1 files changed, 49 insertions, 0 deletions
diff --git a/src/id.rs b/src/id.rs
new file mode 100644
index 0000000..f630107
--- /dev/null
+++ b/src/id.rs
@@ -0,0 +1,49 @@
+use rand::{seq::SliceRandom, thread_rng};
+
+// Make IDs that:
+//
+// * Do not require escaping in URLs
+// * Do not require escaping in hostnames
+// * Are unique up to case conversion
+// * Are relatively unlikely to contain cursewords
+// * Are relatively unlikely to contain visually similar characters in most typefaces
+// * Are not sequential
+//
+// This leaves 23 ASCII characters, or about 4.52 bits of entropy per character
+// if generated with uniform probability.
+pub const ALPHABET: [char; 23] = [
+    '1', '2', '3', '4', '6', '7', '8', '9', 'b', 'c', 'd', 'f', 'h', 'j', 'k', 'n', 'p', 'r', 's',
+    't', 'w', 'x', 'y',
+];
+
+// Pick enough characters per ID to make accidental collisions "acceptably" unlikely
+// without also making them _too_ unwieldy. This gives a fraction under 68 bits per ID.
+pub const ID_SIZE: usize = 15;
+
+// Intended to be wrapped in a newtype that provides both type-based separation
+// from other identifier types, and a unique prefix to allow the intended type
+// of an ID to be determined by eyeball when debugging.
+//
+// By convention, the prefix should be UPPERCASE - note that the alphabet for this
+// is entirely lowercase.
+#[derive(Debug, Hash, PartialEq, Eq, sqlx::Type)]
+#[sqlx(transparent)]
+pub struct Id(String);
+
+impl Id {
+    pub fn generate<T>(prefix: &str) -> T
+    where
+        T: From<Self>,
+    {
+        let mut rng = thread_rng();
+        let id = prefix
+            .chars()
+            .chain(
+                (0..ID_SIZE)
+                    .flat_map(|_| ALPHABET.choose(&mut rng)) /* usize -> &char */
+                    .cloned(), /* &char -> char */
+            )
+            .collect::<String>();
+        T::from(Self(id))
+    }
+}