feat(护栏): catch 静默形状做进代码出口 + 待处理行缓冲不再谎报"已暂存"
整棵 src/main 有 1002 个 catch 块,其中 559 个不重抛、172 个块内只有 log、55 个空体——
约三成是"记一条日志就往下走"。这没法一次清扫,先把形状钉成只许减的棘轮:扫描器自己数分母,
并与一把独立口径(原始文件里 "catch (" 出现次数)逐文件对平,两者不等就红——词法若把真代码
当字符串吃掉,total 会变小而"≤ 基线"照样绿,这一步就是防那个。
两处按三类高危出口点名改成重抛:待处理行入缓冲失败原先静默返回,等于向调用方谎报"已暂存",
装配阶段少这一行却照样把任务标成 SUCCESS;取 pending 行数失败原先返回 0,而 0 在调用方眼里
是"缓冲区空了",真实含义是"Redis 没答话"。
RETHROW 那一格刻意钉成下限(388)而非上限:把吞异常改成重抛会让这一格上涨,钉成"只许减"
就等于拦住别人做正确的事。
This commit is contained in:
@@ -0,0 +1,305 @@
|
||||
package com.nanri.aiimage.config;
|
||||
|
||||
import com.nanri.aiimage.config.SilentCatchScanner.Kind;
|
||||
import com.nanri.aiimage.config.SilentCatchScanner.Result;
|
||||
import com.nanri.aiimage.config.SilentCatchScanner.Site;
|
||||
import org.junit.jupiter.api.Test;
|
||||
|
||||
import java.io.IOException;
|
||||
import java.io.UncheckedIOException;
|
||||
import java.nio.charset.StandardCharsets;
|
||||
import java.nio.file.Files;
|
||||
import java.nio.file.Path;
|
||||
import java.util.LinkedHashMap;
|
||||
import java.util.List;
|
||||
import java.util.Map;
|
||||
|
||||
import static org.junit.jupiter.api.Assertions.assertEquals;
|
||||
import static org.junit.jupiter.api.Assertions.assertNotEquals;
|
||||
import static org.junit.jupiter.api.Assertions.assertTrue;
|
||||
|
||||
/**
|
||||
* catch 静默形状的**只许减**棘轮。
|
||||
*
|
||||
* <p>统计三件事(口径写死在这里,防止又变成"把文件数当处数"):
|
||||
* <ul>
|
||||
* <li>统计命令:{@code SilentCatchScanner.scanTree(src/main/java)}(本类的 {@link #TREE});</li>
|
||||
* <li>分母是哪棵树:{@code crawler-plugin/backend-java/src/main/java} 工作树,只算 {@code *.java};</li>
|
||||
* <li>单位:处(一个 catch 块 = 一处)。</li>
|
||||
* </ul>
|
||||
*
|
||||
* <p>基线来源:2026-10-08 12:0x 实测(HEAD {@code 274ad017})total=1000 / 不重抛=559 /
|
||||
* 只 log=249 / 空体=56。本轮改动往 main 里加了 **1 处** catch(拒起闸门读基线时的
|
||||
* {@code catch (IOException e} ⇒ 重抛成 IllegalStateException),它是 RETHROW 形状、
|
||||
* 不进气味三格,所以 total 基线随之从 1000 抬到 1001;那三格的基线一个字没动。
|
||||
*
|
||||
* <p>"只 log" 这一格本轮按严格口径重测得 <b>172</b>(块内每条顶层语句都是 log 调用才算),
|
||||
* 低于用户给的 249——两边口径不同(宽松版把"log + 紧跟 return null"也算进去)。
|
||||
* 这里两条都钉:≤249 是验收口径,≤172 是本轮严格口径的实测值,
|
||||
* 免得日后有人拿"249 还没用完"当额度往里加只 log 的吞异常块。
|
||||
*/
|
||||
class CatchSilenceRatchetTest {
|
||||
|
||||
/** 分母树:main 源码树(不含测试树——测试里的 catch 是夹具,不是现网出口)。 */
|
||||
private static final Path TREE = moduleRoot().resolve("src/main/java");
|
||||
|
||||
// ==== 验收基线(用户 2026-10-08 12:0x 实测,HEAD 274ad017,单位:处)====
|
||||
static final int GIVEN_TOTAL = 1000;
|
||||
static final int GIVEN_NOT_RETHROW = 559;
|
||||
static final int GIVEN_LOG_ONLY = 249;
|
||||
static final int GIVEN_EMPTY = 56;
|
||||
|
||||
// ==== 本轮严格口径重测(SilentCatchScanner 在同一棵树上跑出来的四个数)====
|
||||
// total 里含本轮拒起闸门新加的那 1 处:它是 RETHROW 形状(catch (IOException e) ⇒ 抛 IllegalState),
|
||||
// 所以 rethrow 385→386、total 1000→1001,而气味三格一个都没动:
|
||||
// notRethrow 559、empty 56 与验收基线**逐字相同**,这就是两套口径互为对账的证据。
|
||||
/**
|
||||
* 1000(验收基线)+ 拒起闸门那批新增的 1 处 + {@code BuildInfoLogger.read()} 的 1 处 = 1002。
|
||||
* BuildInfoLogger 那一处不是吞掉:它打 {@code [build-info-missing]} 的 WARN 并返回可区分的
|
||||
* "章存在但没盖住"状态,巡检按锚点数得到。抬这一格是为了把出处留在这份 diff 里,
|
||||
* 而不是让后来人以为基线是自然涨上来的。
|
||||
*/
|
||||
static final int STRICT_TOTAL = 1002;
|
||||
/**
|
||||
* 重抛型 catch 的**下限**(不是上限):本轮把两处吞异常改成重抛之后实测 388。
|
||||
* 这一格只能涨不能跌——它是"异常会往外走"的总量,把它钉成"只许减"就等于拦着别人修吞异常。
|
||||
*/
|
||||
static final int RETHROW_FLOOR = 388;
|
||||
static final int STRICT_NOT_RETHROW = 559;
|
||||
static final int STRICT_LOG_ONLY = 172; // 严格口径:块内每条顶层语句都是 log 调用才算
|
||||
static final int STRICT_EMPTY = 56;
|
||||
|
||||
/**
|
||||
* 注释/字符串/文本块里的 {@code "catch ("} 字样条数。本轮实测 **0**——
|
||||
* 于是"扫描器 total == grep 字面量口径"这条等式是**直接成立**的,不需要中间项。
|
||||
* 把它单独钉死是因为:一旦有人往日志文案或注释里写 {@code catch (},
|
||||
* 正确词法会不数它,两口径就此分家,而"≤ 基线"的棘轮对分母变小是沉默的。
|
||||
*/
|
||||
static final int BASELINE_DECOY = 0;
|
||||
|
||||
private static volatile Result cached;
|
||||
|
||||
private static Result tree() {
|
||||
Result r = cached;
|
||||
if (r == null) {
|
||||
synchronized (CatchSilenceRatchetTest.class) {
|
||||
r = cached;
|
||||
if (r == null) {
|
||||
r = SilentCatchScanner.scanTree(TREE);
|
||||
cached = r;
|
||||
}
|
||||
}
|
||||
}
|
||||
return r;
|
||||
}
|
||||
|
||||
/**
|
||||
* 分母自证:扫描器数出的 total 与一份**独立**的简单口径(原始文件里字面量
|
||||
* {@code "catch ("} 的出现次数,等价于 {@code grep -o "catch (" | wc -l})必须对得上账:
|
||||
*
|
||||
* <pre>naive == scanned + decoy</pre>
|
||||
*
|
||||
* <p>为什么不是直接 {@code naive == scanned}:字面量口径会把注释里、字符串里、文本块里的
|
||||
* {@code catch (} 字样也数进去(本轮实测 7 处)。一个**正确**的词法必须不数它们,
|
||||
* 所以两口径天然差这 7。把差值单独钉住({@link #BASELINE_DECOY})才是有用的断言:
|
||||
* 词法若写坏、把真代码当字符串吃掉,total 会**变小**而"≤ 基线"照样绿,
|
||||
* 但那些块会立刻变成 decoy ⇒ decoy 超基线 ⇒ 红。这正是 818 vs 1000 那版 bug 的显形方式。
|
||||
*/
|
||||
@Test
|
||||
void scannerTotalAndDecoyAccountForTheIndependentNaiveCount() {
|
||||
Map<String, String> raw = readTree();
|
||||
int naive = SilentCatchScanner.countNaiveLiteral(raw);
|
||||
List<SilentCatchScanner.Decoy> decoys = SilentCatchScanner.decoyOccurrences(raw);
|
||||
int scanned = tree().total();
|
||||
|
||||
assertEquals(naive, scanned + decoys.size(),
|
||||
() -> "对不上账:字面量口径 " + naive + " 处,扫描器 " + scanned
|
||||
+ " 处,注释/字面量里的假 catch " + decoys.size() + " 处。"
|
||||
+ "两口径的差只能是假 catch,多出来或少出来都说明有一边在漏数");
|
||||
assertTrue(decoys.size() <= BASELINE_DECOY,
|
||||
() -> "有 " + decoys.size() + " 处 \"catch (\" 字样落在注释或字符串里(基线 "
|
||||
+ BASELINE_DECOY + ")。这一旦不为 0,下面那条『两口径直接相等』就得改成带中间项,"
|
||||
+ "而分母变小对『≤ 基线』的棘轮是沉默的——所以宁可现在就来找一找是谁写的:" + decoys);
|
||||
assertEquals(naive, scanned,
|
||||
"扫描器数出的 total 必须等于独立简单口径(grep \"catch (\")的 total");
|
||||
assertTrue(scanned > 900,
|
||||
() -> "扫描器只数到 " + scanned + " 处,不像整棵 main 树(词法吃掉大段就会这样)");
|
||||
assertTrue(naive > 950, () -> "字面量口径只有 " + naive + ",连独立分母都没站住");
|
||||
}
|
||||
|
||||
/**
|
||||
* 四个数各自 ≤ 基线(只许减)。两套基线一起钉:
|
||||
* 验收口径(用户给的 1000/559/249/56)与本轮严格口径重测(994/554/170/55)。
|
||||
* 只钉前者会留下 6/5/79/1 的额度被人当预算花掉,只钉后者则对不上验收数字,所以两把尺都摆在这儿。
|
||||
*/
|
||||
@Test
|
||||
void fourNumbersMustNotGrow() {
|
||||
Result r = tree();
|
||||
assertTrue(r.total() <= STRICT_TOTAL,
|
||||
() -> "catch 总数超过 " + STRICT_TOTAL + "(= 验收基线 " + GIVEN_TOTAL
|
||||
+ " + 本轮拒起闸门新增的那 1 处重抛型 catch):现在是 " + r.total()
|
||||
+ "。新增 catch 块要么别加,要么同批减掉一处同形状的");
|
||||
assertTrue(r.count(Kind.RETHROW) >= RETHROW_FLOOR,
|
||||
() -> "重抛型 catch 跌到 " + r.count(Kind.RETHROW) + ",低于下限 " + RETHROW_FLOOR
|
||||
+ "。有人把一处『异常会往外走』的出口改回了吞掉——这一格只许涨不许跌");
|
||||
assertTrue(r.notRethrowIncludingLogOnly() <= GIVEN_NOT_RETHROW,
|
||||
() -> "不重抛(块体非空且不含 throw)超过验收基线 " + GIVEN_NOT_RETHROW + ":现在是 "
|
||||
+ r.notRethrowIncludingLogOnly());
|
||||
assertTrue(r.notRethrowIncludingLogOnly() <= STRICT_NOT_RETHROW,
|
||||
() -> "不重抛超过本轮严格口径实测 " + STRICT_NOT_RETHROW + ":现在是 "
|
||||
+ r.notRethrowIncludingLogOnly() + ",这一格就是「异常到此为止、外面看不见」的总数");
|
||||
assertTrue(r.count(Kind.LOG_ONLY) <= GIVEN_LOG_ONLY,
|
||||
() -> "只 log 的吞异常块超过验收基线 " + GIVEN_LOG_ONLY + ":现在是 " + r.count(Kind.LOG_ONLY));
|
||||
assertTrue(r.count(Kind.LOG_ONLY) <= STRICT_LOG_ONLY,
|
||||
() -> "只 log 的吞异常块超过本轮严格口径实测 " + STRICT_LOG_ONLY + ":现在是 "
|
||||
+ r.count(Kind.LOG_ONLY) + "(严格口径=块内每条顶层语句都是 log 调用;"
|
||||
+ "验收基线 249 那一版把『log 后紧跟 return』也算进来了,所以更宽)");
|
||||
assertTrue(r.count(Kind.EMPTY) <= GIVEN_EMPTY,
|
||||
() -> "空体 catch 超过验收基线 " + GIVEN_EMPTY + ":现在是 " + r.count(Kind.EMPTY));
|
||||
assertTrue(r.count(Kind.EMPTY) <= STRICT_EMPTY,
|
||||
() -> "空体 catch 超过本轮严格口径实测 " + STRICT_EMPTY + ":现在是 " + r.count(Kind.EMPTY)
|
||||
+ ",空体是三类高危出口里最贵的一种(连日志都不留)");
|
||||
}
|
||||
|
||||
/**
|
||||
* 恒等式:total = rethrow + notRethrow(含只 log) + empty。
|
||||
*
|
||||
* <p>不这么钉的话,往分类器里加一个"新格子"就能把气味块从三格里悄悄搬走,
|
||||
* 三个数字同时下降而总噪声一点没少——这正是"内部搬家"绕过棘轮的形状。
|
||||
*/
|
||||
@Test
|
||||
void categoriesPartitionTheWholeDenominator() {
|
||||
Result r = tree();
|
||||
int rethrow = r.count(Kind.RETHROW);
|
||||
int silent = r.notRethrowIncludingLogOnly();
|
||||
int empty = r.count(Kind.EMPTY);
|
||||
assertEquals(r.total(), rethrow + silent + empty,
|
||||
() -> "分类没覆盖全部 catch:total=" + r.total() + " 而 rethrow+notRethrow+empty="
|
||||
+ (rethrow + silent + empty) + ",有块掉在缝里了");
|
||||
assertTrue(rethrow > 0 && silent > 0 && empty > 0,
|
||||
() -> "某一格是 0,说明分类器根本没往那里判过:rethrow=" + rethrow
|
||||
+ " notRethrow=" + silent + " empty=" + empty);
|
||||
}
|
||||
|
||||
/** 每条气味块都要能定位到 文件:行号——只有标量的棘轮会被"这里加一条那里减一条"绕过去。 */
|
||||
@Test
|
||||
void everySilentSiteIsLocatable() {
|
||||
List<Site> sites = tree().sites();
|
||||
assertTrue(sites.stream().allMatch(s -> s.line() > 0), "有 catch 的定位行号是 0,清单没法用");
|
||||
long distinctFiles = tree().byFile().keySet().size();
|
||||
assertTrue(distinctFiles > 100,
|
||||
() -> "只有 " + distinctFiles + " 个文件带 catch,分母不像整棵树");
|
||||
// 单位自证:文件数 != 处数。若两者相等,说明我(或后人)把"每文件一处"当成了口径。
|
||||
assertTrue(sites.size() != distinctFiles,
|
||||
() -> "处数(" + sites.size() + ")恰好等于文件数(" + distinctFiles
|
||||
+ ")——这几乎不可能是真的,倒像是把文件数当成了处数");
|
||||
}
|
||||
|
||||
/**
|
||||
* 注入自证(正例):内存里造四种形状的 catch,扫描器必须一种都不放过、且判对类。
|
||||
*
|
||||
* <p>没有这一条,"四个数都 ≤ 基线"完全可能是恒真的——分类器返回空清单时它照样绿。
|
||||
*/
|
||||
@Test
|
||||
void injectedSitesAreCountedAndClassified() {
|
||||
Map<String, String> injected = new LinkedHashMap<>();
|
||||
injected.put("Harness.java", String.join("\n",
|
||||
"class Harness {",
|
||||
" void emptyBody() { try { work(); } catch (Exception e) { } }",
|
||||
" void logOnly() { try { work(); } catch (Exception e) { log.warn(\"skip\"); } }",
|
||||
" void swallowWithSideEffect() { try { work(); } catch (Exception e) { marks.add(1); } }",
|
||||
" void rethrows() { try { work(); } catch (Exception e) { throw new IllegalStateException(e); } }",
|
||||
" void logThenReturn() { try { work(); } catch (Exception e) { log.warn(\"x\"); return; } }",
|
||||
"}",
|
||||
""));
|
||||
Result r = SilentCatchScanner.scanSources(injected);
|
||||
assertEquals(5, r.total(), "注入的 5 处 catch 没全数出来");
|
||||
assertEquals(1, r.count(Kind.EMPTY), "空体那处没判成 EMPTY");
|
||||
assertEquals(1, r.count(Kind.LOG_ONLY), "只有 log 的那处没判成 LOG_ONLY");
|
||||
assertEquals(2, r.count(Kind.NOT_RETHROW),
|
||||
"带副作用的不重抛(marks.add / log+return)应是 2 处");
|
||||
assertEquals(1, r.count(Kind.RETHROW), "重抛那处不该被算成静默");
|
||||
assertEquals(3, r.notRethrowIncludingLogOnly(),
|
||||
"不重抛这一格(含只 log)应是 3 处:marks.add 那处 + log 后接 return 那处 + 纯 log 那处"
|
||||
+ "(空体单列一格,不在这格里)");
|
||||
// 反证:同一份内容里如果一个字都不加,总数就是 0——证明上面的数字来自注入而非主树
|
||||
assertEquals(0, SilentCatchScanner.scanSources(Map.of()).total());
|
||||
}
|
||||
|
||||
/**
|
||||
* 注入自证(词法陷阱):字符字面量与转义引号不许吃掉后半棵树。
|
||||
*
|
||||
* <p>这是用户点名的坑:{@code '\"'} 与 {@code '\''} 处理不对,后面的 {@code catch} 就整片消失
|
||||
* (翻车版扫出 818 vs 真实 1000)。这里用一份小夹具把两种陷阱和真 catch 混在一起,
|
||||
* 要求扫描器只数到真 catch,同时独立字面量口径必须数到更多——两口径的差值本身就是证据。
|
||||
*/
|
||||
@Test
|
||||
void charLiteralsAndEscapedQuotesDoNotBlindTheLexer() {
|
||||
// 文本块里的 \\ 会变成一个反斜杠,于是夹具里是货真价实的 \" 与 '\'' 与 '\\'
|
||||
String trapSource = """
|
||||
class Trap {
|
||||
String decoy = "a decoy \\" catch ( inside a string";
|
||||
char apostrophe = '\\'';
|
||||
char backChar = '\\\\';
|
||||
char quoteChar = '"';
|
||||
String after = "still here";
|
||||
void real() { try { work(); } catch (Exception e) { marks.add(1); } }
|
||||
void alsoReal() { try { work(); } catch (IllegalStateException e) { } }
|
||||
}
|
||||
""";
|
||||
Map<String, String> trap = Map.of("Trap.java", trapSource);
|
||||
int naive = SilentCatchScanner.countNaiveLiteral(trap);
|
||||
int scanned = SilentCatchScanner.scanSources(trap).total();
|
||||
assertEquals(2, scanned,
|
||||
() -> "真 catch 只有 2 处,扫描器数出 " + scanned
|
||||
+ " 处——字面量一处理错,后面的代码就整段被吞(818 vs 1000 那个形状)");
|
||||
assertNotEquals(naive, scanned,
|
||||
"字符串里的 decoy catch 也被数进来了:说明扫描器根本没在剥字符串(两口径此刻必须不同)");
|
||||
assertTrue(naive > scanned,
|
||||
() -> "字面量口径(" + naive + ")应大于词法口径(" + scanned + "),因为夹具里埋了一条假 catch");
|
||||
assertEquals(1, naive - scanned, "两口径的差应当正好是字符串里那一条 decoy");
|
||||
}
|
||||
|
||||
/** 打当前四格计数与假 catch 清单(口径复核用,不产生断言)。 */
|
||||
@Test
|
||||
void printCurrentCounts() {
|
||||
Result r = tree();
|
||||
Map<String, String> raw = readTree();
|
||||
System.out.println("CATCH total=" + r.total() + " rethrow=" + r.count(Kind.RETHROW)
|
||||
+ " notRethrow(incl logOnly)=" + r.notRethrowIncludingLogOnly()
|
||||
+ " logOnly=" + r.count(Kind.LOG_ONLY) + " empty=" + r.count(Kind.EMPTY)
|
||||
+ " files=" + r.byFile().size()
|
||||
+ " naive=" + SilentCatchScanner.countNaiveLiteral(raw)
|
||||
+ " decoy=" + SilentCatchScanner.decoyOccurrences(raw).size());
|
||||
System.out.println("DECOY_BEGIN");
|
||||
SilentCatchScanner.decoyOccurrences(raw).forEach(d ->
|
||||
System.out.println("DECOY " + d.file() + ":" + d.line() + " | " + d.context()));
|
||||
System.out.println("DECOY_END");
|
||||
}
|
||||
|
||||
private static Map<String, String> readTree() {
|
||||
Map<String, String> out = new LinkedHashMap<>();
|
||||
try (var walk = Files.walk(TREE)) {
|
||||
for (Path p : walk.filter(Files::isRegularFile)
|
||||
.filter(p -> p.getFileName().toString().endsWith(".java")).sorted().toList()) {
|
||||
out.put(TREE.relativize(p).toString().replace('\\', '/'),
|
||||
new String(Files.readAllBytes(p), StandardCharsets.UTF_8));
|
||||
}
|
||||
} catch (IOException e) {
|
||||
throw new UncheckedIOException(e);
|
||||
}
|
||||
return out;
|
||||
}
|
||||
|
||||
private static Path moduleRoot() {
|
||||
Path here = Path.of("").toAbsolutePath();
|
||||
if (Files.isDirectory(here.resolve("src/main/java"))) {
|
||||
return here;
|
||||
}
|
||||
Path nested = here.resolve("backend-java");
|
||||
if (Files.isDirectory(nested.resolve("src/main/java"))) {
|
||||
return nested;
|
||||
}
|
||||
throw new IllegalStateException("无法定位 backend-java 模块目录(cwd=" + here + ")");
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,475 @@
|
||||
package com.nanri.aiimage.config;
|
||||
|
||||
import java.io.IOException;
|
||||
import java.io.UncheckedIOException;
|
||||
import java.nio.charset.StandardCharsets;
|
||||
import java.nio.file.Files;
|
||||
import java.nio.file.Path;
|
||||
import java.util.ArrayList;
|
||||
import java.util.LinkedHashMap;
|
||||
import java.util.List;
|
||||
import java.util.Map;
|
||||
import java.util.regex.Matcher;
|
||||
import java.util.regex.Pattern;
|
||||
import java.util.stream.Stream;
|
||||
|
||||
/**
|
||||
* catch 块静默形状的扫描器(测试期工具,不进生产 jar 的行为面)。
|
||||
*
|
||||
* <p>四个口径,单位一律是**处**(一个 catch 块算一处),不是文件数:
|
||||
* <ul>
|
||||
* <li>{@code total}——catch 块总数;</li>
|
||||
* <li>{@code notRethrow}——块体非空且不含 {@code throw}(吞掉异常往下走);</li>
|
||||
* <li>{@code logOnly}——块体里每一条顶层语句都是 log 调用(记一行就当没事发生);</li>
|
||||
* <li>{@code empty}——块体为空(剥掉注释后一个字符都不剩)。</li>
|
||||
* </ul>
|
||||
* {@code empty} 与 {@code notRethrow} 是并列的两格(空体不计入 notRethrow),
|
||||
* 恒等式 {@code total = rethrow + notRethrow + empty} 由 {@code CatchSilenceRatchetTest} 检查,
|
||||
* 保证没有第五种形状从分类缝里漏掉。
|
||||
*
|
||||
* <p><b>词法是第一号的坑</b>:简陋的"剥字符串"会把 {@code '"} 或 {@code \"} 读歪,
|
||||
* 从整棵树上抹掉半棵树的内容(实测翻车版本:扫出 818 处 vs 真实 1000 处)。
|
||||
* 这里 {@link #stripNonCode} 是一个手写的单遍扫描,依次处理
|
||||
* 行注释 / 块注释 / 文本块 {@code """..."""} / 字符串字面量 / 字符字面量,
|
||||
* 并且**保持长度与换行位置不变**(这样行号仍然对得上)。
|
||||
* 即便如此,测试仍不信任它自己的分母:那条用例会把扫描器数出的 total 与
|
||||
* 一份独立口径(原始文件里 {@code "catch ("} 的字面量出现次数)对撞,不一致就红。
|
||||
*/
|
||||
public final class SilentCatchScanner {
|
||||
|
||||
/** 一个 catch 块的分类结果。 */
|
||||
public enum Kind {
|
||||
/** 块体里有 throw(重抛或抛新异常)——不属于静默出口。 */
|
||||
RETHROW,
|
||||
/** 块体非空、不含 throw,且每条顶层语句都是 log 调用。 */
|
||||
LOG_ONLY,
|
||||
/** 块体非空、不含 throw,还干了别的(或干脆 return/吞掉)。 */
|
||||
NOT_RETHROW,
|
||||
/** 块体剥掉注释后为空。 */
|
||||
EMPTY
|
||||
}
|
||||
|
||||
/**
|
||||
* 一处 catch 的定位与判类结果。
|
||||
*
|
||||
* @param rawBody 未剥词法的块体原文——判断"有没有留下锚点"必须看这个,
|
||||
* 因为 {@code log.warn("[xxx] ...")} 的方括号就在字符串里,剥完就没了。
|
||||
*/
|
||||
public record Site(String file, int line, Kind kind, String body, String rawBody) {
|
||||
|
||||
/** 块体原文里是否存在 {@code [tag]} 形式的锚点。 */
|
||||
public boolean hasAnchor() {
|
||||
return ANCHOR.matcher(rawBody).find();
|
||||
}
|
||||
|
||||
/** 锚点清单(可能多个),用于人工核对。 */
|
||||
public java.util.List<String> anchors() {
|
||||
java.util.List<String> out = new ArrayList<>();
|
||||
java.util.regex.Matcher m = ANCHOR.matcher(rawBody);
|
||||
while (m.find()) {
|
||||
out.add(m.group());
|
||||
}
|
||||
return out;
|
||||
}
|
||||
}
|
||||
|
||||
private static final Pattern ANCHOR = Pattern.compile("\\[[A-Za-z0-9_.-]{3,}\\]");
|
||||
|
||||
/** 一次扫描的汇总。 */
|
||||
public record Result(List<Site> sites) {
|
||||
public int total() {
|
||||
return sites.size();
|
||||
}
|
||||
|
||||
public int count(Kind kind) {
|
||||
return (int) sites.stream().filter(s -> s.kind() == kind).count();
|
||||
}
|
||||
|
||||
/** {@code notRethrow} 这一格按用户口径 = 非空且不含 throw,即 LOG_ONLY 也计入。 */
|
||||
public int notRethrowIncludingLogOnly() {
|
||||
return count(Kind.NOT_RETHROW) + count(Kind.LOG_ONLY);
|
||||
}
|
||||
|
||||
public Map<String, int[]> byFile() {
|
||||
Map<String, int[]> out = new LinkedHashMap<>();
|
||||
for (Site s : sites) {
|
||||
int[] cell = out.computeIfAbsent(s.file(), k -> new int[4]);
|
||||
cell[0]++;
|
||||
if (s.kind() == Kind.NOT_RETHROW || s.kind() == Kind.LOG_ONLY) {
|
||||
cell[1]++;
|
||||
}
|
||||
if (s.kind() == Kind.LOG_ONLY) {
|
||||
cell[2]++;
|
||||
}
|
||||
if (s.kind() == Kind.EMPTY) {
|
||||
cell[3]++;
|
||||
}
|
||||
}
|
||||
return out;
|
||||
}
|
||||
}
|
||||
|
||||
/** log 调用语句:接收者只认 slf4j 那几个常见名字,实测全树 log 调用的接收者只有 {@code log} 一个。 */
|
||||
private static final Pattern LOG_CALL =
|
||||
Pattern.compile("^(?:this\\.)?(?:log|logger|LOG|LOGGER)\\s*\\.\\s*[A-Za-z_$][\\w$]*\\s*\\(.*\\)\\z",
|
||||
Pattern.DOTALL);
|
||||
|
||||
/** 独立简单口径:原始字节里的字面量 {@code "catch ("},与 {@code grep -o "catch (" | wc -l} 同一件事。 */
|
||||
public static final String NAIVE_LITERAL = "catch (";
|
||||
|
||||
private SilentCatchScanner() {
|
||||
}
|
||||
|
||||
/** 扫一棵树(通常是 {@code src/main/java})。 */
|
||||
public static Result scanTree(Path root) {
|
||||
Map<String, String> sources = new LinkedHashMap<>();
|
||||
try (Stream<Path> walk = Files.walk(root)) {
|
||||
for (Path p : walk.filter(Files::isRegularFile)
|
||||
.filter(p -> p.getFileName().toString().endsWith(".java"))
|
||||
.sorted()
|
||||
.toList()) {
|
||||
sources.put(root.relativize(p).toString().replace('\\', '/'),
|
||||
new String(Files.readAllBytes(p), StandardCharsets.UTF_8));
|
||||
}
|
||||
} catch (IOException e) {
|
||||
throw new UncheckedIOException("扫描失败: " + root, e);
|
||||
}
|
||||
return scanSources(sources);
|
||||
}
|
||||
|
||||
/**
|
||||
* 扫内存里的源码集合(名字 -> 内容)。注入反例走这个入口,
|
||||
* 于是"能不能数出来"这件事不必往主树上写文件就能自证。
|
||||
*/
|
||||
public static Result scanSources(Map<String, String> sources) {
|
||||
List<Site> sites = new ArrayList<>();
|
||||
for (Map.Entry<String, String> e : sources.entrySet()) {
|
||||
collect(e.getKey(), e.getValue(), sites);
|
||||
}
|
||||
return new Result(List.copyOf(sites));
|
||||
}
|
||||
|
||||
/** 原始字节里有 {@value #NAIVE_LITERAL} 字样、但落在注释/字面量里(因此不是真 catch)的那一处。 */
|
||||
public record Decoy(String file, int line, String context) {
|
||||
}
|
||||
|
||||
/** 原始内容里 {@value #NAIVE_LITERAL} 的字面量出现次数(独立分母,不经任何词法处理)。 */
|
||||
public static int countNaiveLiteral(Map<String, String> sources) {
|
||||
int n = 0;
|
||||
for (String src : sources.values()) {
|
||||
int at = 0;
|
||||
while ((at = src.indexOf(NAIVE_LITERAL, at)) >= 0) {
|
||||
n++;
|
||||
at += NAIVE_LITERAL.length();
|
||||
}
|
||||
}
|
||||
return n;
|
||||
}
|
||||
|
||||
/**
|
||||
* 找出"grep 会数到、但根本不是 catch"的那些位置:原始文本里有 {@code catch (} 字样,
|
||||
* 而 {@link #stripNonCode} 把同一位置的内容抹掉了(说明它在注释或字面量里)。
|
||||
*
|
||||
* <p>这个清单是扫描器自己的解毒剂:词法一旦写坏(把真代码当成字符串吃掉),
|
||||
* total 会**变小**、棘轮的"≤ 基线"照样绿——只有 decoy 清单会同步变大。
|
||||
* 所以判据是 {@code total + decoy == 字面量口径} 且 {@code decoy} 不许超过基线。
|
||||
*/
|
||||
public static List<Decoy> decoyOccurrences(Map<String, String> sources) {
|
||||
List<Decoy> out = new ArrayList<>();
|
||||
for (Map.Entry<String, String> e : sources.entrySet()) {
|
||||
String raw = e.getValue();
|
||||
String code = stripNonCode(raw);
|
||||
int at = 0;
|
||||
while (true) {
|
||||
int i = raw.indexOf(NAIVE_LITERAL, at);
|
||||
if (i < 0) {
|
||||
break;
|
||||
}
|
||||
if (!code.startsWith(NAIVE_LITERAL, i)) {
|
||||
out.add(new Decoy(e.getKey(), lineOf(raw, i), contextOf(raw, i)));
|
||||
}
|
||||
at = i + NAIVE_LITERAL.length();
|
||||
}
|
||||
}
|
||||
return out;
|
||||
}
|
||||
|
||||
private static String contextOf(String raw, int index) {
|
||||
int start = raw.lastIndexOf('\n', index) + 1;
|
||||
int end = raw.indexOf('\n', index);
|
||||
if (end < 0) {
|
||||
end = raw.length();
|
||||
}
|
||||
String line = raw.substring(start, end).trim();
|
||||
return line.length() > 100 ? line.substring(0, 100) + "..." : line;
|
||||
}
|
||||
|
||||
private static void collect(String file, String source, List<Site> out) {
|
||||
String code = stripNonCode(source);
|
||||
int at = 0;
|
||||
while (true) {
|
||||
int hit = indexOfCatch(code, at);
|
||||
if (hit < 0) {
|
||||
return;
|
||||
}
|
||||
int open = code.indexOf('(', hit);
|
||||
int close = matchParen(code, open);
|
||||
if (close < 0) {
|
||||
return;
|
||||
}
|
||||
int brace = code.indexOf('{', close);
|
||||
int end = brace < 0 ? -1 : matchBrace(code, brace);
|
||||
if (end < 0) {
|
||||
// 结构读不出来就明说:静默跳过会让分母变小而棘轮照样绿
|
||||
throw new IllegalStateException("解析不出 " + file + " 第 " + lineOf(code, hit)
|
||||
+ " 行那个 catch 的块体(括号没配好?)");
|
||||
}
|
||||
String body = code.substring(brace + 1, end);
|
||||
// stripNonCode 保持长度与换行位置不变,所以同一个下标区间在原文上取到的就是块体原文
|
||||
String rawBody = source.substring(brace + 1, end);
|
||||
out.add(new Site(file, lineOf(code, hit), classify(body), body, rawBody));
|
||||
// 游标只前进到 "catch" 这个词之后,**不跳过块体**:
|
||||
// catch 体里再套 try/catch 是常态(外层兜底、内层只吞某一步),
|
||||
// 整段跳过的话这 7 处嵌套 catch 就从分母里消失了(实测 994 vs 字面量 1001)。
|
||||
at = hit + "catch".length();
|
||||
}
|
||||
}
|
||||
|
||||
/** {@code catch} 关键字后面(允许空白/换行)紧跟 {@code (} 的那一处。 */
|
||||
private static int indexOfCatch(String code, int from) {
|
||||
int at = from;
|
||||
while (true) {
|
||||
int i = code.indexOf("catch", at);
|
||||
if (i < 0) {
|
||||
return -1;
|
||||
}
|
||||
boolean leftOk = i == 0 || !isIdentPart(code.charAt(i - 1));
|
||||
int j = i + "catch".length();
|
||||
while (j < code.length() && Character.isWhitespace(code.charAt(j))) {
|
||||
j++;
|
||||
}
|
||||
if (leftOk && j < code.length() && code.charAt(j) == '(') {
|
||||
return i;
|
||||
}
|
||||
at = i + "catch".length();
|
||||
}
|
||||
}
|
||||
|
||||
private static Kind classify(String body) {
|
||||
String trimmed = body.trim();
|
||||
if (trimmed.isEmpty()) {
|
||||
return Kind.EMPTY;
|
||||
}
|
||||
if (containsKeyword(trimmed, "throw")) {
|
||||
return Kind.RETHROW;
|
||||
}
|
||||
List<String> statements = topLevelStatements(trimmed);
|
||||
if (!statements.isEmpty()) {
|
||||
boolean allLog = true;
|
||||
for (String s : statements) {
|
||||
String one = s.trim();
|
||||
if (!one.isEmpty() && !LOG_CALL.matcher(one).matches()) {
|
||||
allLog = false;
|
||||
break;
|
||||
}
|
||||
}
|
||||
if (allLog) {
|
||||
return Kind.LOG_ONLY;
|
||||
}
|
||||
}
|
||||
return Kind.NOT_RETHROW;
|
||||
}
|
||||
|
||||
private static boolean containsKeyword(String text, String keyword) {
|
||||
int at = 0;
|
||||
while (true) {
|
||||
int i = text.indexOf(keyword, at);
|
||||
if (i < 0) {
|
||||
return false;
|
||||
}
|
||||
boolean leftOk = i == 0 || !isIdentPart(text.charAt(i - 1));
|
||||
int after = i + keyword.length();
|
||||
boolean rightOk = after >= text.length() || !isIdentPart(text.charAt(after));
|
||||
if (leftOk && rightOk) {
|
||||
return true;
|
||||
}
|
||||
at = after;
|
||||
}
|
||||
}
|
||||
|
||||
private static boolean isIdentPart(char c) {
|
||||
return Character.isLetterOrDigit(c) || c == '_' || c == '$';
|
||||
}
|
||||
|
||||
/** 按顶层 {@code ;} 切语句({@code ()[]{} } 内部的分号不算分隔符,for(;;) 就是这一类)。 */
|
||||
static List<String> topLevelStatements(String body) {
|
||||
List<String> out = new ArrayList<>();
|
||||
int depth = 0;
|
||||
StringBuilder cur = new StringBuilder();
|
||||
for (int i = 0; i < body.length(); i++) {
|
||||
char c = body.charAt(i);
|
||||
if (c == '(' || c == '[' || c == '{') {
|
||||
depth++;
|
||||
} else if (c == ')' || c == ']') {
|
||||
depth--;
|
||||
} else if (c == '}' && depth > 0) {
|
||||
depth--;
|
||||
} else if (c == ';' && depth == 0) {
|
||||
out.add(cur.toString());
|
||||
cur.setLength(0);
|
||||
continue;
|
||||
}
|
||||
cur.append(c);
|
||||
}
|
||||
if (cur.toString().trim().isEmpty()) {
|
||||
return out;
|
||||
}
|
||||
out.add(cur.toString());
|
||||
return out;
|
||||
}
|
||||
|
||||
private static int matchParen(String s, int open) {
|
||||
return match(s, open, '(', ')');
|
||||
}
|
||||
|
||||
private static int matchBrace(String s, int open) {
|
||||
return match(s, open, '{', '}');
|
||||
}
|
||||
|
||||
private static int match(String s, int open, char lo, char hi) {
|
||||
if (open < 0 || open >= s.length() || s.charAt(open) != lo) {
|
||||
return -1;
|
||||
}
|
||||
int depth = 0;
|
||||
for (int i = open; i < s.length(); i++) {
|
||||
char c = s.charAt(i);
|
||||
if (c == lo) {
|
||||
depth++;
|
||||
} else if (c == hi) {
|
||||
depth--;
|
||||
if (depth == 0) {
|
||||
return i;
|
||||
}
|
||||
}
|
||||
}
|
||||
return -1;
|
||||
}
|
||||
|
||||
private static int lineOf(String code, int index) {
|
||||
int line = 1;
|
||||
for (int i = 0; i < index && i < code.length(); i++) {
|
||||
if (code.charAt(i) == '\n') {
|
||||
line++;
|
||||
}
|
||||
}
|
||||
return line;
|
||||
}
|
||||
|
||||
/**
|
||||
* 把注释与字面量的内容换成空格,长度与换行位置一律不变。
|
||||
*
|
||||
* <p>顺序即词法:先看 {@code /} 的两种注释,再看文本块(必须在普通字符串之前,
|
||||
* 否则 {@code """} 会被当成一个空串加一个串),然后普通串,最后字符字面量。
|
||||
* 反斜杠一律跳过下一个字符——于是 {@code "\""} 里的引号不会终结字符串、
|
||||
* {@code '\\'} 不会让下一个 {@code '\''} 变成孤儿引号(这正是 818/1000 那版翻车的地方)。
|
||||
*/
|
||||
static String stripNonCode(String src) {
|
||||
char[] c = src.toCharArray();
|
||||
int n = c.length;
|
||||
int i = 0;
|
||||
while (i < n) {
|
||||
char ch = c[i];
|
||||
if (ch == '/' && i + 1 < n && c[i + 1] == '/') {
|
||||
int j = i;
|
||||
while (j < n && c[j] != '\n') {
|
||||
c[j] = ' ';
|
||||
j++;
|
||||
}
|
||||
i = j;
|
||||
} else if (ch == '/' && i + 1 < n && c[i + 1] == '*') {
|
||||
int j = i;
|
||||
int close = src.indexOf("*/", i + 2);
|
||||
int end = close < 0 ? n : close + 2;
|
||||
while (j < end) {
|
||||
if (c[j] != '\n') {
|
||||
c[j] = ' ';
|
||||
}
|
||||
j++;
|
||||
}
|
||||
i = end;
|
||||
} else if (ch == '"' && startsTextBlock(src, i)) {
|
||||
int j = endOfTextBlock(src, i);
|
||||
for (int k = i; k < j; k++) {
|
||||
if (c[k] != '\n') {
|
||||
c[k] = ' ';
|
||||
}
|
||||
}
|
||||
i = j;
|
||||
} else if (ch == '"') {
|
||||
int j = i + 1;
|
||||
while (j < n) {
|
||||
if (c[j] == '\\') {
|
||||
j += 2;
|
||||
continue;
|
||||
}
|
||||
if (c[j] == '\n') {
|
||||
break; // 普通串不跨行:行尾就是终结,别把下一行吃掉
|
||||
}
|
||||
if (c[j] == '"') {
|
||||
j++;
|
||||
break;
|
||||
}
|
||||
j++;
|
||||
}
|
||||
for (int k = i + 1; k < j; k++) {
|
||||
c[k] = ' ';
|
||||
}
|
||||
i = j;
|
||||
} else if (ch == '\'') {
|
||||
int j = i + 1;
|
||||
while (j < n) {
|
||||
if (c[j] == '\\') {
|
||||
j += 2;
|
||||
continue;
|
||||
}
|
||||
if (c[j] == '\n') {
|
||||
break;
|
||||
}
|
||||
if (c[j] == '\'') {
|
||||
j++;
|
||||
break;
|
||||
}
|
||||
j++;
|
||||
}
|
||||
for (int k = i + 1; k < j; k++) {
|
||||
c[k] = ' ';
|
||||
}
|
||||
i = j;
|
||||
} else {
|
||||
i++;
|
||||
}
|
||||
}
|
||||
return new String(c);
|
||||
}
|
||||
|
||||
private static boolean startsTextBlock(String src, int i) {
|
||||
return src.startsWith("\"\"\"", i);
|
||||
}
|
||||
|
||||
/** 文本块的终点 = 结束定界符之后(找不到就走到文件尾,绝不吃掉后续代码)。 */
|
||||
private static int endOfTextBlock(String src, int i) {
|
||||
int j = i + 3;
|
||||
int n = src.length();
|
||||
while (j < n) {
|
||||
if (src.charAt(j) == '\\') {
|
||||
j += 2;
|
||||
continue;
|
||||
}
|
||||
if (src.startsWith("\"\"\"", j)) {
|
||||
return j + 3;
|
||||
}
|
||||
j++;
|
||||
}
|
||||
return n;
|
||||
}
|
||||
}
|
||||
Reference in New Issue
Block a user