Syntax Vs Semantics

Syntax is what the grammar allows, semantics is what it means.

int x = "five"; // syntax is okay (type identifier = value), semantics is wrong ("five" is not an int).

No idea what the following is supposed to mean. It couldn't be more wrong.

Semantics is what matters in an interface because it maps one to one between what the user of the interface wants and the way the implementor of the interface does it.

Syntax is what doesn't matter to either one or the other party on each side of the interface.

If the user doesn't care about the distinction between two functions but the implementor forces them to make a distinction, then that is syntax.

Conversely, if there is no distinction between the implementation of two functions but the user forces the implementor to distinguish them, then that too is syntax.

Semantics is the one to one mapping between how the user thinks about the black box and between the black boxes' actual internals. Syntax is the measure of the discrepancy between the two.

Syntax that supports the user's view of the system is helpful and good. Syntax that violates that view in favour of the implementor's convenience is harmful and evil. An interface that has only good syntax adheres to SyntaxFollowsSemantics.

Since the user's view of the system is different from user to user, SyntaxIsSubjective. This is why syntax should be user-configurable.

Ok, that clarifies it quite a bit more. Certainly an unusual use of the words, but the concepts are now making sense. Hey, that sounds familiar... :-)

Are you assuming an idealized user? What about users who don't care about something, but really really should?

What confused me for a while was the distinction between good syntax and bad syntax.

The idealized user is me. :)

And the distinction kicks in this way; I really don't give a damn about the low-level details that compiler writers and linguists agonize about. Those things aren't even syntax to me, they're chaff that should be completely eliminated from my perceptions, buried so deep in the system that I never, ever encounter it.

Well, you gotta specify the semantics of something somehow. And unfortunately, to a great many people SyntaxMatters. Perhaps more than it should; some Smalltalkers like to complain that Smalltalk might have gained more widespread acceptance had it used curly braces rather than square braces. (I doubt it, but hey--it's a nice StrawMan). But there is a constant tension between what a) what's easy to read; b) what's easy to write (type in); and c) what's easy to process (both for the compiler itself, and for other tools that might process or generate source code). Some languages focus on terseness--C and perl come to mind. Fans of these languages like typing 'em in; as they don't have to repeat themselves too much. Others are very verbose--it's easy to read and figure out what is going on (and the compiler can use the verbosity to detect errors); but many programmers hate programming in them because of all the extra typing you need to do. And some languages--Lisp is the canonical example--emphasize easiness of parsing/processing. EssExpressions are as close to raw AbstractSyntaxTrees as you will get. The lack of syntactic distinction between special forms, macros, and functions is widely seen as a plus. But reading Lisp code can be a royal pain.

Therefore, if we understand correctly, you are trying to say in 3 rambling pages (could you try to make it one ? ) that syntax is altogether unimportant (how unimportant is left as an exercise). Well, if it matters to you syntax (writing) was one of the most significant achievements in the history of mankind. So syntax does seem to matter. You must be grateful that they forced the syntax of the English language on you, otherwise you'd be retarded.

You seem to further advocate that language designers (including, yes, GUI designers as we can take GUIs to be special purpose languages) should focus a great deal on the semantics (-- what kind of semantics, by the way ?) and, if possible, make syntax optional and replaceable like for example skins in UI interfaces. Let's say we try to design the next XML like language and we should therefore "focus" on trees as mathematical objects, certain kind of trees with certain properties vis-a-vis labels and order, and the universe of those trees and the operation defined in that universe will constitute the semantics on our new XML like language. And we somehow leave it as an option for the user to use XML syntax or EssExpressions syntax, or even their ad-hoc syntax they damn well please. Of course, we'd have to teach them how to use BNF or something like it, in order to be able to communicate their syntax to the system.

Well, it sounds very nice in theory, of course. There's one technical detail to resolve that in trying to "think" or "design" those mathematical objects that constitute semantics we have to use mathematical notation, even more bloody syntax.

So what you need resolve at a level better than handwaving is: In favor of extensible or pluggable syntax: maths, sciences, logics, engineering, architecture, medicine, law, market and trade, arts and crafts, and educated professions of all sorts tend to develop jargons and notations to aide in the expression of concepts. The ability to plug in these sorts of DomainSpecificLanguages or DomainSpecificTweaks may go a long way towards supporting developers - especially those who share interest across two domains, or who aim to make programming elements more accessible to an end-user. However, this isn't the same as having each programmer invent a new syntax; rather, it suggests maintenance of several 'popular plug-ins' by interested community groups, and some attempt at standardization within each domain. If the language itself is aware of syntax extensions it can aide developers by providing ProgressiveDisclosure - i.e. the ability to highlight some odd syntax and see its resulting AST; DrScheme already does this.

DSLs are especially tempting for data description: i.e. describing maps, models, animations, choreography, simulations, etc. without trying to fit into the language's form. Because this is common, it is ideal if the developer can also easily obtain these syntaxes and apply them in parsing and processing runtime inputs from users, files, or databases.


View edit of January 10, 2012 or FindPage with title or text search