LLM approve #1

lib/list.tri

@@ -232,54 +232,100 @@ contains?_ self needle haystack =
     (startsWith? needle haystack)
 contains? = needle haystack : y contains?_ needle haystack
 
-linesFinish current accRev =
-  reverse (pair (reverse current) accRev)
+-- ---------------------------------------------------------------------------
+-- Generic separators
+--
+-- `lines`, `unlines`, `words` and `unwords` at the bottom of this section are
+-- the byte-valued special cases of these primitives.
+--
+-- Joining takes any separator; splitting takes one byte. Separators are removed
+-- rather than kept, and empty fields are preserved.
+--
+-- The workers below follow notes/tricu-normalization-rules.md: consumed data
+-- first, lazy eliminators around every recursive branch, `y` only inside the
+-- public wrapper, and `pair`-only state updates.
+-- ---------------------------------------------------------------------------
 
-lines_ self str accRev current =
-  matchList
-    (linesFinish current accRev)
+takeWhile_ self xs f =
+  lazyList
+    (_ : t)
     (h r :
-      matchBool
-        (self r (pair (reverse current) accRev) t)
-        (self r accRev (pair h current))
-        (equal? h 10))
-    str
-lines = str : y lines_ str t t
+      lazyBool
+        (_ : pair h (self r f))
+        (_ : t)
+        (f h))
+    xs
+takeWhile = f xs : y takeWhile_ xs f
 
-unlines_ self lines =
-  matchList
-    ""
-    (h r : append h (append "\n" (self r)))
-    lines
-unlines = lines : y unlines_ lines
-
-wordsAdd current accRev =
-  matchBool
-    accRev
-    (pair (reverse current) accRev)
-    (emptyList? current)
-
-words_ self str accRev current =
-  matchList
-    (reverse (wordsAdd current accRev))
+dropWhile_ self xs f =
+  lazyList
+    (_ : t)
     (h r :
-      matchBool
-        (self r (wordsAdd current accRev) t)
-        (self r accRev (pair h current))
-        (equal? h 32))
-    str
-words = str : y words_ str t t
+      lazyBool
+        (_ : self r f)
+        (_ : pair h r)
+        (f h))
+    xs
+dropWhile = f xs : y dropWhile_ xs f
 
-unwords_ self words =
-  matchList
-    ""
+-- Byte-level whitespace only: space and horizontal tab (HTTP OWS).
+spaceByte? = b : equal? b 32
+tabByte?   = b : equal? b 9
+trimByte?  = b : or? (spaceByte? b) (tabByte? b)
+
+trim = xs : dropWhile trimByte? (reverse (dropWhile trimByte? (reverse xs)))
+
+intercalate_ self xs sep =
+  lazyList
+    (_ : t)
     (h r :
-      matchBool
-        h
-        (append h (append " " (self r)))
+      lazyBool
+        (_ : h)
+        (_ : append h (append sep (self r sep)))
         (emptyList? r))
-    words
-unwords = words : y unwords_ words
+    xs
+intercalate = sep xs : y intercalate_ xs sep
+
+-- Separator after every field, including the last one. Line-oriented formats
+-- want this: `joinSuffix "\n" xs` terminates the final line while
+-- `intercalate "\n" xs` does not.
+joinSuffix_ self xs sep =
+  lazyList
+    (_ : t)
+    (h r : append (append h sep) (self r sep))
+    xs
+joinSuffix = sep xs : y joinSuffix_ xs sep
+
+-- Split on a single byte. A separator byte is never stored, so the state
+-- updates stay `pair`s and every recursive argument is a variable: the input is
+-- walked exactly once and the fields are reversed back once, when it ends.
+--
+-- Splitting on a multi-byte separator is deliberately not here. Detecting a
+-- separator longer than a byte means re-walking the remaining input at every
+-- split point (or splicing the field), which is quadratic in the best case and
+-- blew up when tried. `http.tri` wants CRLF and `:` splits; that wants a shape
+-- where the separator drives the recursion instead of the input.
+--
+-- Empty fields are preserved: `splitOnByte 58 "a::b"` is ["a" "" "b"].
+splitByte_ self str byte acc current =
+  lazyList
+    (_ : map reverse (reverse (pair current acc)))
+    (h r :
+      lazyBool
+        (_ : self r byte (pair current acc) t)
+        (_ : self r byte acc (pair h current))
+        (equal? h byte))
+    str
+splitOnByte = byte str : y splitByte_ str byte t t
+
+-- Every one of these keeps its arguments bound: partially applying a
+-- multi-argument function at the top level leaves a fixed point exposed.
+lines   = str : splitOnByte 10 str
+unlines = xs : joinSuffix "\n" xs
+
+-- Runs of separators collapse: empty fields are dropped.
+words   = str : filter (w : not? (emptyList? w)) (splitOnByte 32 str)
+unwords = xs : intercalate " " xs
 
 zipWith_ self f xs ys =
   matchList