libclang：缺少 AST 中的一些语句？

Question

我写了一个测试程序（parse_ast.c）来解析一个c源文件（tt.c）来看看libclang是如何工作的，输出是AST的层次结构：

这是测试文件：

/* tt.c */                                    // line 1
#include <unistd.h>
#include <stdio.h>

typedef ssize_t (*write_fn_t)(int, const void *, size_t);

void indirect_write(write_fn_t write_fn) {    // line 7
    (*write_fn)(1, "indirect call\n", 14);
}

void direct_write() {                         // line 11
    write(1, "direct call\n", 12);            // line 12 mising in the ast?
}

int main() {                                  // line 15
    direct_write();
    indirect_write(write);                    // line 17 missing in the ast?

    return 0;
}

输出显示如下：

 ...
 ...
 inclusion directive at tt.c (2, 1) to (2, 20)
 inclusion directive at tt.c (3, 1) to (3, 19)
 TypedefDecl at tt.c (5, 1) to (5, 57)
 TypeRef at tt.c (5, 9) to (5, 16)
 ParmDecl at tt.c (5, 31) to (5, 35)
 ParmDecl at tt.c (5, 36) to (5, 49)
 ParmDecl at tt.c (5, 50) to (5, 56)
 FunctionDecl at tt.c (7, 1) to (9, 2)
 ParmDecl at tt.c (7, 21) to (7, 40)
  TypeRef at tt.c (7, 21) to (7, 31)
 CompoundStmt at tt.c (7, 42) to (9, 2)
  CallExpr at tt.c (8, 5) to (8, 42)
   UnexposedExpr at tt.c (8, 5) to (8, 16)
    ParenExpr at tt.c (8, 5) to (8, 16)
     UnaryOperator at tt.c (8, 6) to (8, 15)
      UnexposedExpr at tt.c (8, 7) to (8, 15)
       DeclRefExpr at tt.c (8, 7) to (8, 15)
   IntegerLiteral at tt.c (8, 17) to (8, 18)
   UnexposedExpr at tt.c (8, 20) to (8, 37)
    UnexposedExpr at tt.c (8, 20) to (8, 37)
     StringLiteral at tt.c (8, 20) to (8, 37)
   IntegerLiteral at tt.c (8, 39) to (8, 41)
 FunctionDecl at tt.c (11, 1) to (13, 2)
 CompoundStmt at tt.c (11, 21) to (13, 2)        <- XXX no line 12?
 FunctionDecl at tt.c (15, 1) to (20, 2)
 CompoundStmt at tt.c (15, 12) to (20, 2)
  CallExpr at tt.c (16, 5) to (16, 19)
   UnexposedExpr at tt.c (16, 5) to (16, 17)
    DeclRefExpr at tt.c (16, 5) to (16, 17)      <- XXX no line 17?
  ReturnStmt at tt.c (19, 5) to (19, 13)
   IntegerLiteral at tt.c (19, 12) to (19, 13)

我们可以看到三个函数（第 7 行的 direct_write/第 11 行的 indirect_write/第 15 行的 main），大多数语句都可以在 AST 中找到，但我找不到任何代表语句的内容第 12 行和第 17 行。有人知道原因吗？

我正在使用 debian 2.6.32 挤压，在 clang 3.1 和 3.2 上进行了测试（从源代码编译）。

这里是程序 parse_ast.c:

#include <stddef.h>
#include <stdio.h>
#include <clang-c/Index.h>

enum CXChildVisitResult visit_fn(CXCursor cr, CXCursor parent,
        CXClientData client_data) {

    unsigned depth;
    unsigned line, column, offset;
    enum CXCursorKind kind;
    CXSourceRange extent;
    CXSourceLocation start, end;
    CXString kind_spelling, filename;
    CXFile file;

    depth = (unsigned)client_data;

    // print cursor kind
    kind = clang_getCursorKind(cr);
    kind_spelling = clang_getCursorKindSpelling(kind);
    fprintf(stdout, "%*s%s at", depth, " ", clang_getCString(kind_spelling));
    clang_disposeString(kind_spelling);

    // get extent
    extent = clang_getCursorExtent(cr);
    start = clang_getRangeStart(extent);
    end = clang_getRangeEnd(extent);

    // print start position
    clang_getExpansionLocation(start, &file, &line, &column, &offset);
    filename = clang_getFileName(file);
    fprintf(stdout, " %s (%u, %u) to", clang_getCString(filename), line,
            column);
    clang_disposeString(filename);

    // print end position
    clang_getExpansionLocation(end, &file, &line, &column, &offset);
    fprintf(stdout, " (%u, %u)\n", line, column);

    // recursive
    clang_visitChildren(cr, visit_fn, (CXClientData)(depth + 1));

    return CXChildVisit_Continue;

}

int main(int argc, const char * const *argv) {
    CXIndex Index = clang_createIndex(0, 0);
    CXTranslationUnit TU = clang_parseTranslationUnit(Index, NULL,
            argv, argc, 0, 0, CXTranslationUnit_DetailedPreprocessingRecord);

    clang_visitChildren(clang_getTranslationUnitCursor(TU),
            visit_fn, 0);
    clang_disposeTranslationUnit(TU);
    clang_disposeIndex(Index);

    return 0;
}

更新：

问题是由于缺少头文件stddef.h，它在libclang的邮件列表http://clang-developers.42468.n3.nabble.com/libclang-missing-some-statements-in-the-AST-td4029641.html中得到了回答

Answer 1

检查clang_parseTranslationUnit() 生成的诊断信息——即使遇到错误，也会生成一个 AST，但显然不能保证它是有意义的。

我发现注释掉 #include 行会导致编译错误，但生成的 AST 与您的相似（具体来说，缺少第 17 行）。

将#include 行替换为size_t 和ssize_t（如int）的typedef 会导致关于write() 的隐式声明的编译警告，但AST 包含第17 行。

因此，我认为您的头文件中存在问题，诊断应该会显示该问题。 可以通过例如检索诊断信息

for (unsigned I = 0, N = clang_getNumDiagnostics(TU); I != N; ++I) { 
    CXDiagnostic Diag = clang_getDiagnostic(TU, I);
    CXString String = clang_formatDiagnostic(Diag, clang_defaultDiagnosticDisplayOptions());
    fprintf(stderr, "%s\n", clang_getCString(String));
    clang_disposeString(String);
}

Answer 2

我正在使用 libclang 来解析和优化 c 代码，但是用你的代码解析 c 源文件我看不到 CompoundStmt 中的 CXCursor_BinaryOperator

例如

void OCTS_C_TimerMiliseconds_reset_Timers(OCTS_outC_C_TimerMiliseconds_Timers *outC)
{
    outC->init = kcg_true;
    /* 1 */ OCTS_Sign_INT_reset_Math(&outC->_1_Context_1);
    /* 2 */ OCTS_Sign_INT_reset_Math(&outC->Context_2);
    /* 1 */ OCTS_FallingEdge_reset_Edge(&outC->Context_1);
} 

结果是：

FunctionDecl at s.cpp (10, 1) to (17, 2) OCTS_C_TimerMiliseconds_reset_Timers 
ParmDecl at s.cpp (11, 3) to (11, 44) outC 
 TypeRef at s.cpp (11, 3) to (11, 38) OCTS_outC_C_TimerMiliseconds_Timers 
CompoundStmt at s.cpp (12, 1) to (17, 2)