Bogus warning with semi-initialised struct variables and static inline functions

[llvmbot]

Bugzilla Link	9628
Resolution	FIXED
Resolved on	Mar 23, 2012 00:11
Version	2.9
OS	MacOS X
Reporter	LLVM Bugzilla Contributor
CC	@tkremenek

Extended Description

Consider the following program:

#include <stdio.h>

struct Foo { int x; int y; };

static inline int foogetx(struct Foo foo) { return foo.x; }

int main() {
struct Foo bar;
bar.x = 5;
printf("%d\n", foogetx(bar));
return 0;
}

$ clang --version
Apple clang version 2.0 (tags/Apple/clang-137) (based on LLVM 2.9svn)
Target: x86_64-apple-darwin10
Thread model: posix

$ clang --analyze a.c
a.c:10:20: warning: Passed-by-value struct argument contains uninitialized data (e.g., field: 'y')
printf("%d\n", foogetx(bar));
^ ~~~
1 warning generated.

This is not an actual issue since that particular member (.x) has been initialised. It seems that this should be fixed with full interprocedural analysis, but that might not be necessary considering that foogetx() is a static inline function.

At any rate, using bar.x instead of foogetx(bar) is a simple enough workaround. This situation happens with some functions in Apple's Foundation library, too:

NSRect frame;
frame.size = NSZeroSize;
NSHeight(frame) // static analyser warning
frame.size.height // no static analyser warning

[llvmbot]

assigned to @tkremenek

[tkremenek]

Consider the following program:

#include <stdio.h>

struct Foo { int x; int y; };

static inline int foogetx(struct Foo foo) { return foo.x; }

int main() {
struct Foo bar;
bar.x = 5;
printf("%d\n", foogetx(bar));
return 0;
}

$ clang --version
Apple clang version 2.0 (tags/Apple/clang-137) (based on LLVM 2.9svn)
Target: x86_64-apple-darwin10
Thread model: posix

$ clang --analyze a.c
a.c:10:20: warning: Passed-by-value struct argument contains uninitialized data
(e.g., field: 'y')
printf("%d\n", foogetx(bar));
^ ~~~
1 warning generated.

This is not an actual issue since that particular member (.x) has been
initialised. It seems that this should be fixed with full interprocedural
analysis, but that might not be necessary considering that foogetx() is a
static inline function.

You are right this is an IPA issue. We plan on doing basic IPA sooner-rather-than-later that includes the bodies of static inline functions.

At any rate, using bar.x instead of foogetx(bar) is a simple enough workaround.
This situation happens with some functions in Apple's Foundation library, too:

NSRect frame;
frame.size = NSZeroSize;
NSHeight(frame) // static analyser warning
frame.size.height // no static analyser warning

[tkremenek]

cloned to rdar://problem/9233046

[tkremenek]

The analyzer now handles this correctly.