What is the modern popular standard Prolog that hobby Prolog interpreters should aim for?

Question

I'm writing a Prolog interpreter as an exercise and wondering what I should be aiming for. Unfortunately there are many versions of Prolog to choose from and they are documented to various degrees. I quickly found this question from someone who was apparently expecting far too much from the internet by wanting a detailed html specification of Prolog. The answer to that was that you can get the ISO standard for $30, but that's rather impractical. Users are never going to pay $30 just to read about Prolog when they can get a Prolog interpreter for even less money, so if you pay the money and conform to the standard few people will ever recognize your effort. Therefore it doesn't surprise me at all that the ISO standard isn't universally respected.

Starting from the assumption that the ISO standard is a joke, what is the real version of Prolog that an interpreter should be aiming for? I don't mean that every little Prolog interpreter should fully implement every feature, but when constructing a Prolog interpreter there's no end to the little decisions that must be made. How should someone discover the consensus of the Prolog community about what Prolog should be?

Answer 1

If you are writing a Prolog system as an exercise, don't expect that you get too much done. After all, it is quite an effort.

To start with, aim for the core of the ISO standard, that is 13211-1:1995 including Cor.1:2007 and Cor.2:2012. That core is pretty much supported by many systems like: IF, SWI, YAP, B, GNU, SICStus, Jekejeke, Minerva. So while this core just covers the very basics, it will be still a lot of work to you.

Then, you can consider what further direction you want to go. From a standard's viewpoint these are implementation specific extensions. Systems pretty much differ in the way they offer extensions, so there is no clear way to choose. The most popular systems are SICStus (commercial) and SWI (open source). An open source system with better conformance than SWI is GNU.

You are putting a lot of quite debatable implications into your question, so let me try to sort some out:

Price of standards. ISO standards do cost something - these documents have a certain legal status - depending on your country and legislation. Freely available web documents can serve as evidence only. See for example the C standard which you get for the same prices: One official high price (USD 285) and a reduced one by INCITS (USD 30). The difference is only the cover sheet. At least, you can get the Prolog standard for a significantly reduced price.

Relevance. There is just one standard. And systems conform quite closely. Where they differ, they differ rather randomly. As an example, look at this detailed comparison of syntax which covers both reading and writing terms. Typically, such differences are reported by users who get hit by one or another difference. These differences are nowhere formally defined.

Answer 2

I don't agree with your assumption about ISO Prolog, indeed I would suggest to try to implement a small subset of ISO Prolog (i.e. be 'sure' to properly implement findall/setof).

A principal problem with ISO standard it's the module directive. Then choose an implementation to model modules, or skip them altogether.

Even some 'undiscussed' builtin will be difficult to implement, depending on the language you use (C, Haskell, Lisp, SQL, Javascript, C++...) and the choices you will do about the degree of translation. Most implementations out there are not interpreters, but bytecode compilers with various degrees of runtime support. The most used choice for the bytecode level is Warren's Abstract Machine (WAM, as you surely know).

When I wrote my Prolog interpreter, many years ago, I designed and implemented an object oriented database model, using algorithm ABC instead of the WAM, and I designed and implemented the Variables handling with ingenuity... but I left out setof/bagof, for instance...

Answer 3

I think SWI-Prolog pretty-much drives the Prolog standard nowdays, so take a look at their documentation... Other than that, what you are asking is being debated over and over again during the past years in the Prolog Standardization meetings. Some argue that tabling should be made into the standard, others claim the same for automatic indexing, and so on. So, in my humble opinion, the best you can do is "mimic" what SWI does for most stuff, and you'll be almost certainly in the standard