Programming language is an artificial [formal](formal_language.md) (mathematically precise) language created in order to allow humans to relatively easily write [algorithms](algorithm.md) for [computers](computer.md). It basically allows a human to very specifically and precisely but still relatively comfortably tell a computer what to do. We call a program written in programming language the program's **[source code](source_code.md)**. Programming languages often try to mimic some human language -- practically always [English](english.md) -- so as to be somewhat close to humans but programming language is actually MUCH simpler so that a computer can actually analyze it and understand it precisely (as computers are extremely bad at understanding actual [natural language](natural_language.md)), without ambiguity, so in the end it all also partially looks like [math](math.md) expressions. A programming language can be seen as a middle ground between pure [machine code](machine_code.md) (the computer's native language, very hard to handle by humans) and natural language (very hard to handle by computers).
For beginners: a programming language is actually much easier to learn than a foreign language, it will typically have fewer than 100 "words" to learn (out of which you'll mostly use like 10) and once you know one programming language, learning another becomes a breeze because they're all (usually) pretty similar in basic concepts. The hard part may be learning some of the concepts.
A programming language is distinct from a general computer language by its purpose to express algorithms and be used for creation of [programs](program.md). This is to say that there are computer languages that are NOT programming languages (at least in the narrower sense), such as [HTML](html.md), [json](json.md) and so on.
- **compiled** languages: Meant to be transformed by a [compiler](compiler.md) to a [native](native.md) (directly executable) binary program, i.e. before running the program we have to run it through the process of compilation into runnable form. These languages are typically more efficient but usually more difficult to program in, less flexible and the compiled programs are non-portable (can't just be copy-pasted to another computer with different [architecture](isa.md) and expected to run; note that this doesn't mean compiled languages aren't [portable](portability.md), just that the compiled EXECUTABLE is not). These languages are usually [lower level](low-level), use static and strong [typing](typing.md) and more of manual [memory management](memory_management.md). Examples: [C](c.md), [C++](cpp.md), [go](go.md), [Haskell](haskell.md) or [Pascal](pascal.md).
- **interpreted** languages: Meant to be interpreted by an [interpreter](interpreter.md) "on-the-go", i.e. what we write we can also immediately run; these languages are often used for **[scripting](scripting.md)**. To run such program you need the interpreter of the language installed on your computer and this interpreter reads the [source code](source_code.md) as it is written and performs what it dictates (well, this is actually simplified as the interpreter normally also internally does a kind of quick "lightweight" compilation, but anyway...). These languages are generally less efficient (slower, use more RAM) but also more flexible, easier to program in and [independent of platforms](platform_independent.md). These languages usually [higher-level](high_level.md), use weak and dynamic [typing](typing.md) and automatic [memory management](memory_management.md) ([garbage collection](garbage_collection.md), ...). Examples: [Python](python.md), [Perl](perl.md), [JavaScript](js.md) and [BASH](bash.md).
Sometimes the distinction here may not be completely clear, for example Python is normally considered an interpreted language but it can also be compiled into [bytecode](bytecode.md) and even native code. [Java](java.md) is considered more of a compiled language but it doesn't compile to native code (it compiles to bytecode). [C](c.md) is traditionally a compiled language but there also exist C interpreters. [Comun](comun.md) is meant to be both compiled and interpreted etc.
Another common division is by **level of [abstraction](abstraction.md)** roughly to (keep in mind the transition is gradual and depends on context, the line between low and high level is extremely fuzzy):
- **[low level](low_level.md)**: Languages which are so called "closer to [hardware](hardware.md)" ("glorified [assembly](assembly.md)"), using little to no abstraction (reflecting more how a computer actually works under the hood without adding too many artificial concepts above it, allowing direct access to memory with [pointers](pointer.md), ...), for this they very often use plain [imperative](imperative.md) paradigm), being less comfortable (requiring the programmer to do many things manually), less flexible, less safe (allowing shooting oneself in the foot). However (because [less is more](less_is_more.md)) they have great many advantages, e.g. being [simple](kiss.md) to implement (and so more [free](freedom.md)) and **greatly efficient** (being fast, memory efficient, ...). One popular definition is also that "a low level language is that which requires paying attention to the irrelevant"; another definition says a low level language is that in which one command usually corresponds to one machine instruction. Low level languages are **typically compiled** (but it doesn't have to be so). Where exactly low level languages end is highly subjective, many say [C](c.md), [Fortran](fortran.md), [Forth](forth.md) and similar languages are low level (normally when discussing them in context of new, very high level languages), others (mainly the older programmers) say only [assembly](assembly.md) languages are low level and some will even say only [machine code](machine_code.md) is low level.
- **[high level](high_level.md)**: Languages with higher level of abstraction than low level ones -- they are normally more complex (though not always), interpreted (again, not necessarily), comfortable, dynamically typed, beginner friendly, "safe" (having various safety mechanism, automatic checks, automatic memory management such as [garbage collection](garbage_collection.md)) etc. For all this they are typically slower, less memory efficient, and just more [bloated](bloat.md). Examples are [Python](python.md) or [JavaScript](js.md).
We can divide language in many more ways, for example based on their **[paradigm](paradigm.md)** (roughly its core idea/model/"philosophy", e.g. [impertaive](imperative.md), [declarative](declarative.md), [object-oriented](oop.md), [functional](functional.md), [logical](logical.md), ...), **purpose** (general purpose, special purpose), computational power ([turing complete](turing_complete.md) or weaker, many definitions of a programming language require Turing completeness), [typing](data_type.md) (strong, weak, dynamic, static) or function evaluation (strict, lazy).
- **[syntax](syntax.md)**: The grammar rules and words, i.e. how the language "looks", what expressions we are allowed to write in it. Syntax says which words can follow other words, if indentation has to follow some rules, how to insert comments in the source code, what format numbers can be written in, what kinds of names variables can have etc. Syntax is the surface part, it's often considered not as important or hard as semantics (e.g. syntax errors aren't really a big deal as the language processor immediately catches them and we correct them easily), but a good design of syntax is nevertheless still very important because that's what the programmer actually deals with a great amount of time.
- **[semantics](semantics)**: The meaning of what we write, i.e. semantics says what the syntax actually stands for. E.g. when syntax says it is possible to write`a / b`, semantics says this means the mathematical operation of division and furthermore specifies what *a* and *b* can actually be, what happens if *b* is zero etc. Semantics is the deeper part as firstly it is more difficult to define and secondly it gives the language its [features](feature.md), its power to compute, usability, it can make the language robust or prone to errors, it can make it efficient or slow, easy and hard to compile, optimize etc.
**What is the best programming language and which one should you learn?** (See also [programming](programming.md).) These are the big questions, the topic of programming languages is infamous for being very [religious](holy_war.md) and different people root for different languages like they do e.g. for [football](football.md) teams. For [minimalists](minimalism.md), i.e. [suckless](suckless.md), [LRS](lrs.md) (us), [Unix](unix.md) people, [Plan9](plan9.md) people etc., the standard language is **[C](c.md)**, which is also probably the most important language in [history](history.md). It is not in the league of the absolutely most minimal and objectively best languages, but it's relatively minimalist (much more than practically any [modern](modern.md) language) and has great advantages such as being one of the absolutely fastest languages, being extremely well established, long tested, supported everywhere, having many compilers etc. But C isn't easy to learn as a first language. Some minimalist also promote [go](golang.md), which is kind of like "new C". Among the most minimal usable languages are traditionally [Forth](forth.md) and [Lisp](lisp.md) which kind of compete for who really is the smallest, then there is also our [comun](comun.md) which is a bit bigger but still much smaller than C. To learn programming you may actually want to start with some ugly language such as [Python](python.md), but you should really aim to transition to a better language later on.
**Can you use multiple programming languages for one project?** Yes, though it may be a burden, so don't do it just because you can. Combining languages is possible in many ways, e.g. by embedding a [scripting](scripting.md) language into a compiled language, linking together object files produces by different languages, creating different programs that communicate over network etc.
What really IS a programming language -- is it software? Is it a standard? Can a language be [bloated](bloat.md)? How does the languages evolve? Where is the exact line between a programming language and non-programming language? Who makes programming languages? Who "owns" them? Who controls them? Why are there so many and not just one? These are just some of the questions one may ask upon learning about programming. Let's try to quickly answer some of them.
Strictly speaking programming language is a [formal language](formal_language.md) with [semantics](semantics.md), i.e. just something akin a "mathematical idea" -- as such it cannot be directly "owned", at least not on the grounds of [copyright](copyright.md), as seems to have been quite strongly established by a few court cases now. However things related to a language can sadly be owned, for example their specifications (official standards describing the language), [trademarks](trademark.md) (the name or logo of the language), implementations (specific software such as the language's compiler), [patents](patent.md) on some ideas used in the implementation etc. Also if a language is very complex, it can be owned practically; typically a corporation will make an extremely complicated language which only 1000 paid programmers can maintain, giving the corporation complete control over the language -- see [bloat monopoly](bloat_monopoly.md) and [capitalist software](capitalist_software.md).
At this point we should start to distinguish between the pure language and its **[implementation](implementation.md)**. As has been said, the pure language is just an idea -- this idea is explained in detail in so called **language specification**, a document that's kind of a standard that precisely describes the language. Specification is a technical document, it is NOT a tutorial or promotional material or anything like that, its purpose is just to DEFINE the language for those who will be implementing it -- sometimes specification can be a very official standard made by some standardizing organization (as e.g. with C), other times it may be just a collaborative online document that at the same time serves as the language reference (as e.g. with Lua). In any case it's important to [version](version_numbering.md) the specification just as we version programs, because when specification changes, the specified languages usually changes too (unless it's a minor change such as fixing some typos), so we have to have a way to exactly identify WHICH version of the language we are referring to. Theoretically specification is the first thing, however in practice we usually have someone e.g. program a small language for internal use in a company, then that language becomes more popular and widespread and only then someone decides to standardize it and make the official specification. Specification describes things like syntax, semantics, conformance criteria etc., often using precise formal tools such as [grammars](grammar.md). It's hugely difficult to make good specification because one has to decide what depth to go to and even what to purposefully leave unspecified! One would thought that it's always better to define as many things as possible, but that's naive -- leaving some things up to the choice of those who will be implementing the language gives them freedom to implement it in a way that's fastest, most elegant or convenient in any other way.
It is possible for a language to exist without official specification -- the language is then basically specified by some of its implementations, i.e. we say the language is "what this program accepts as valid input". Many languages go through this phase before receiving their specification. Language specified purely by one implementation is not a very good idea because firstly such specification is not very readable and secondly, as said, here EVERYTHING is specified by this one program (the language EQUALS that one specific compiler), we don't know where the freedom of implementation is. Do other implementations have to produce exactly the same compiled binary as this one (without being able to e.g. optimize it better or produce binaries for other platforms)? If not, how much can they differ? Can they e.g. use different representation of numbers (may be important for compatibility)? Do they have to reproduce even the same bugs as the original compiler? Do they have to have the same technical limitations? Do they have to implement the same command line interface (without potentially adding improvements)? Etc.
Specification typically gets updated just as software does, it has its own version and so we then also talk about version of the language (e.g. C89, C99, C11, ...), each one corresponding to some version of the specification.
Now that we have a specification, i.e. the idea, someone has to realize it, i.e. program it, make the implementation; this mostly means programming the language's [compiler](compiler.md) or [interpreter](interpreter.md) (or both), and possibly other tools (debugger, optimizer, [transpiler](transpiler.md), etc.). A language can (and often does) have multiple implementations; this happens because some people want to make the language as fast as possible while others e.g. want to rather have small, [minimalist](minimalism.md) implementation that will run on limited computers, others want implementation under a different license etc. The first implementation is usually so called **reference implementation** -- the one that will serve as a kind of authority that shows how the language should behave (e.g. in case it's not clear from the specification) to those who will make newer implementations; here the focus is often on correctness rather than e.g. efficiency or minimalism, though it is often the case that reference implementations are among the best as they're developed for longest time. Reference implementations guide development of the language itself, they help spot and improve weak points of the language etc. Besides this there are third party implementations, i.e. those made later by others. These may add extensions and/or other modifications to the original language so they spawn **dialects** -- slightly different versions of the language. We may see dialects as [forks](fork.md) of the original language, which may sometimes even evolve into a completely new language over time. Extensions of the languages may sound like a good thing as they add more "comfort" and "features", however they're usually bad as they create a [dependency](dependency.md) and fuck up the standardization -- if someone writes a program in a specific compiler's dialect, the program won't compile under other compilers.
A new language comes to existence just as other things do -- when there is a reason for it. I.e. if someone feels there is no good language for whatever he's doing or if someone has a brilliant idea and want to write a PhD thesis or if someone smokes too much weed or if a corporation wants to control some software platform etc., a new language may be made. This often happen gradually (again, like with many things), i.e. someone just starts modifying an already existing language -- at first he just makes a few macros, then he starts making a more complex preprocessor, then he sees it's starting to become a new language so he gives it a name and makes it a new language -- such language may at first just be transpiled to another language (often [C](c.md)) and over time it gets its own full compiler. At first a new language is written in some other language, however most languages aim for **[self hosted](self_hosting.md) implementation**, i.e. being written in itself. This is natural and has many advantages -- a language written in itself proves its maturity, it becomes independent and as it itself improves, so does its own compiler. Self hosting a language is one of the greatest milestones in its life -- after this the original implementation in the other language often gets deletes as it would just be a burden to keep [maintaining](maintenance.md) it.
**So can a language be inherently fast, [bloated](bloat.md) etc.?** When we say a language is fast, bloated, memory efficient and so on, we often refer to its implementations because, as mentioned, a language is just an idea which can be implemented in many ways with different priorities and tradeoffs, so just keep in mind that talking about languages like this usually refers to the implementations. But on the other hand yes, a language CAN itself be seen as inherently having a similar property because it's simply such that its implementations more or less have to have this property. A very complicated language just cannot be implemented in a simple, non-bloated way, an extremely high level and flexible language cannot be implemented to be among the fastest -- so referring to language implementations we also a little bit refer to the language itself as an implementation reflects the abstract language's properties. **How to tell if language is bloated?** One can get an idea from several things, e.g. list of features, [paradigm](paradigm.md), size of its implementations, number of implementations, size of the specification, year of creation (newer mostly means more bloat) and so on. However be careful, many of these are just clues, for example small specification may just mean it's vague. Even a small self hosted implementation doesn't have to mean the language is small -- imagine e.g. a language that just does what you write in plain English; such language will have just one line self hosted implementation: "Implement yourself." But to actually [bootstrap](boot.md) the language will be immensely difficult and will require a lot of bloat.
Judging languages may further be complicated by the question of what the language encompasses because some languages are e.g. built on relatively small "pure language" core while relying on a huge library, preprocessor, other embedded languages and/or other tools of the development environment coming with the language -- for example [POSIX shell](posix_shell.md) makes heavy use of separate programs, utilities that should come with the POSIX system. Similarly [Python](python.md) relies on its huge library. So sometimes we have to make it explicitly clear about this.
Here is a table of notable programming languages in chronological order (keep in mind a language usually has several versions/standards/implementations, this is just an overview).
Some programming languages may be [interesting](interesting.md) rather than directly useful, however these are important too as they teach us a lot and may help us design good practically usable languages. In fact professional researches in theory of computation spend their whole lives dealing with practically unusable languages and purely theoretical computers.
One such language is e.g. **[Unary](unary_lang.md)**, a programming language that only uses a single character while being Turing complete (i.e. having the highest possible "computing power", being able to express any program). All programs in Unary are just sequences of one character, differing only by their length (i.e. a program can also be seen just as a single natural number, the length of the sequence). We can do this because we can make an ordered list of all (infinitely many) possible programs in some simple programming language (such as a [Turing machine](turing_machine.md) or [Brainfuck](brainfuck.md)), i.e. assign each program its ordinal number (1st, 2nd, 3rd, ...) -- then to express a program we simply say the position of the program on the list.
There is a community around so called **[esoteric programming languages](esolang.md)** which takes great interest in such languages, from mere [jokes](jokes.md) (e.g. languages that look like cooking recipes or languages that can compute everything but can't output anything) to discussing semi-serious and serious, even philosophical and metaphysical questions. They make you think about what really is a programming language; where should we draw the line exactly, what is the absolute essence of a programming language? What's the smallest thing we would call a programming language? Does it have to be Turing complete? Does it have to allow output? What does it even mean to compute? And so on. If you dare, kindly follow the rabbit hole.
