Document the semantics of Atom's generic parameter

Provide a short example that shows why interning temporary atoms will
not blow up memory consumption.

Fixes #212

Author: federicomenaquintero

src/atom.rs

@@ -193,8 +193,46 @@ pub type DefaultAtom = Atom<EmptyStaticAtomSet>;
 
 /// Represents a string that has been interned.
 ///
-/// In reality this contains a complex packed datastructure and the methods to extract information
-/// from it, along with type information to tell the compiler which static set it corresponds to.
+/// While the type definition for `Atom` indicates that it generic on a particular
+/// implementation of an atom set, you don't need to worry about this.  Atoms can be static
+/// and come from a `StaticAtomSet` generated by the `string_cache_codegen` crate, or they
+/// can be dynamic and created by you on an `EmptyStaticAtomSet`.
+///
+/// `Atom` implements `Clone` but not `Copy`, since internally atoms are reference-counted;
+/// this means that you may need to `.clone()` an atom to keep copies to it in different
+/// places, or when passing it to a function that takes an `Atom` rather than an `&Atom`.
+///
+/// ## Creating an atom at runtime
+///
+/// If you use `string_cache_codegen` to generate a precomputed list of atoms, your code
+/// may then do something like read data from somewhere and extract tokens that need to be
+/// compared to the atoms.  In this case, you can use `Atom::from(&str)` or
+/// `Atom::from(String)`.  These create a reference-counted atom which will be
+/// automatically freed when all references to it are dropped.
+///
+/// This means that your application can safely have a loop which tokenizes data, creates
+/// atoms from the tokens, and compares the atoms to a predefined set of keywords, without
+/// running the risk of arbitrary memory consumption from creating large numbers of atoms —
+/// as long as your application does not store clones of the atoms it creates along the
+/// way.
+///
+/// For example, the following is safe and will not consume arbitrary amounts of memory:
+///
+/// ```ignore
+/// let untrusted_data = "large amounts of text ...";
+///
+/// for token in untrusted_data.split_whitespace() {
+///     let atom = Atom::from(token); // interns the string
+///
+///     if atom == Atom::from("keyword") {
+///         // handle that keyword
+///     } else if atom == Atom::from("another_keyword") {
+///         // handle that keyword
+///     } else {
+///         println!("unknown keyword");
+///     }
+/// } // atom is dropped here, so it is not kept around in memory
+/// ```
 pub struct Atom<Static: StaticAtomSet> {
     /// This field is public so that the `atom!()` macros can use it.
     /// You should not otherwise access this field.