[cfe-dev] clang and clang-sa do not detect uninitialized variable

[Nathan Sidwell via cfe-dev]

On 5/4/21 10:30 AM, Ali Shuja Siddiqui (alissidd) wrote:
> Hi Nathan,
> 
> Thanks for the reply. Can you please elaborate on how to let clang-sa to 
> learn about printf? I tried using CTU analysis [1], but I’m unsure how 
> to add mappings for libc functions.

Sorry, I don't know.  I don't know if clang/llvm have support for 
knowing about various builtin functions the standard library provides. 
It must have some (like say abs and sqrt).  If it knew about printf, it 
could provide diagnostics when the variadic args do not match the format 
string, for instance.

nathan

> Thanks,
> 
> Ali
> 
> [1] https://clang.llvm.org/docs/analyzer/user-docs/CrossTranslationUnit.html
> 
> *From: *Nathan Sidwell <nathanmsidwell at gmail.com> on behalf of Nathan 
> Sidwell <nathan at acm.org>
> *Date: *Friday, April 30, 2021 at 9:00 AM
> *To: *Ali Shuja Siddiqui (alissidd) <alissidd at cisco.com>, 
> cfe-dev at lists.llvm.org <cfe-dev at lists.llvm.org>
> *Subject: *Re: [cfe-dev] clang and clang-sa do not detect uninitialized 
> variable
> 
> On 4/29/21 4:41 PM, Ali Shuja Siddiqui (alissidd) via cfe-dev wrote:
>  > Hello,
>  >
>  > With the following code
>  >
>  > //-------------------------
>  >
>  > #include <stdio.h>
>  >
>  > extern int t;
>  >
>  > void use_b (int *b){
>  >
>  >      printf("%p\n",b);
>  >
>  > }
>  >
>  > void func(){
>  >
>  >      int b;
>  >
>  >      use_b(&b);
>  >
>  >      if (b)
>  >
>  >          b+=33;
>  >
>  > }
>  >
>  > //---------------------------
>  >
>  > Running clang -Wuninitialized or running clang –analyze, I don’t see any
>  > warning for uninitialized variables. However, if I change the code to:
>  >
>  > //----------------------
>  >
>  > #include <stdio.h>
>  >
>  > extern int t;
>  >
>  > void use_b (int *b){
>  >
>  >      if (t==5)
>  >
>  >          return;
>  >
>  >      printf("%p\n",b);
>  >
>  > }
>  >
>  > void func(){
>  >
>  >      int b;
>  >
>  >      use_b(&b);
>  >
>  >      if (b)
>  >
>  >          b+=33;
>  >
>  > }
>  >
>  > //--------------------
>  >
>  > I do see this warning with the static analyzer:
>  >
>  > sa_try.c:13:9: warning: Branch condition evaluates to a garbage value
>  > [core.uninitialized.Branch]
>  >
>  >      if (b)
>  >
>  >          ^
>  >
>  > 1 warning generated.
>  >
>  > My question is why am I not getting any warning for the first case? Is
>  > it being considered that printf is updating the value of b in some way?
> 
> Presuming it has no knowledge of printf's semantics, then yes.  that
> function could write through the pointer (and indeed would, if 'b' was
> an output parameter).
> 
> It could learn that printf doesn't do that (except for %n).
> 
> nathan
> 
> 
> nathan
> -- 
> Nathan Sidwell
> 


-- 
Nathan Sidwell