The Mercury Programming Language

Low-level versus high-level

A low-level language requires closer attention to the specific hardware features of the computer system running the program than a higher-level language would demand. At the lowest level is the binary command set for the cpu. C and C++ are higher-level than that, and Fortran, Ruby and Ada are high level languages. Although many higher-level languages have special features allowing contact to specific hardware features, it is easy in these languages to write programs that run properly with little or no attention to the specific hardware it runs on. Mercury is a high-level language.

Practical versus research

A practical computer language is directed toward usage by those who are interested in getting data processed at a reasonable time and cost. They are interested in answers, not arcane computer science questions. They want their analysts and programmers to write software with a minimal amount of bugs and which is not hard to maintain. A research language supports scientific investigations, either computer science or any other investigations requiring something special not easily available using the usual languages (like Fortran, C, Visual Basic, Perl, Python etc.). In order to satisfy unique research requirements, investigators will put up with a remarkable amount of inconvenience with respect to obtaining an exotic language, building the compiler or other support software, struggling with the documentation, and so on. This is not what the commercial user wants to deal with. Mercury aims to be a practical language.

Compiled versus interpreted

This refers to the expected run-time environment for the language. The same program could be either compiled or interpreted, depending on what run-time software is used. When a language is compiled, it is transformed to a form which can be run as process under the operating environment of the computer system. When a language is interpreted, it is transformed to data which is to be used by an additional program, the interpreter, which is running on the target computer system. The introduction of the interpreter as a middleman results in additional consumption of computer resources (time and/or space) when the program is run. Mercury is a compiled language. This is in the interest of speed, as Mercury aims to be used for large practical problems.

Imperative versus declarative

Computer languages started out all being imperative. That means the main emphasis is on the algorithms used to process the data. There has been a movement among both computer scientists and actual software developers in which some persons want the computer languages they use to more closely reflect the problems they are solving. This is in opposition to closely reflecting the computer architecture they are using, and constitutes a shift toward a higher level. Some speak of "compilable requirements", meaning that by the time you have an accurate requirements specification there is enough exact information available that a computer could produce the acutual runnable program. Lisp is a good example of a declarative language, as is Prolog, which is a logic language. Mercury is in the declarative camp.

Iteration versus recursion

Iteration refers to repeating a process on some data by repeating an action on each chunk of data. In some languages, iteration is not available directly. Instead, operation on a big structure is accomplished by finding that the big structure is made up of a bunch of structures to the big structure, so that the routine after acting with respect to the big structure calls itself to deal with the similar component structures. (If this seems confusing, it is not surprising—most people not already familiar with recursive methods need more than an introductory paragraph to get comfortable with it). Mercury uses recursion, not iteration. This goes with being a declarative logic language.

Object-oriented programming

This is a technique in which the problem environment is modeled in terms of "objects", which are entities each of which have data attributes and operations upon them. Some languages do not directly support this concept: examples, C or Fortran. Some languages provide support for this style of programming, but it is not exactly built into the language, and it is not necessary to progam in an "object-oriented" way. Examples: Perl, ocaml. Some languages are defined from the ground up for object-oriented programming. Examples: C++,Python,Ruby,Java. Mercury is like Perl or ocaml, in that features are provided to aid programmers who wish to adopt an object-oriented style.

Functional programming

This refers to languages which organize computations into functions which have the property that the result of applying the function algorithm to its inputs depends only on the inputs to the function. This means that a routine cannot have so-called "side effects", that is, do something not implied just by the inputs. Printing out results is a side effect, and so are various other things you might want your computer to do. The implication is that you can have the bulk of your code purely functional, but you must also somewhere have a part that is not purely functional. For some languages this involves a few contortions, but this is OK if they are minimal, and you gain the desired benefit of the rest of your code (presumably the hard part) being especially easy to understand and maintain. Mercury is a functional programming language like ML, Haskell or ocaml. It has very precise tracking of the necessary "impure" code.

Logic programming

Logic programming languages address the problem of finding valid consequences of given axioms and postulates. These languages are based on mathematical logic, comparable to, for instance relational database being based on set theory. Logic programs have application in artificial intelligence. They also have been used to prove that certain programs are valid, i.e., are doing what they are supposed to. An example would be showing an encryption program is valid. Program proving is so difficult it is not widely done. "Research Continues" Mercury is a logic programming language.

Backtracking versus other

Backtracking refers to the feature of a language which applies when at a certain point the computation can proceed along alternative paths (each of which may require extensive calculations) but it is not known which (if any) will provide an acceptable result. The backtracking feature allows a checkpoint to be taken so that if this path does not work out, the computation revert to before the alternatives, and another alternative can be tried. This could be all arranged by the programmer, but it is immensely easier when the language supports it. Mercury provides backtracking, like Prolog or transaction rollback in SQL.

Compile-time type-checking versus run-time type-checking

In programs, data structures are assigned types, which determine what operations may be performed on them. It is easy to make mistakes, and write a programming command which specifies an invalid operation. There are three options for what happens next: 1. The error is not found except possibly by detection of bad output when the program is used. This is the worst outcome. 2. The error is found when the program is run, because the program contains checks which signal that an invalid operation was attempted, or that the result of an operation was invalid. Here at least we know that there was a problem. 3. The error is found when the program is compiled. The programmer can fix the problem before delivering a bug. Which option is desirable depends on how important is is for the program to be reliable. Option 2 means a lot of work by the programmer if reliability is to be achieved. However, it is not hard to get programs to compile, and compilers are relatively cheap to develop and thus to buy. Option 3 requires plenty of work to develop a compiler, and demands careful type definition and use by the programmer. All this is to detect bugs earlier and make it less likely that a bug will be delivered. Mercury, like Ada or ocaml, performs extensive compile-time checking.

Mercury is the result of adding functional programming (like Haskell) and full compile-time type checking (like Ada or ocaml) to logic programming (like Prolog). Some might look at this as having the both the desirablity and the likelyhood of a successful effort to hybridize armadillos and hyenas. Nonetheless, the project has produced a running compiler and a full build environment. It works on Linux and Windows and some other places. The compiler is fast and so are the resulting programs (relatively speaking).