[clang-sa] Macro definitions of header TUs not included in the plist file

[ankurkraj]

When analyzing a particular file ( using CTU ) . For a macro that is part of the bug-path - the definition of the macro is not included in the plist reports , if the macro is not called in the parent file but in any other intermediate file

Versions:
clang-14,clang-18,clang-20,main branch binaries

Steps to reproduce the behaviour:

Following is a relatively simpler representation of the workflow

file1.c -> calls demo_function(10,0) defined in file2.c
file2.c -> uses macro DIVISION(a,b) in demo_function(int a, int b), macro DIVISION is defined in file3.h
file3.h -> defines DIVISION(a,b) as ((a)/(b))

I have followed the steps for PCH analysis The command I used was

clang --analyze -Xclang -analyzer-config -Xclang experimental-enable-naive-ctu-analysis=true -Xclang -analyzer-config -Xclang ctu-dir=. -Xclang -analyzer-config -Xclang ctu-phase1-inlining=all -Xclang -analyzer-config -Xclang expand-macros=true  -Xclang -analyzer-output=pl
ist-multi-file  -I. file1.c file2.c file3.c

The result displayed on the screen includes the macro definition but the plist files do not contain these value

In file included from file1.c:1:
file2.c:15:50: warning: Division by zero [core.DivideZero]
    printf("Division of %d / %d is: %d\n", a, b, DIVISION(a,b));
                                                 ^~~~~~~~~~~~~
file3.h:10:29: note: expanded from macro 'DIVISION'
#define DIVISION(a, b) ((a) / (b))
                        ~~~~^~~~~
1 warning generated.

The expected behavious is to have the macro included in the plist file,
This is present , but only in the cases when the entry function of the bug contains the MACRO,

for eg. If in the above case, file2.c directly calls the demo_function(10,0) function in its main, the macro definition is found in that case

Is this an expected behaviour or a bug ? Are there any configurations I can tweak to bring the desirable results ?

If this is the expected behaviour, Is there any heuristic behind this decision to only display the macros for first file ?

[llvmbot]

@llvm/issue-subscribers-clang-static-analyzer

Author: None (ankurkraj)

When analyzing a particular file ( using CTU ) . For a macro that is part of the bug-path - the definition of the macro is not included in the plist reports , if the macro is not called in the parent file but in any other intermediate file

Versions:
clang-14,clang-18

Steps to reproduce the behaviour:

Following is a relatively simpler representation of the workflow

file1.c -> calls demo_function(10,0) defined in file2.c
file2.c -> uses macro DIVISION(a,b) in demo_function(int a, int b), macro DIVISION is defined in file3.h
file3.h -> defines DIVISION(a,b) as ((a)/(b))

I have followed the steps for PCH analysis The command I used was

clang --analyze -Xclang -analyzer-config -Xclang experimental-enable-naive-ctu-analysis=true -Xclang -analyzer-config -Xclang ctu-dir=. -Xclang -analyzer-config -Xclang ctu-phase1-inlining=all -Xclang -analyzer-config -Xclang expand-macros=true  -Xclang -analyzer-output=pl
ist-multi-file  -I. file1.c file2.c file3.c

The result displayed on the screen includes the macro definition but the plist files do not contain these value

In file included from file1.c:1:
file2.c:15:50: warning: Division by zero [core.DivideZero]
    printf("Division of %d / %d is: %d\n", a, b, DIVISION(a,b));
                                                 ^~~~~~~~~~~~~
file3.h:10:29: note: expanded from macro 'DIVISION'
#define DIVISION(a, b) ((a) / (b))
                        ~~~~^~~~~
1 warning generated.

The expected behavious is to have the macro included in the plist file,
This is present , but only in the cases when the entry function of the bug contains the MACRO,

for eg. If in the above case, file2.c directly calls the demo_function(10,0) function in its main, the macro definition is found in that case

Is this an expected behaviour or a bug ? Are there any configurations I can tweak to bring the desirable results ?

If this is the expected behaviour, Is there any heuristic behind this decision to only display the macros for first file ?

[EugeneZelenko]

Could you please try 20 or main branch?

[ankurkraj]

Could you please try 20 or main branch?

@EugeneZelenko , I tried with both 20 and the main branch, still exactly the same results.

[ankurkraj]

Hi @EugeneZelenko is there any update on this issue ?

[llvmbot]

@llvm/issue-subscribers-bug

Author: None (ankurkraj)

When analyzing a particular file ( using CTU ) . For a macro that is part of the bug-path - the definition of the macro is not included in the plist reports , if the macro is not called in the parent file but in any other intermediate file

Versions:
clang-14,clang-18,clang-20,main branch binaries

Steps to reproduce the behaviour:

Following is a relatively simpler representation of the workflow

file1.c -> calls demo_function(10,0) defined in file2.c
file2.c -> uses macro DIVISION(a,b) in demo_function(int a, int b), macro DIVISION is defined in file3.h
file3.h -> defines DIVISION(a,b) as ((a)/(b))

I have followed the steps for PCH analysis The command I used was

clang --analyze -Xclang -analyzer-config -Xclang experimental-enable-naive-ctu-analysis=true -Xclang -analyzer-config -Xclang ctu-dir=. -Xclang -analyzer-config -Xclang ctu-phase1-inlining=all -Xclang -analyzer-config -Xclang expand-macros=true  -Xclang -analyzer-output=pl
ist-multi-file  -I. file1.c file2.c file3.c

The result displayed on the screen includes the macro definition but the plist files do not contain these value

In file included from file1.c:1:
file2.c:15:50: warning: Division by zero [core.DivideZero]
    printf("Division of %d / %d is: %d\n", a, b, DIVISION(a,b));
                                                 ^~~~~~~~~~~~~
file3.h:10:29: note: expanded from macro 'DIVISION'
#define DIVISION(a, b) ((a) / (b))
                        ~~~~^~~~~
1 warning generated.

The expected behavious is to have the macro included in the plist file,
This is present , but only in the cases when the entry function of the bug contains the MACRO,

for eg. If in the above case, file2.c directly calls the demo_function(10,0) function in its main, the macro definition is found in that case

Is this an expected behaviour or a bug ? Are there any configurations I can tweak to bring the desirable results ?

If this is the expected behaviour, Is there any heuristic behind this decision to only display the macros for first file ?

[steakhal]

Hi EugeneZelenko is there any update on this issue?

FYI EugeneZelenko just sometimes labels issues and asks questions about reproducibility. EugeneZelenko is unlikely to work on this ticket.

I think the best we could do is to label this issue and wait for someone to investigate and figure out a fix for it.
I was tempted to mark this as a good first issue, where new contributors would try to fix this, but I'm conflicted to pass an issue with macros and CTU to a newcomer.

[ankurkraj]

@steakhal , thanks for clarification. do you mind if i work on it ?

[steakhal]

@steakhal , thanks for clarification. do you mind if i work on it ?

Sure, you can work on it. Thank you for going the extra mile by not only reporting an issue but also trying to fix it. That makes this sustainable.

For anyone else reading this, one does not need to announce that intent or started working on an issue.

If you need some help, dont hesitate and reach out. As for first steps, I'd look at the text diagnostic consumer and the plist diag consumer. Make a debug build and then hop into gdb and set a conditional break point (or patch the code to have an explicit condition for hooking the relevant note with the macro expansion) then see how did you end up there and why does the plist diag consumer miss it.