[BUG] CodeQL Fails to Process --include=<arg> in clang Commands

[cradiator]

Summary

CodeQL does not handle the --include=<arg> format correctly when used in Clang commands.

In Clang, the --include option instructs the compiler to include macros from a file before parsing. It can be specified in three formats: -include<file>, --include<file>, and --include=<arg>. However, CodeQL fails to correctly interpret the --include=<arg> format.

https://clang.llvm.org/docs/ClangCommandLineReference.html

Steps to Reproduce

codeql database create --language=cpp --command="clang --include=header.h -c main.c -o main.o" --overwrite ./db

Expected Behavior

The database should contain information from main.c.

Actual Behavior

The resulting database does not include information from main.c.

Error Log

The extractor's error log shows the following warning:

linux64/extractor ... --preinclude '=header.h'

...
[E 00:10:05 3930451] Warning[extractor-c++]: In construct_text_message: Command-line error: cannot open source file "=header.h"

This indicates that the extractor incorrectly interprets =header.h as the file name, rather than recognizing header.h as the intended include file.

[aibaars]

You are right. Even though clang's documentation does not mention the = sign in its man pages, it is allowed

       -include <filename>
              Adds an implicit #include into the predefines buffer which is read before the source file is preprocessed.

I'll let the team know. As a workaround, could you try leaving out the = sign in the command?

Thanks for reporting!

[jketema]

Hi @cradiator,

Could you share the build-tracer.log file created during database creation? It's not clear to me whether clang is ran with an integrated cc1 in your case, and the log file should clear that up (any fix will depend on that).

[intrigus-lgtm]

You are right. Even though clang's documentation does not mention the = sign in its man pages, it is allowed

       -include <filename>
              Adds an implicit #include into the predefines buffer which is read before the source file is preprocessed.

I guess the man page is just outdated as the online documentation lists it:
https://clang.llvm.org/docs/ClangCommandLineReference.html#include-path-management

[cradiator]

@aibaars Yes, I tried "--includeheader.h" and it works, the output database contains main.c's information.

@jketema Here is is the build-tracer.log for your reference.
build-tracer.log