C/C++: extern "c" will cause parse error #2647

chongchai · 2020-09-17T03:49:48Z

I find the definition will not be parsed in the #else and extern "C".

The sample code:

#ifdef __cplusplus
extern "C" {
#endif

#ifdef MAX
  void testMax1() {}
  #define NB1 11
#else
  void testMax2() {}
  #define NB2 22
#endif

#ifdef __cplusplus
}
#endif

#ifdef MAX
  void testMax3() {}
  #define NB3 33
#else
  void testMax4() {}
  #define NB4 44
#endif

I use following command to parse:

ctags.exe --output-format=json --langmap=c:+.inc --kinds-c=+plz --fields=+nie -o - D:\tmp\test\testIfMacro.c

Result is:

{"_type": "tag", "name": "NB1", "path": "D:\\tmp\\test\\testIfMacro.c", "pattern": "/^  #define NB1 /", "file": true, "line": 7, "kind": "macro", "end": 7}
{"_type": "tag", "name": "NB3", "path": "D:\\tmp\\test\\testIfMacro.c", "pattern": "/^  #define NB3 /", "file": true, "line": 19, "kind": "macro", "end": 19}
{"_type": "tag", "name": "NB4", "path": "D:\\tmp\\test\\testIfMacro.c", "pattern": "/^  #define NB4 /", "file": true, "line": 22, "kind": "macro", "end": 22}
{"_type": "tag", "name": "testMax1", "path": "D:\\tmp\\test\\testIfMacro.c", "pattern": "/^  void testMax1() {}$/", "line": 6, "typeref": "typename:void", "kind": "function", "end": 6}
{"_type": "tag", "name": "testMax3", "path": "D:\\tmp\\test\\testIfMacro.c", "pattern": "/^  void testMax3() {}$/", "line": 18, "typeref": "typename:void", "kind": "function", "end": 18}
{"_type": "tag", "name": "testMax4", "path": "D:\\tmp\\test\\testIfMacro.c", "pattern": "/^  void testMax4() {}$/", "line": 21, "typeref": "typename:void", "kind": "function", "end": 21}

The testMax2 and NB2 are not parsed, while testMax4 and NB4 are parsed.
I thinkextern "C" will cause this error. Can you support this situation?

The text was updated successfully, but these errors were encountered:

masatake · 2020-09-17T05:03:55Z

I thinkextern "C" will cause this error. Can you support this situation?

It seems that your guessing is correct.
Following change is for ignoring the code between "#ifdef __cplusplus ~ #endif.
@pragmaware, how do you think about my approach?

$ git diff | cat
diff --git a/parsers/cpreprocessor.c b/parsers/cpreprocessor.c
index 668171f0..4ad6087d 100644
--- a/parsers/cpreprocessor.c
+++ b/parsers/cpreprocessor.c
@@ -953,10 +953,31 @@ static void directivePragma (int c)
 	Cpp.directive.state = DRCTV_NONE;
 }
 
+static bool isDefCondition (const int c, const char *condition)
+{
+	if (*condition == '\0')
+		return true;
+	else if (c == EOF)
+		return false;
+
+	if (*condition != '\0' && c == condition[0])
+	{
+		const int next = cppGetcFromUngetBufferOrFile ();
+		return isDefCondition (next, condition + 1);
+	}
+
+	return false;
+}
+
 static bool directiveIf (const int c)
 {
 	DebugStatement ( const bool ignore0 = isIgnore (); )
-	const bool ignore = pushConditional ((bool) (c != '0'));
+	bool firstBranchChosen = true;
+
+	if (c == '0' || isDefCondition (c, "__cplusplus"))
+		firstBranchChosen = false;
+
+	const bool ignore = pushConditional (firstBranchChosen);
 
 	Cpp.directive.state = DRCTV_NONE;
 	DebugStatement ( debugCppNest (true, Cpp.directive.nestLevel);
$ u-ctags --output-format=json --langmap=c:+.inc --kinds-c=+plz --fields=+nie -o - /tmp/foo.c

u-ctags --output-format=json --langmap=c:+.inc --kinds-c=+plz --fields=+nie -o - /tmp/foo.c

{"_type": "tag", "name": "NB1", "path": "/tmp/foo.c", "pattern": "/^  #define NB1 /", "file": true, "line": 7, "kind": "macro", "end": 7}
{"_type": "tag", "name": "NB2", "path": "/tmp/foo.c", "pattern": "/^  #define NB2 /", "file": true, "line": 10, "kind": "macro", "end": 10}
{"_type": "tag", "name": "NB3", "path": "/tmp/foo.c", "pattern": "/^  #define NB3 /", "file": true, "line": 19, "kind": "macro", "end": 19}
{"_type": "tag", "name": "NB4", "path": "/tmp/foo.c", "pattern": "/^  #define NB4 /", "file": true, "line": 22, "kind": "macro", "end": 22}
{"_type": "tag", "name": "testMax1", "path": "/tmp/foo.c", "pattern": "/^  void testMax1() {}$/", "line": 6, "typeref": "typename:void", "kind": "function", "end": 6}
{"_type": "tag", "name": "testMax2", "path": "/tmp/foo.c", "pattern": "/^  void testMax2() {}$/", "line": 9, "typeref": "typename:void", "kind": "function", "end": 9}
{"_type": "tag", "name": "testMax3", "path": "/tmp/foo.c", "pattern": "/^  void testMax3() {}$/", "line": 18, "typeref": "typename:void", "kind": "function", "end": 18}
{"_type": "tag", "name": "testMax4", "path": "/tmp/foo.c", "pattern": "/^  void testMax4() {}$/", "line": 21, "typeref": "typename:void", "kind": "function", "end": 21}

pragmaware · 2020-09-17T10:49:43Z

Well, if it fixes the problem and doesn't break tests then it looks good.

However, I don't understand the relation between the #ifdef __cplusplus define and the #ifdef MAX one.
They shouldn't influence each other in terms of choosing one or multiple branches... or I'm missing something?

masatake · 2020-09-17T14:54:13Z

My analysis is that it is related to how c parser handles extern "C" { }.

$ cat /tmp/foo.c
cat /tmp/foo.c
extern "C" {

#ifdef MAX
  void testMax1() {}
  #define NB1 11
#else
  void testMax2() {}
  #define NB2 22
#endif

}
$ u-ctags --output-format=xref --kinds-c=+plz --fields=+nie -o - /tmp/foo.c

u-ctags --output-format=xref --kinds-c=+plz --fields=+nie -o - /tmp/foo.c

NB1              macro         5 /tmp/foo.c       #define NB1 11
testMax1         function      4 /tmp/foo.c       void testMax1() {}

$ cat /tmp/bar.c 
cat /tmp/bar.c 
// extern "C" {

#ifdef MAX
  void testMax1() {}
  #define NB1 11
#else
  void testMax2() {}
  #define NB2 22
#endif

// }
$ u-ctags --output-format=xref --kinds-c=+plz --fields=+nie -o - /tmp/bar.c

u-ctags --output-format=xref --kinds-c=+plz --fields=+nie -o - /tmp/bar.c

NB1              macro         5 /tmp/bar.c       #define NB1 11
NB2              macro         8 /tmp/bar.c       #define NB2 22
testMax1         function      4 /tmp/bar.c       void testMax1() {}
testMax2         function      7 /tmp/bar.c       void testMax2() {}

If cpreprocessor parser suppresses the area #ifdef __cplusplus ~ #end, c parser doesn't read the line extern "C" { and }.

I have to add one more condition to the patch (#2647 (comment)).
The condition is "if the client parser of cpreprocessor parser is c parser" or "if the client parser of cpreprocessor parser is not c++ parser".

Close universal-ctags#2647. C parser doesn't work when C++ code "extern "C" {" and "}" is given. $ cat /tmp/extern-c.c extern "C" { #ifdef X void f() {} #else void g() {} #endif } $ u-ctags --output-format=xref --kinds-c=+plz --fields=+nie -o - /tmp/extern-c.c f function 3 /tmp/extern-c.c void f() {} $ cat /tmp/none-extern-c.c #ifdef X void f() {} #else void g() {} #endif $ u-ctags --output-format=xref --kinds-c=+plz --fields=+nie -o - /tmp/none-extern-c.c f function 2 /tmp/none-extern-c.c void f() {} g function 4 /tmp/none-extern-c.c void g() {} The function g() is missed if the code is surrounded by extern "C" { and }. This change uses a heuristic rule to skip the code in the C parser. In most of cases, the code is in #ifdef __cplusplus ~ #endif branch. So when the CPreProcessor parser detects #ifdef __cplusplus ~ #endif, with this change, the CPreProcessor parser doesn't pass the code inside __cplusplus branch to its client parser that requests to the CPreProcessor parser to do so. C parser requests it. C++ parser doesn't. Signed-off-by: Masatake YAMATO <yamato@redhat.com>

pragmaware · 2020-09-18T03:42:59Z

Though I still don't understand why.
In theory the behaviour of an #ifdef should not be influenced by any previous #ifdef...
... it seems to be more of a bug in the handling of the second #ifdef, not the first one...

masatake · 2020-09-19T06:37:01Z

I found this is nothing to do with extern "C" { ... }.
I can reproduce the behavior with struct s { ... }:

[yamato@control]~/var/ctags-github% cat /tmp/f.c
cat /tmp/f.c
struct s {
#ifdef X
  int i;
#else
  int j;
#endif  
};
[yamato@control]~/var/ctags-github% ./ctags -o - /tmp/f.c
./ctags -o - /tmp/f.c
i	/tmp/f.c	/^  int i;$/;"	m	language:C	struct:s	typeref:typename:int	file:
s	/tmp/f.c	/^struct s {$/;"	s	language:C	file:
[yamato@control]~/var/ctags-github% cat /tmp/g.c
cat /tmp/g.c

#ifdef X
  int i;
#else
  int j;
#endif  

[yamato@control]~/var/ctags-github% ./ctags -o - /tmp/g.c
./ctags -o - /tmp/g.c
i	/tmp/g.c	/^  int i;$/;"	v	language:C	typeref:typename:int
j	/tmp/g.c	/^  int j;$/;"	v	language:C	typeref:typename:int
[yamato@control]~/var/ctags-github%

chongchai · 2020-09-22T06:18:46Z

I can add another testcase for thier behavior.
Local variable In function body like follows

[yklhard@gotpc12]$ cat a.c
void test() {
#ifdef X
  int i = 1;
#else
  int j = 2;
#endif 
}
[yklhard@gotpc12]$ ./ctags-src/ctags --kinds-c=+l -o - a.c
i	a.c	/^  int i = 1;$/;"	l	function:test	typeref:typename:int	file:
test	a.c	/^void test() {$/;"	f	typeref:typename:void
[yklhard@gotpc12]$ cat b.c
#ifdef X
  int i = 1;
#else
  int j = 2;
#endif 
[yklhard@gotpc12]$ ./ctags-src/ctags --kinds-c=+l -o - b.c
i	b.c	/^  int i = 1;$/;"	v	typeref:typename:int
j	b.c	/^  int j = 2;$/;"	v	typeref:typename:int

ArcsinX · 2021-04-07T08:49:58Z

Seems that we should not call cppBeginStatement() here https://github.com/universal-ctags/ctags/blob/master/parsers/cxx/cxx_parser_block.c#L262 :

the fact that we are inside some code block (function, struct, namespace, extern "C", etc.) is not a good enough reason to skip (resolve) branches.
this change does not break any test

masatake linked a pull request Sep 17, 2020 that will close this issue

CPreProcessor,C: skip #ifdef __cplusplus ~ #endif branch #2648

Open

ArcsinX mentioned this issue Apr 7, 2021

CPreprocessor: branches skip inside code block #2944

Closed

masatake added the C/C++ label Apr 2, 2022

masatake self-assigned this Apr 2, 2022

masatake pinned this issue Apr 2, 2022

masatake removed the C/C++ label Nov 29, 2022

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

C/C++: extern "c" will cause parse error #2647

C/C++: extern "c" will cause parse error #2647

chongchai commented Sep 17, 2020

masatake commented Sep 17, 2020

pragmaware commented Sep 17, 2020

masatake commented Sep 17, 2020

pragmaware commented Sep 18, 2020

masatake commented Sep 19, 2020

chongchai commented Sep 22, 2020

ArcsinX commented Apr 7, 2021

C/C++: extern "c" will cause parse error #2647

C/C++: extern "c" will cause parse error #2647

Comments

chongchai commented Sep 17, 2020

masatake commented Sep 17, 2020

pragmaware commented Sep 17, 2020

masatake commented Sep 17, 2020

pragmaware commented Sep 18, 2020

masatake commented Sep 19, 2020

chongchai commented Sep 22, 2020

ArcsinX commented Apr 7, 2021