Andrea "6502" Griffini
Hi, my name is Andrea Griffini.
Picture of me I have been born in 1966 in Vigevano and I have always been living here except for a few nice years during high school when I was in Intra (on Lago maggiore) and for about one year when I was living and working in California.
I also have a few social accounts/profiles, but I'm not very active on any of them.
If you want to contact me for any reason feel free to send a mail to x@y.z.
In case you are a recruiter I'm not offended at all (why should someone be?) but please consider that I am not currently looking for a new job.
Except for spam or spam-looking messages I normally reply to personal emails, so if I don't please consider that your email could have been lost somewhere. It may also take some time for me to reply (even days).
I am a computer programmer.
I have been programming since 12, starting with programmable calculators (a TI-57 and an HP that my brother got as a bargain because the seller thought it was broken when it was instead RPN based).
The first "real computer" I worked on was an Apple ][, 8 bit, 16Kb RAM and 1MHz clock (the computer I'm typing this text now is 64 bit, 16Gb RAM and 3GHz clock x 4 CPUs, in other words is roughly 8×1,000,000×3,000×4=96,000,000,000 i.e. ninety-six billions times better, not counting the storage - magnetic tape cassette instead of a SSD).
I program in many computer languages.
Actually programming languages is one of my interests and I have experimented quite a bit in many of them, including exotic ones, and I also have created mines.
The languages that I currently use most often are
    ## Python
    A very nice language. The first Python code I saw was in an usenet post on it.comp.lang.c++ and I thought it was pseudo-code.
    It is so nice that this can even be a problem as I got the impression that sometimes the temptation is to write overly complex solutions because the basic solution that one would implement in C++ is too easy and a programmer's brain is always looking for a challenge.
    The other problem is speed compared to C++, but this can be mitigated easily by writing C++ modules interfaced using sip, writing cython modules or even more simply by using PyPy if the dependencies allow it.
    Python is my first choice in many cases and my rule is roughly "Python if you can, C++ if you need".
    ## C++
    A powerful but complex language. It has many defects and problems, but still it can get the job done without compromising too much (for the end user).
    The biggest problem in my opinion is the "undefined behavior" concept that requires programmers to never do mistakes. Paired with the absurd complexity of the language this makes most complex C++ software a ticking bomb just waiting to explode.
    C++ is not and will never be a high-level language than can let you forget about its problems and concentrate on the problem you are working on. Sometimes it gives you that illusion, but at the very first segfault you are reminded that you're just a few millimeters above the metal.
    One quite annoying issue of C++ is that it's hard to experiment with
    I think that the best way to really understand something in programming is by writing, not by reading. Reading a bit is very important and all good, but reading too much is not, because reading only gives you the illusion you understood something.
    Unfortunately, C++ language is in my opinion quite hostile to experimentation for three main reasons
    • It's complex
      C++ is actually terribly complex. Some of the complexity is essential (the problem being considered is hard) but sometimes it's artificial (it's just more complex than it could have been).
    • It's often illogical
      The C++ semantic (and even its syntax) can be really surprising and appears to be in a few places completely illogical. There are in my opinion mainly two different reasons for this 1) it's the output of a committee and 2) it's an old language that changed over time.
      The first reason can generate illogical choices is because a committee itself is a complex entity with complex dynamics. Sometimes the decisions are not just the most logical decision for the problem but are explainable considering other aspects like the agendas and views of the different members or even psychological dynamics in a meeting (e.g. approving a little nonsense by someone that didn't say anything else in the meeting because it would look bad to be too hard about it).
      The second reason (history and evolution) can generate problems because C++ has been, at least in the past, very very strong about keeping backward compatibility with both C and previous versions of itself. This means that when a mistake slips in, and it's not just an obvious "typo" but a semantic decision that just turned out later to be a bad idea, then it's impossible to fix it because that would be a change, breaking all existing code that is depending on that questionable idea. Also over the years even the focus shifted a little and thus different parts of the language don't seem to be aligned on the same "view" (but earlier parts cannot be changed because of backward compatibility.
      This strong backward compatibility value started to change some years ago, and for example now there is even a "deprecation" model that makes conceivable to stop compiling C++ code that was compiling before. Even changes in runtime behavior of the same code is now acceptable, provided the cases in which the difference is visible are considered absurd and not common.
    • There are no runtime errors, but undefined behavior
      C++ has always been focused on speed (the idea is to leave no room for a lower-level language between C++ and the metal... with possibly the exception of a little bit of assembly). The focus on speed is very important and thus for example doing checks at runtime that would trigger only for buggy code is considered a bad idea (why slow down correct code to help bad code? fix the bad code instead!).
      For example, accessing the element with index 10 of a ten-element array doesn't trigger an out of bound error (like it does in Python, Java and most other languages) but just is declared Undefined Behavior (UB) that means that anything may happen.
      This UB concept is problematic and subtle because often "anything" translates in "nothing" and buggy code simply seems to be working (just to bite back when the damage will be maximal).
      Undefined behavior can be extremely surprising, and consider that it can even travel back in time.

    To sum it up C++ is hard to experiment with because it's complex, logic doesn't always help and when you make a mistake there is no guaranteed negative feedback. In such an environment it's not trivial to build a mental model of how things work and thus, in my opinion, a much saner path starts with a bit of reading.
    I'd suggest:

    • The C++ Programming Language
      By Bjarne Stroustrup, the original designer of C++.
    • Effective *
      A series of books by Scott Meyers. Those are very nice also because of the format: a collection of small mostly self-contained tips and specific cases discussed in detail that are much easier to "swallow" than a huge discussion about complex parts.
    ## Javascript
    A terrible language but with a fantastic runtime environment.
    Actually the main architecture is ok, but the details about the syntax and semantic are terrible and make, in my opinion, the language unsuitable for complex software.
    What is really great is the runtime environment in an HTML5 browser. The amazing combination of three badly designed and quirk-riddled languages like HTML, CSS and Javascript still make a powerful combination (for the end user).
    ## Lisp (my own dialect)
    Lisp is a real eye-opener. More than a language is a distilled essence of programming on which you can build any language you like.
    With "Lisp" I mean here the generic Lisp idea with generic macros without self-inflicted amputations like the ones of Scheme (template-based macros only, only recursion for iteration) or of Clojure (functional approach only).
    For reasons that are not completely clear to me basically no one (numerically speaking) uses Lisp today. While ignorance (not knowing about it) may be a reason for the majority, even who knows very well (Peter Norvig, director of research at Google, for example is an expert lisper) still is not using it and this puzzles me.
    I write most Lisp code in my own implementation of Lisp named JsLisp (a Lisp compiler targeting Javascript).
    JsLisp is a compile-only implementation and this is possible because Javascript provides eval that allows the creation of new Javascript code at runtime.
    When such a feature is not present a Lisp compiler needs also to embed a Lisp interpreter because Lisp macros require the ability of execute arbitrary Lisp code at compile/read time.
    The project is hosted on github and some documentation (albeit a bit outdated) is available on JsLisp website.
    Here you can see a small video showing the JsLisp IDE (running inside a browser) and some demo programs written with JsLisp.
    ## C
    I don't use C very often now because I am currently working on mostly on PCs where resources are abundant. Still the language is a nice "portable assembler" with which you can implement nice efficient algorithms without being annoyed by registers and other limitations of CPUs.
    For quite a long time I have been writing software using the object-oriented paradigm in C using a base object as first element of a derived object for inheritance and using function pointers to emulate dynamic dispatching.
    Now I don't do this anymore and I simply work with C++ when it makes sense and there are no resource or interoperability problems.
    ## Assembler
    I started writing code in assembler and, in retrospect, I think this is a great path to programming.
    Assembler is conceptually simple (or at least it was simple on Apple ][, with the fantastic 6502 processor from which I took my nickname).
    I strongly believe that the human brain is wired for a concrete-to-abstract path when understanding and thus it makes sense to start with something so simple and concrete that even the "function" or "variable" concept are not natively present.
    Once you have a mental model of how things are working it is easy to build abstractions like higher-level languages as BASIC. Even things like a BASIC string are not "magic", but something you have a rough idea of how are implemented.
    For quite a long time I coded mostly in assembler only and even the first commercial program I completed (PaintStar, a pixel oriented paint program for the Apple ][ family) was written entirely in assembler.
    I have devoted to my memories of 6502 a few emulators. One is a simple text-mode only Apple ][ emulator written long ago in C and assembler and another is an emulator of the processor only written in Javascript (my first and so far only attempt at writing a JIT compiler - something slightly more complex for a processor where self-modifying code is permitted and commonly used).
    After 6502 I worked quite a bit using 68k processors and x86 in 16/32 bit (including manually selecting instructions for u/v pipelines parallel execution in early Pentium :-D).
    I've a much smaller experience in assembler-level 64-bit coding but I'm currently filling this gap (I'm working on my first 64-bit native-code compiler for a Lisp dialect that generates x86-64 machine code directly, without depending on a C compiler or an assembler).
    The calling convention uses rdi for as a pointer to the argument list and rsi as the argument count. Values are all stored in 64-bit unions tagged using lowest 1-3 bits (one bit only for floating point values, leaving 63 bits for the numeric value itself) and the call convention is compatible with a "C" declaration of Value foo(Value *args, int count).
    The first argument is always the closure object itself; variadic calls do transform the arglist parameter into a list object in the function prologue.
    I'm still reworking on the base object representation but I expect that leaving parameters out of the call stack will allow a simpler gc logic still not paying too much in performance.
    So far I'm impressed with the speed that even for such a naïve implementation is not far from much more sophisticated JIT compilers (there is however no gc yet).
There are also programming languages I avoid.
I think that language can shape the thought process, and this is true also for computer programming languages.
Some languages really allow you to think the previously unthinkable and provide you with new weapons that can be used when fighting with real world problems.
Other languages instead cripple your brain and to keep it confined to predefined schemas. The sad part is that this, in some cases, is not an accident but the target for the language: the aim is not enabling programmers to do great things, but just avoid doing big damages.
While of course big damages are not a good thing the solution, in my opinion, is more on the education and practice than in using blunt tools. A bad programmer is not going to write good code if provided a bad language.
Languages that I prefer to avoid are
    ## Java (crippled and limited by design).
    Long ago, when I was working for Enel, I was given the opportunity of following a Java training and I got a Java 1.1 certification.
    Even in the course I however got the impression that the language was full of bureaucracy and wrinkles designed to constrain.
    I did not follow closely the evolution of the language as I decided back then that I would try to avoid the language if possible. So far my strategy has been successful.
    ## PHP (obscenely crippled because designed by illiterate).
    PHP is bad, so bad that's not even funny. It's so bad that it's hard to distinguish it from a caricature of bad programming languages.
    Amazingly enough it's a winner on the web: the worst possible language is probably one of the most popular. Go figure.
    In my opinion the bad parts of PHP (most of it) are however due to simple ignorance and not to a deliberate design. Moreover the limitations in the language are there because the implementers didn't know how to get certain features in, not because they didn't want programmers to use them.
    ## COBOL (maybe ok in '70s as better than assembler; inexcusable today).
    ## FORTRAN (ditto).
    ## C++ extreme metaprogramming
    C++ as a programming language is full of problems but still reasonable. The metaprogramming part however is awful.
    Exploiting what I think was an unwanted side-effect of the implementation of C++ templates, after some years people discovered that the absurdly intricate rules for overloading resolution and the strange rule SFINAE ([template] Substitution Failure Is Not An Error) could be used to make IFs and recursion-based loops at compile time during template expansion.
    This would have been no problem if many programmers (not understanding there is a difference between what can be done and what should be done) wouldn't have started using these unwanted and unanticipated "features" (basically bugs) to try to write real metaprogramming code.
    This amazingly enough ended up in huge template-based undebuggable libraries that take forever to compile, that exhaust at compile time all existing RAM, that give tens of screenfuls of error message when you make a typo.
    Even more amazingly this absurd way of programming made it into the standard. Compilers over the years have been fixed to handle "complex" (i.e. non completely trivial) templates when in the early days for example recursion limit was a problem. C++11 now incorporates some of the monstrosities that template metaprogramming obsessed guys designed.
    Thanks to this absurd choice of chosing brainf**k to do metaprogramming it's still too much to ask for a real loop, data structures or even just enumerating members of a class at compile time. Non-trivial metaprogramming in C++ is still impossible (and IMO it's not something that will be fixed in any foreseeable future).
    My choice for metaprogramming in C++ is writing external code generators in Python. Of course not an ideal solution as there is no compiler telling you what it knows and you're forced instead to work at the text level (a nightmare thanks to C++ very complex grammar), but still way better than writing metaprograms in such a sad and poor environment.
I am quite active on StackOverflow and I like programming competitions and problem solving sites like Project Euler, TopCoder or HackerRank.
It is now several years that I am not using Windows anymore neither at home nor at work. At home I am currently using a Linux Arch system and I also have an old Mac mini and an old iPad on which I experiment a bit on OsX. I only use Windows to check that software works reasonably on those systems.
I'm considered old and I started behaving as it's customary for old people.
I am now 55 and I noticed that the industry seems thinking that I am too old to be interesting as a resource (the number of job offers I get diminished in recent years).
I'm however currently a lot less publicly involved compared to long ago so maybe this is also playing a key role.
I found recently in myself a sort of automatic repulsion for anything that is new but I'm consciously trying to compensate that. Maybe it is normal for aging people to become more reluctant to changes, maybe now a default evaluation of "rubbish" is the statistically correct point of view.
I never started a serious blog or video channel (only experimented with different platforms) but none the less here are a few points that I've come to over the years:
  • Reading is good, reading too much is bad
    In coding, reading can give some directions but the only way to learn is by writing a lot of code. If you read too much you may get the ILLUSION you know a topic, but in reality you are not able to do anything useful in that area.
  • Ignore non-english material, avoid producing it
    There's nothing to gain by splitting knowledge by country. If you've problems with technical English, first fix that. I was born and I'm currently living in Italy but, in my opinion, it's better bad English than good Italian if we're talking about programming.
    Localizing is focusing on differences and divisions instead of focusing unity and on what is common; this is true in my opinion for everything, but while localization may be acceptable or even the right thing for non-technical people (for historical reasons, not as a value per se), I think it's absolutely incomprehensible and inexcusable in STEM.
  • Once done, do it again
    When writing code it's normal to make decisions and later in the construction one may realize that a better approach was possible. Changing would mean however to restart from scratch and rewriting a lot of existing code. My suggestion is to still complete and get to "the end" of the project (i.e. if you're writing a chess-playing program get to a program that maybe is not very strong but can at least play a full game). When you finish your project, start it over from scratch. This time you have the experience to fix those early mistakes and get a better one... but it wouldn't be surprising if during the new implementation you get another even better idea about the basics. My suggestion is to still finish the second version but, after that, start again. And over. And over. You can move to something else only after a few iterations, not before (if you care about writing decent stuff).
  • If it works, you didn't look close enough
    In complex software (i.e. not completely trivial) there will be bugs or at least behaviors that were unintended. When everything works as expected it simply means that you didn't look close enough. Note that this is not an excuse for leaving bugs in, or for coding without thinking too much because there will be bugs anyway... just a reminder that the code is most probably not doing what you think is doing. When debugging is also important to accept that you don't know what the program is doing (if that was the case the bug wouldn't be present) and carefully measure and inspect everything. It can be really surprising how things could go wildly and badly internally but can still be apparently (or even effectively) "working as expected" by mere coincidence. When facing bad behavior and debugging looking for the issue it's not uncommon to find other genuine bugs that are however unrelated to the observed issue.
  • Your time is precious, choose wisely what to invest on
    When I started IT information was scarce and expensive; I was paying for compilers (not kidding). Today we're in the opposite situation (free compilers, documentation, even free tiers of cloud computing and storage power are available) and everyone in coding is flooded with information, tips, techniques, frameworks, libraries, languages, methodologies, tools... all for free. Most of that is useless nonsense (90% of everything is crap) so it's important to choose correctly.
    I don't think there is an easy way to guess, but a few of rules I gave myself are:
    • Never invest free time on details of a commercial product
      No matter how popular it is, no matter how wonderful it is, a commercial product is in the hand of a single company that will have its own agenda that most probably will be different from yours. What is good for them will not be what is good for you. A marketing department will destroy the product or change it into something else because they think it will be for their advantage. All your knowledge about the details of that product will vanish instantly and will be pointless. It's ok to invest time on a commercial product if you're getting some other compensation (e.g. money) but you're not "growing" by learning it because in a few years all you added to yourself will disappear instantly. It's not about if, it's only about when. You're just accumulating future nonsense rubbish.
    • Never invest free time on details of a fake open source product
      Something may be "open source" in theory, but with a firm grasp by a single company or paying entity, and in that case the situation is not very far from a commercial product. While IN THEORY if the "owner" steers development in a direction you or the community doesn't like, you or the community could simply fork the project... this in practice is often unfeasible. For example, if one great asset of the product is the documentation then a truly volunteer-driven open source version has no hope to stay relevant in my opinion. While everyone loves to read good documentation, writing it requires a lot of work and it's not that funny... so must be paid work.
      What you can get from volunteer work is at most a crappy documentation generator; the documentation itself (where the real value is) requires a LOT of repetitive work that never ends (you must keep updating it as something evolves).
    • Popularity is a very bad proxy for quality
      If something is very popular maybe it's good and worth your attention, but probably it's not. Popularity is a strange property and the best proxy for popularity is popularity (i.e. the more something is popular and the more will gain in popularity). It's a lottery. There is good stuff that will never be popular and literal rubbish that will be number 1 on all charts for a long time.
    • Apply the 10 years rule
      If you plan to invest your time on some topic, where was that topic 10 years ago? What makes you think it will be still relevant in 10 years? IT is moving at an amazing speed and most of what is super trendy now will be completely forgotten soon. If what you're interested in has a version number then the risk is high that a new version will come out pretty soon. Maybe it's not a problem, but maybe it is.
  • The trick is that there is no trick
    This last one is in my opinion the most important. Everyone is looking for tricks but the only good trick I found is realizing that there is no trick and that you must just work your ass off. It's that simple.
I studied pure math at university (and loved it) but I always worked programming computers, something for which I only have a high-school diploma.
I took my degree in math while working full time because I liked the subject (I fell in love with the book Algebra of I. N. Herstein, one of the reasons of my change from engineering to math).
I never formally used my degree for work (except that I was able to work in the US thanks to it because immigration H1-B rules at the time were translating "specialized" with "university degree").
I don't believe too much in the value of formal studies: much much more important to me is the personal dedication and interest into the subject. For example, I know very good programmers that have no formal training at all (not even high-school level) and instead I know IT graduated that literally don't understand anything they are talking about.
When I'm requested to evaluate someone, having a sci/tech degree only means to me that the subject can (or could at the time) provide some continuity: it is not a proxy for being smart or being able to deliver in the specific field.
Of course NOT having a degree doesn't mean the subject cannot provide continuity.
I am also a bad pianist, a bad chess player, bad speedcuber and bad in sports. If you however never attempted seriously any of these activities I am probably much better than you (with bad I mean "worse than the average of who does it seriously").


I started studying piano when a kid because my mother wanted us to try many different paths when young (thanks mom!). I didn't practice much however, and in the end I dropped out of the musical school after about three years of piano.
Many years later I fell in love with Chopin and got back to piano this time self-teaching. The following is a video of one of my early takes at what is considered the dream of hobbyist piano players: Chopin Fantaisie Impromptu Op. 66...


I am a FIDE rated player with a current rating of about 1800. This means that I am weaker than the average of rated players but that I just don't make random moves on the board so even a professional player needs to pay some attention. I am also a chess arbiter and a few times every year I spend time in chess tournaments.
I also take care of online broadcasting of chess events using a software I wrote myself after reverse-engineering the serial protocol used by DGT chessboards (the software provided with the board was pricey and, more importantly, really really bad).
I own 5 DGT boards and about 20 wooden tournament boards and clocks with which I'm happy to help friends when they need to organize a chess tournament.
I also wrote several chess-playing programs some of which ended up being stronger than me (writing a chess engine is a very good programming exercise in my opinion, especially for debugging).


I first met Rubik's cube thanks to my brother in 1983, but after a short time I forgot about it. Recently I found it back in my mother's house and decided to look at what it is now. I re-started learning it using more efficient methods (I'm using CFOP) and my average is now around 35 seconds... my aim is to get to have an average below 30 seconds (it's going to be still a long way :-D ).
I wonder why Rubik's cube is not used in math classes when teaching about finite groups, it's IMO a very good example and a lot of apparently strange concepts like subgroups, cycles, closures, can be shown quite clearly on the cube and other twisty puzzle variations.
I also own and love a Megaminx, but I'm still horribly slow at solving it... somehow I've always had sort of a fetish for dodecahedrons.


I've been regularly practicing karate, ski and swimming for some time; now I enjoy running for medium distances (about 40min). I've ran a few half-marathons (I was tempted by the full marathon but read that while being of course feasible by most people it's not really good for health as it's a bit past the sustainable limit).
I live in Vigevano, a small city (~60k residents) not far from Milan, in northern Italy.
I work for COMELZ, a company that produces machines and software for the shoe/apparel/leather goods market.
The company recently acquired develer a company created by long time friends from Florence.
Now the software team working on comelz products is much bigger (tens of developers) and much more organized with all pros and cons of the situation.
I've been basically of a lone wolf for very long time and there are a few aspects of working in big groups I'm slowly getting used to again (like meetings ;-) ).
I wrote much of the initial version of software for our cutting machine family (CMxx/CZxx) and this allowed me to work in several interesting areas:
    ## Axis control
    In our company we produce ourselves the brushless motors we use in our machines, both the mechanical parts and the control electronic.
    I wrote the software that computes online the space positioning with 1ms accuracy minimizing total time while keeping the motion below dynamic limits.
    ## Work plan optimization
    The main cutting machine family is multi-headed, in other words there are multiple cutting heads working simultaneously on the same work area.
    Simplifying a bit, work plan optimization is like solving TSP instances adding the time dimension and where there are multiple travelers that cannot get closer than a certain distance from each other.
    ## Artificial vision
    Our machines can also be equipped with cameras than can be used for several purposes; from user interface (pen tracking) to positioning error compensation to printed material logo detection for cut shape alignment to leather boundary detection.
    I also recently worked on the interesting problem of texture classification and found a novel "smart" filtering approach that allows our machines to work on previously difficult to handle materials (for example single color or low contrast digital fabrics where the design is visible only because of texture change).
    ## Cut waste minimization
    This is the very interesting problem of placing parts inside a leather, sheet or roll so that the minimum amount of material is wasted. I was able to design an implement algorithms that performed quite well compared to the current state of the art in industry.
    nesting example
    Nesting of a shoe model on leather avoiding defects.
    Of course even the most basic version of the problem is NP-complete as it can be trivially seen that as 1-D subcase you can model the knapsack.
    This is also my main area of interest at the moment.
to name a few.
I also designed and partly implemented early versions of our specialized CAD/CAM software for the shoe/apparel/leather good industry Caligola4: a full reimplementation and extension of a previous CAD software of the company.