Uncaught AssertionError from CopybookParser.parseTree on copybooks with duplicated clauses (8 crash paths, JVM -ea independent)

オープン
#878 コメント 1 件 リアクション 0 件 担当者 0 名 GitHub で見る

まだ誰も着手していません。

評価

難易度
4/5
見積もり時間
3〜5日
初心者へのやさしさ
54/100
issue の種類
バグ
明瞭さ
明確に書かれている
活発さ
活発
技術スタック
scala, spark
領域
backend, data

調査の方向性

Start with ParserVisitor.scala at visitGroup and visitPrimitive (the issue identifies lines 506–509 and 796–800), then inspect the grammar rules that admit repeated clauses. Add parser tests for the listed duplicate clauses and for the duplicated SIGN control case. Done means malformed copybooks produce SyntaxErrorException consistently with and without JVM assertions enabled.

索引モデルが issue の本文から書いたものです。

説明

bug

Describe the bug

CopybookParser.parseTree(...) — the public parser entry point used by the spark-cobol data source — throws an uncaught java.lang.AssertionError when a copybook contains a duplicated clause on the same data entry (e.g. two PICTURE, two OCCURS, two REDEFINES, duplicate VALUE).

  • Class: CWE-617 (Reachable Assertion). Availability impact only (no memory-safety).
  • Affected: za.co.absa.cobrix:cobol-parser_2.13:2.11.2 (Maven Central latest, 2026-10-12); the same asserts are still on master @ 2026-10-23 at the same lines.
  • Always-on: the crash is not a debug-only invariant failure — it fires with the default JVM flags (-da) too.

Three factors combine:

  1. The ANTLR grammar admits any number of these clauses (( ... )* repetition), so the invalid input passes syntax validation and reaches the visitor.
  2. The visitor enforces its "at most one / exactly one" invariants only with scala.Predef.assert (ParserVisitor.scala L506–509 visitGroup; L796–800 visitPrimitive). No semantic check rejects duplicates with the library's own SyntaxErrorException.
  3. scala.Predef.assert is @elidable(ASSERTION) and the published artifact has the calls compiled in, so they fire regardless of -ea/-da. Nothing catches between the visitor and the public API, so the Error escapes parseTree(...).

Confirmed crash paths (all deterministic, identical with and without -ea — 8 crash paths × 2 modes = 16/16; row 5 is the control, see below):

# data entry duplicated clause result failure location
1 primitive PIC X(2) PIC X(5) AssertionError visitPrimitive :800
2 primitive COMP COMP-3 (USAGE ×2) AssertionError visitPrimitive :797
3 primitive OCCURS 2 TIMES OCCURS 3 TIMES AssertionError visitPrimitive :798
4 primitive REDEFINES FOO REDEFINES BAR AssertionError visitPrimitive :796
5 primitive SIGN LEADING SEPARATE SIGN TRAILING SEPARATE SyntaxErrorException (control — grammar rejects, every tested flavor) — (:799 assert unreachable)
6 group REDEFINES A1 REDEFINES A2 AssertionError visitGroup :506
7 group COMP-3 PACKED-DECIMAL (usageGroup ×2) AssertionError visitGroup :507
8 group OCCURS 2 TIMES OCCURS 3 TIMES AssertionError visitGroup :508
9 group VALUE 'A' VALUE 'B' AssertionError visitGroup :509

Row 5 is the control: duplicated SIGN clauses are rejected with the library's intended error
type (SyntaxErrorException) in every flavor tested — adjacent, with SEPARATE CHARACTER,
reversed order, SIGN IS form, with another clause in between, and triple — in both -ea
and default JVM modes. It shows the error model that the other 8 paths bypass, and makes
the :799 (ctx.separateSign.size() < 2) assert unreachable from the accepted language —
that dead assert can be fixed or dropped together with the others.

Escape path / stack:

java.lang.AssertionError: assertion failed
	at scala.Predef$.assert(Predef.scala:264)
	at za.co.absa.cobrix.cobol.parser.antlr.ParserVisitor.visitPrimitive(ParserVisitor.scala:800)
	at za.co.absa.cobrix.cobol.parser.antlr.copybookParser$PrimitiveContext.accept(copybookParser.java:3559)
	...
	at za.co.absa.cobrix.cobol.parser.antlr.ANTLRParser$.parse(ANTLRParser.scala:88)
	at za.co.absa.cobrix.cobol.parser.CopybookParser$.parseTree(CopybookParser.scala:336)

Found by coverage-guided fuzzing (Jazzer 0.30 around parseTree) and minimized. A plain java.lang.Error aborts the calling Spark job, and re-reading the same input re-crashes deterministically (no retry tolerance).

Code snippet that caused the issue

import za.co.absa.cobrix.cobol.parser.CopybookParser

// duplicated PICTURE clause on one data entry (row 1)
val copybook =
  """      01  REC1.
    |          05  MES   PIC X(2) PIC X(5).
    |""".stripMargin

CopybookParser.parseTree(copybook)
// java.lang.AssertionError: assertion failed
//   at za.co.absa.cobrix.cobol.parser.antlr.ParserVisitor.visitPrimitive(ParserVisitor.scala:800)

Reachable the same way from Spark:

spark.read
  .format("cobol")
  .option("copybook_contents", copybook) // copybook with a duplicated clause
  .load(path)

The other 7 crash cases from the table (rows 2–4 and 6–9; each is a single duplicated clause) all reproduce identically — e.g. 05 F PIC X COMP COMP-3., 05 G PIC 9 OCCURS 2 TIMES OCCURS 3 TIMES., 05 A REDEFINES X REDEFINES Y., group-level REDEFINES/usage/OCCURS/VALUE duplicates, etc. A standalone Java probe covering all 9 cases was used (runs only against the published Maven artifacts, no rebuild).

Expected behavior

A malformed copybook should be rejected with the library's own za.co.absa.cobrix.cobol.parser.exceptions.SyntaxErrorException (as already happens for genuine syntax errors via the ANTLR error listener, and for semantic errors such as in checkBounds), with a clear message like "duplicate clause X on data entry Y" — not an uncaught java.lang.AssertionError.

The behaviour must be identical with and without -ea.

Context

  • Cobrix version: za.co.absa.cobrix:cobol-parser_2.13:2.11.2 (Maven Central latest, published 2026-10-12); same issue on master @ 2026-10-23
  • Spark version: n/a for the standalone repro (parser-level); also reachable via spark-cobol
  • Scala version: 2.13.18 (scala-library; artifact built for 2.13)
  • Operating system: Linux x86_64 (OpenJDK 21.0.13 Temurin)

CVSS v3.1 (suggested, copybook from a not-fully-trusted feed): AV:N/AC:L/PR:L/UI:N/S:U/C:N/I:N/A:H → 6.5 (Medium).

Copybook (if possible)

(rows 6–9: each line is its own one-entry copybook; the error fires on the line the probe focuses on — shown together for compactness)
Primary crashing copybook (realistic multi-field layout; note 10 MES carries two PICTURE clauses):

      01  MESSAGE.
          05  MESCAGE-TOP.
              10  MESSAGE-TYPE            PIC X(2).
              10  MES       PIC X         PIC X(1).
              10  MESSAGE-DATA            PIC X(10).

Minimal header-only group variants (rows 6–9):

      01  R.
          05  A   REDEFINES X REDEFINES Y.
          05  B   COMP-3 PACKED-DECIMAL.
          05  C   OCCURS 2 TIMES OCCURS 3 TIMES.
          05  D   VALUE 'A' VALUE 'B'.

Attachments: standalone probe PocAssertFamily.java (9 cases, prints crash location per case), full probe output for -ea and -da (identical for the 8 crash cases, 16/16), and the fuzzer crash input + Java reproducer. Data file is not needed to reproduce (the crash is parse-level and data-independent), but a small one can be attached if useful.

poc_results.txt
PocAssertFamily.java
Crash_cc10fa3d5bfb9296bd39fee83dd200380925f302.java
crash-cc10fa3d5bfb9296bd39fee83dd200380925f302.txt

主要言語
Scala
スター
170
フォーク
96
平均マージ
7日 5時間
マージ済み PR(30日)
4

環境構築

このプロジェクトには開発コンテナ、Dockerfile、コントリビューションガイドがありません。まず README を読み、一般的な手順ははじめてのコントリビューションガイドを参照してください。

はじめの一歩

  1. issue を最後まで読み、次にプロジェクトのコントリビューションガイドを読みます。
  2. 着手することを issue にコメントします — 二人が同じ作業をするのを防げます。
  3. リポジトリをフォークし、ブランチを切って変更します。
  4. issue 番号を参照したプルリクエストを送ります。

AbsaOSS/cobrix のほかの issue

AbsaOSS/cobrix の issue をすべて見る

似ている issue

Scala の issue をもっと見る

新しい issue をメールで受け取る

初心者向けの GitHub issue を短くまとめたダイジェスト。