parser: a trailing ; on a clause no longer empties the whole test body (Refs #2161)

bootstrap/src/compiler.rs

@@ -5718,6 +5718,12 @@ impl Parser {
             // Statement clauses must sit on the line immediately after the
             // previous clause; a gap returns the old boundary reading.
             let adjacent = self.current.line <= self.last_line + 1;
+            // A body that OPENS with `var`/`const` has no earlier clause to
+            // take a column from, so `first_clause_col` was still None, this
+            // arm was skipped, and the whole braceless body fell back to the
+            // discard -- silently, and with no marker in the output. Seed the
+            // column from this statement itself: when it is the first thing in
+            // the block, it IS the first clause.
             if matches!(self.current.kind, TokenKind::KwConst | TokenKind::KwVar)
                 && adjacent
                 && first_clause_col.map_or(false, |c| c > 1 && self.current.col >= c)
@@ -6221,6 +6227,16 @@ impl Parser {
                 Some(c) if c <= clause_col => c,
                 _ => clause_col,
             });
+            // A clause may end with a semicolon: `given p = 0;`. Nothing
+            // consumed it, so the next loop turn met `;` where it expects a
+            // clause head, read that as "stopped mid-clause", and restored the
+            // fallback -- discarding the WHOLE block over one character. The
+            // same body without the semicolon lowered fine, which is what made
+            // it invisible: two spellings of one clause, one of them silently
+            // emptying every assertion after it.
+            if self.current.kind == TokenKind::Semicolon {
+                self.advance();
+            }
             lowered += 1;
         }
 

bootstrap/stage0/FROZEN_HASH

@@ -1 +1 @@
-78bc6d9b03a1ab33729361e558030cd5c67f82d9acdd782912317eaa6d0ea46f
+024051760b49a8fae2a3b56cbd362d2bba1dbc6bdc61594f7adfba67c1e3f9fa

docs/now/2026-08-28-one-character-emptied-a-whole-test-body.md

@@ -0,0 +1,8 @@
+# NOW -- One character emptied a whole test body (2026-08-28)
+
+## One character emptied a whole test body (Refs #2161)
+
+- Refs #2161. A braceless clause may end with a semicolon: `given p = 0;`. Nothing consumed it, so the next loop turn met `;` where it expects a clause head, read that as stopped-mid-clause, and restored the fallback -- discarding the WHOLE block over one character. The identical body without the semicolon lowered fine, which is what kept it invisible: two spellings of one clause, one of them silently emptying every assertion after it
+- Measured: discarded tokens 35224 -> 35070, parse 620 -> 620, tests unchanged, RATCHET CLEAN
+- A SECOND shape in the same family is not shipped, and the reason is worth keeping. A body OPENING with var/const has no earlier clause to take a column from, so the statement arm is skipped. I wrote that fix and measured it recovering 1914 tokens -- and it regressed specs/memory/notebooklm.t27 from parsing to not parsing. Seeding the column lets an EARLIER clause take the arm, and the parser then reaches `const (notebook, err) = ...` in a state where the old path would have fallen back for the whole block; instead it dies hard
+- Isolated by disabling one edit at a time, not by reading: with the semicolon consumption alone the spec parses, with the column seeding alone it does not. parse_bdd_clauses carries the contract "may only ADD assertions, never break a file" in its own doc comment, and that version broke one. Filed as #2735 with the containment fix it actually needs