diff options
Diffstat (limited to 'static/src')
-rw-r--r-- | static/src/_posts/2022-01-21-ginger-names.md | 173 |
1 files changed, 173 insertions, 0 deletions
diff --git a/static/src/_posts/2022-01-21-ginger-names.md b/static/src/_posts/2022-01-21-ginger-names.md new file mode 100644 index 0000000..81d90e5 --- /dev/null +++ b/static/src/_posts/2022-01-21-ginger-names.md @@ -0,0 +1,173 @@ +--- +title: >- + Ginger Names +description: >- + Thoughts about a fundamental data type. +tags: tech +series: ginger +--- + +The ginger language has, so far, 2 core types implemented: numbers and names. +Obviously there will be more coming later, but at this stage of development +these are all that's really needed. Numbers are pretty self explanatory, but +it's worth talking about names a bit. + +As they are currently defined, a name's only property is that it can either be +equal or not equal to another name. Syntactically they are encoded as being any +alphanumeric token starting with an alphabetic character. We might _think_ of +them as being strings, but names lack nearly all capabilities that strings have: +they cannot be iterated over, they cannot be concatenated, they cannot be split. +Names can only be compared for equality. + +## Utility + +The use-case for names is self-explanatory: they are words which identify +something from amongst a group. + +Consider your own name. It _might_ have an ostensible meaning. Mine, Brian, +means "high" (as in... like a hill, which is the possible root word). But when +people yell "Brian" across the room I'm in, they don't mean a hill. They mean +me, because that word is used to identify me from amongst others. The etymology +is essentially background information which doesn't matter. + +We use names all the time in programming, though we don't always call them that. +Variable names, package names, type names, function names, struct field names. +There's also keys which get used in hash maps, which are essentially names, as +well as enumerations. By defining name as a core type we can cover a lot of +ground. + +## Precedence + +This is not the first time a name has been used as a core type. Ruby has +symbols, which look like `:this`. Clojure has keywords, which also look like +`:this`, and it has symbols, which look like `this`. Erlang has atoms, which +don't have a prefix and so look like `this`. I can't imagine these are the only +examples. They are all called different things, but they're all essentially the +same thing: a runtime value which can only be compared for equality. + +I can't speak much about ruby, but I _can_ speak about clojure and erlang. + +Clojure is a LISP language, meaning the language itself is described using the +data types and structures built into the language. Ginger is also a LISP, though +it uses graphs instead of lists. + +Clojure keywords are generally used as keys to hash maps, sentinel values, and +enumerations. Besides keywords, clojure also makes use of symbols, which are +used for variable and library names. There seems to be some kind of split +ability on symbols, as they are expected to be separated on their periods when +importing, as in `clojure.java.io`. There's also a quoting mechanism in clojure, +where prefixing a symbol, or other value, with a single quote, like `'this`, +prevents it from being evaluated as a variable or function call. + +It's also possible to have something get quoted multiple layers deep, like +`'''this`. This can get confusing. + +Erlang is not a LISP language, but it does have atoms. These values are used in +the same way that clojure keywords are used. There is no need for a +corresponding symbol type like clojure has, since erlang is not a LISP and has +no real macro system. Atoms are sort of used like symbols, in that functions and +packages are identified by an atom, and so one can "call" an atom, like +`this()`, in order to evaluate it. + +## Just Names + +I don't really see the need for clojure's separation between keywords and +symbols. Symbols still need to be quoted in order to prevent evaluation either +way, so you end up with three different entities to juggle (keywords, symbols, +and symbols which won't be evaluated). Erlang's solution is simpler, atoms are +just atoms, and since evaluation is explicit there's no need for quoting. Ginger +names are like erlang atoms in that they are the only tool at hand. + +The approaches of erlang vs clojure could be reframed as explicit vs implicit +evaluation of operations calls. + +In ginger evaluation is currently done implicitly, but in only two cases: + +* A value on an edge is evaluated to the first value which is a graph (which + then gets interpreted as an operation). + +* A leaf vertex with a name value is evaluated to the first value which is not a + name. + +In all other cases, the value is left as-is. A graph does not need to be quoted, +since the need to evaluate a graph as an operation is already based on its +placement as an edge or not. So the only case left where quoting is needed (if +implicit evaluation continues to be used) is a name on a leaf vertex, as in the +example before. + +As an example to explore explicit vs implicit quoting in ginger, if we want to +programatically call the `AddValueIn` method on a graph, which terminates an +open edge into a value, and that value is a name, it might look like this with +implicit evaluation (the clojure-like example): + +``` +out = addValueIn < (g (quote < someName;) someValue; ); + +* or, to borrow the clojure syntax, where single quote is a shortcut: + +out = addValueIn < (g; 'someName; someValue; ); +``` + +In an explicit evaluation language, which ginger so far has not been and so this +will look weird, we might end up with something like this: + +``` +out = addValueIn < (eval < g; someName; eval < someValue; ); + +* with $ as sugar for the `eval`, like ' is a shortcut for `quote` in clojure:` + +out = addValueIn < ($g; someName; $someValue; ); +``` + +I don't _like_ either pattern, and since it's such a specific case I feel like +something less obtrusive could come up. So no decisions here yet. + +## Uniqueness + +There's another idea I haven't really gotten to the bottom of yet. The idea is +that a name, _maybe_, shouldn't be considered equal to the same name unless they +belong to the same graph. + +For example: + +``` +otherFoo = { out = 'foo } < (); + +out = equal < ('foo; otherFoo; ); +``` + +This would output false. `otherFoo`'s value is the name `foo`, and the value +it's being compared to is also a name `foo`, but they are from different graphs +and so are not equal. In essence, names are automatically namespaces. + +This idea only really makes sense in the context of packages, where a user +(a developer) wants to import functionality from somewhere else and use it +in their program. The code package which is imported will likely use name +values internally to implement its functionality, but it shouldn't need to worry +about naming conflicts with values passed in by the user. While it's possible to +avoid conflicts if a package is designed conscientiously, it's also easy to mess +up if one isn't careful. This becomes especially true when combining +functionality of packages with overlapping functionality, where the data +returned from one might looks _similar_ to that used by the other, but it's not +necessarily true. + +On the other hand, this could create some real headaches for the developer, as +they chase down errors which are caused because one `foo` isn't actually the +same as another `foo`. + +What it really comes down to is the mechanism which packages use to function as +packages. Forced namespaces will require packages to export all names which they +expect the user to need to work with the package. So the ergonomics of that +exporting, both on the user's and package's side, are really important in order +to make this bearable. + +So it's hard to make any progress on determining if this idea is gonna work +until the details of packaging are worked out. But for this idea to work the +packaging is going to need to be designed with it in mind. It's a bit of a +puzzle, and one that I'm going to marinate on longer, in addition to the quoting +of names. + +And that's names, their current behavior and possible future behavior. Keep an +eye out for more ginger posts in.... many months, because I'm going to go work +on other things for a while (I say, with a post from a month ago having ended +with the same sentiment). |