OutputStreamMonitor.getContents() inserts U+FFFD when a multibyte character is split across two reads

未关闭
#3,058 2 条评论 0 个 reaction 已指派 0 人 在 GitHub 查看

维护者通常 1 天内回复

@dimitribekale 已经在做这个了。

开始于 2026年11月5日。

  • #1 来自 @dimitribekale —— 已关闭,未合并
  • #3060 来自 @dimitribekale —— 未关闭

评估

难度
4/5
预计耗时
1-2 天
新手友好度
25/100
Issue 类型
缺陷
描述清晰度
描述清楚
活跃度
停滞
技术栈
java
领域
backend

调研方向

Start with debug/org.eclipse.debug.core/core/org/eclipse/debug/internal/core/OutputStreamMonitor.java, especially getContents(), flushContents(), and setBuffered(), and read the related OutputStreamMonitorTests coverage. Reproduce the split UTF-8 read described in the issue and check how decoding behaves when getContents() is called after each read. Done means the returned text is o and then oöO without an intervening flush, while the existing flush test still passes.

由索引模型根据 Issue 内容生成。

描述

bug

org.eclipse.debug.internal.core.OutputStreamMonitor.getContents() returns corrupted text when a
multibyte character arrives split across two reads of the process stream and getContents() is
called after each read without flushContents() in between.

Cause. getContents() (master c49150d04204, 2026-11-04, lines 191–202) decodes the whole accumulated byte
buffer with fBufferedDecoder, a stateful StreamDecoder that keeps the bytes of an incomplete
trailing character between calls. The read thread clears fCachedDecodedContents on every append,
so the next call re-feeds bytes the decoder has already consumed. The retained lead byte is then
followed by the first byte of the buffer again, which the decoder reports as malformed input (U+FFFD).

Reproduce. UTF-8 monitor, input "oöO" (6F C3 B6 4F), written as 6F C3 then B6 4F,
getContents() after each write (the attached Repro.java does this with a PipedOutputStream and
the same subclass trick as OutputStreamMonitorTests):

  • Expected: "o", then "oöO"
  • Actual: "o", then "�oöO"

The existing testBufferedOutputStreamMonitor uses this input but flushes between the reads, so it
does not catch the case.

Impact. IStreamMonitor.getContents() is API and monitors are buffered from construction.
ProcessConsole reads the contents once and then unbuffers, so the Console view is not affected;
clients that poll a buffered monitor while the process is writing get wrong text. Minor.

Proposed fix. Decode only the bytes not yet fed to the decoder and accumulate the result, so
nothing is re-fed; flushContents() and setBuffered() reset the accumulated text and the fed-byte
count and keep the decoder state, as today. Verified fail-then-pass against master compiled
standalone, and the flush sequence of testBufferedOutputStreamMonitor keeps its current result.
A regression test in the style of that test is included below. Happy to open a PR if this is
confirmed.

Patch
--- a/debug/org.eclipse.debug.core/core/org/eclipse/debug/internal/core/OutputStreamMonitor.java
+++ b/debug/org.eclipse.debug.core/core/org/eclipse/debug/internal/core/OutputStreamMonitor.java
@@ -70,7 +70,9 @@
 	 * of an incomplete character.
 	 */
 	private StreamDecoder fBufferedDecoder;
+	/** Decoded form of the first {@link #fDecodedLength} bytes of {@link #fContents}. */
 	private String fCachedDecodedContents;
+	private int fDecodedLength;
 
 	/**
 	 * The thread which reads from the stream
@@ -193,11 +195,12 @@
 		if (!isBuffered()) {
 			return ""; //$NON-NLS-1$
 		}
-		if (fCachedDecodedContents != null) {
-			return fCachedDecodedContents;
-		}
 		byte[] data = getData();
-		fCachedDecodedContents = fBufferedDecoder.decode(data, 0, data.length);
+		if (fDecodedLength < data.length) {
+			// decode only the bytes not yet fed to the stateful decoder
+			fCachedDecodedContents += fBufferedDecoder.decode(data, fDecodedLength, data.length - fDecodedLength);
+			fDecodedLength = data.length;
+		}
 		return fCachedDecodedContents;
 	}
 
@@ -236,7 +239,6 @@
 					if (read > 0) {
 						synchronized (this) {
 							if (isBuffered()) {
-								fCachedDecodedContents = null;
 								fContents.write(buffer, 0, read);
 							}
 							fireStreamAppended(buffer, 0, read);
@@ -306,7 +308,8 @@
 	@Override
 	public synchronized void setBuffered(boolean buffer) {
 		if (isBuffered() != buffer) {
-			fCachedDecodedContents = null;
+			fCachedDecodedContents = ""; //$NON-NLS-1$
+			fDecodedLength = 0;
 			if (buffer) {
 				fContents = new ByteArrayOutputStream();
 				fBufferedDecoder = new StreamDecoder(fCharset == null ? Charset.defaultCharset() : fCharset);
@@ -320,7 +323,8 @@
 	@Override
 	public synchronized void flushContents() {
 		if (isBuffered()) {
-			fCachedDecodedContents = null;
+			fCachedDecodedContents = ""; //$NON-NLS-1$
+			fDecodedLength = 0;
 			fContents.reset();
 		}
 	}
Regression test for OutputStreamMonitorTests
/**
 * A multibyte character split across two reads must not be corrupted when
 * getContents() is called after each read without flushing in between.
 */
@Test
public void testBufferedContentsWithSplitCharacterAndNoFlush() throws Exception {
	String input = "oöO";
	byte[] byteInput = input.getBytes(StandardCharsets.UTF_8);
	BinaryListener binaryListener = new BinaryListener();
	monitor.addBinaryListener(binaryListener);

	sysout.write(byteInput, 0, 2);
	sysout.flush();
	monitor.startMonitoring();
	binaryListener.waitForBytes(2);
	assertThat(monitor.getContents()).isEqualTo(input.substring(0, 1));

	sysout.write(byteInput, 2, byteInput.length - 2);
	sysout.flush();
	binaryListener.waitForBytes(byteInput.length);
	assertThat(monitor.getContents()).isEqualTo(input);
	binaryListener.assertNoExceptions();
}
Standalone reproducer (Repro.java)

Compile OutputStreamMonitor.java and StreamDecoder.java from master together with this file
against the Eclipse SDK plug-ins (JDK 17+), run Repro; exit code 1 on the current code, 0 with
the patch.

import java.io.PipedInputStream;
import java.io.PipedOutputStream;
import java.nio.charset.StandardCharsets;

import org.eclipse.debug.core.IBinaryStreamListener;
import org.eclipse.debug.internal.core.OutputStreamMonitor;

public class Repro {
	static class TestOutputStreamMonitor extends OutputStreamMonitor {
		TestOutputStreamMonitor(java.io.InputStream in) { super(in, StandardCharsets.UTF_8); }
		public void startMonitoring() { super.startMonitoring("repro"); }
	}

	public static void main(String[] args) throws Exception {
		String input = "oöO";
		byte[] bytes = input.getBytes(StandardCharsets.UTF_8);   // 6F C3 B6 4F
		PipedOutputStream sysout = new PipedOutputStream();
		PipedInputStream in = new PipedInputStream(sysout);
		TestOutputStreamMonitor monitor = new TestOutputStreamMonitor(in);
		int[] received = {0};
		monitor.addBinaryListener((IBinaryStreamListener) (data, m) -> { synchronized (received) { received[0] += data.length; received.notifyAll(); } });

		sysout.write(bytes, 0, 2);                        // 'o' + first byte of 'ö'
		sysout.flush();
		monitor.startMonitoring();
		waitFor(received, 2);
		String first = monitor.getContents();             // "o", decoder keeps 0xC3

		sysout.write(bytes, 2, 2);                        // second byte of 'ö' + 'O', no flushContents in between
		sysout.flush();
		waitFor(received, 4);
		String second = monitor.getContents();
		sysout.close();

		System.out.println("first  getContents(): " + show(first));
		System.out.println("second getContents(): " + show(second) + "   expected " + show(input));
		System.exit(second.equals(input) ? 0 : 1);
	}

	static void waitFor(int[] c, int n) throws InterruptedException {
		synchronized (c) { long end = System.currentTimeMillis() + 10000; while (c[0] < n && System.currentTimeMillis() < end) c.wait(100); }
	}

	static String show(String s) {
		StringBuilder b = new StringBuilder("\"");
		for (char ch : s.toCharArray()) b.append(ch < 128 ? String.valueOf(ch) : String.format("<U+%04X>", (int) ch));
		return b.append('"').toString();
	}
}

Provenance. This defect was found during an academic study of LLM-assisted code review at Jeonbuk National University (Jeonju, South Korea). The initial analysis was produced by an automated LLM-based reviewer and then manually verified by the authors: we confirmed the defect mechanism by reading the source at HEAD, and the reproducer above compiles and exhibits the failure against the real code. The reviewer's own patch (a fresh decoder per call) fixed this case but broke the flush sequence that the existing testBufferedOutputStreamMonitor asserts, so the patch proposed here is the authors' correction. We are disclosing this origin in line with project policies on AI-assisted contributions. We report it because we believe it is a genuine defect, and we will respect the project's triage either way; a "not a bug" outcome is valuable to us as a correction.

主要语言
Java
星标
165
派生
174
平均合并
4 天 56 分钟
30 天内合并 PR
36

环境准备

从这里开始

  1. 先读完整个 Issue,再读项目的贡献指南。
  2. 在 Issue 下留言说明你要接手 —— 这能避免两个人做同样的事。
  3. Fork 仓库,在一个分支上完成修改。
  4. 提交 Pull Request,并在描述里引用这个 Issue 编号。

eclipse-platform/eclipse.platform 的其他 Issue

查看 eclipse-platform/eclipse.platform 的全部 Issue

相似的 Issue

更多 Java Issue

把新 issue 发到你的邮箱

精选适合新手参与的 GitHub issue 摘要。