team Gnome voglia passare a Rust per l'implementazione delle librerie e della infrastruttura ad oggetti
del famoso desktop. O meglio, si vuole pluggare componenti Rust nel sistema ad oggetti GTK+. Conosco GTK+, anche se sono molti
anni che non lo "utilizzo" da vicino: l'idea è bella ed elegante e si basa sulla costruzione di un sistema OOP puramente in C.
E riesce bene nello scopo, a patto di ricordarsi alcune regole boilerplate per la definizione dei nuovi tipi (classi).
Evidentemente, proponendosi Rust come un super-C ad oggetti, l'idea del team Gnome è quella di usare Rust per ridurre il codice
boilerplate pur mantenendo la dorsale GTK+ (per ovvia retrocompatibilità).

Francamente non mi pare che Gnome negli ultimi anni si sia mosso nelle giuste direzioni, e questa di Rust mi sembra l'ennesima
"cantonata" dettata quasi piu' dalla moda che non da problematiche tecnologiche. GTK+ è un sistema solido, testato, affermato
e modificarlo adesso creando una sorta di super-binding per Rust rischia di confondere i programmatori e creare piu' incompatibilità
rispetto ai problemi risolti. Sono sicuro che Gnome ha valutato molto bene vantaggi e svantaggi di questo approccio, e la mia resta
una pura riflessione personale, ma onestamente trovo abbastanza complesso e poco elegante il linguaggio scelto.

Da ultimo, per una visione piu' simile alla mia, si legga questo interessante articolo sull'adozione di nuovi linguaggi (fra i quali Rust).

sabato 8 aprile 2017

PostgreSQL ltree

Da un thread su una delle mailing list ufficiali ho appreso di un tipo di dato a me sconosciuto: ltree.
Un ltree rappresenta un label tree, quello che nei linguaggi di programmazione è un meccanismo di properties. In sostanza
si definiscono delle etichette, ordinate gerarchicamente in base ad un separatore (il carattere .) e a queste si associa un valore.
Esistono poi funzioni apposite di utilità per la navigazione e la ricerca nell'albero delle etichette.


Vediamo un esempio pratico: supponiamo di voler catalogare in un albero alcune informazioni basilari riguardo la nostra associazioni
(dati assolutamente casuali e a puro scopo didattico!).


CREATE TABLE 
itpug( tipologia ltree, counter integer );

Ora si supponga di voler trovare il dettaglio dei soci, magari la loro somma partendo dalla foglia dell'albero:


SELECT sum( counter )  
FROM itpug  
WHERE tipologia @> 'itpug.soci';
 sum
----
  50
(1 row)

Supponiamo di voler trovare tutte le informazioni relativi al web (si noti l'uso della tilde):



SELECT *  
FROM itpug  
WHERE tipologia ~ '*.web.*';
         tipologia         | counter
---------------------------+---------
 itpug.web.siti            |       3
 itpug.web.ssl.certificati |       1
(2 rows)

Insomma, ltree si presenta come estensione sicuramente interessante, anche se forse un po' sorpassata dall'uso di altri formati quali json e jsonb.

venerdì 7 aprile 2017

Emacs yank at point (dal mouse)

Finalmente, dopo tanto tempo, ho trovato un trucchetto per far si che
Emacs incolli la clipboard tramite mouse dove c'è il cursore.
Mi spiego meglio: quando in Emacs si effettua un yank (incolla) dal mouse, mediante il pulsante centrale,
la parte copiata viene posizionata dove si trova il puntatore del mouse nel buffer, spostando il cursore in quel
punto (point). La cosa può risultare noiosa, perché spesso si vuole copiare un comando o un link da altre applicazioni
senza spostare il cursorse e senza dover posizionare esattamente il puntatore del mouse.
Niente di piu' semplice nell'editor piu' configurabile del mondo:


(setq mouse-yank-at-point t)

Valorizzando la variabile mouse-yank-at-point si ottiene il comportamento desiderato: il mouse incolla (yank)
dove si trova il cursore (at-point) senza spostare quest'ultimo.