[cfe-dev] CLang 2.1 doesn't catch assert within if

[Ted Kremenek]

The analyzer doesn't find this, but it would be an interesting check to write.  Conditional code for branches are just asking for trouble.

On Oct 7, 2011, at 10:02 PM, Don Quixote de la Mancha wrote:

> I've been looking around for an analysis I could add to clang, but I
> used Xcode 4.1's clang to test this, which only has CLang 2.1.  Could
> someone please check whether the current SVN catches this?  I would
> check myself but I am having some trouble getting the current SVN to
> build.  It's not the source code but some arcane problems with the
> configure shell scripts.  If I can't figure out my configure problem
> soon I will ask about it here.
> 
> If the current SVN does not catch this, I'd like to do the work to add
> the analysis myself.  I've been following the list for a couple weeks
> and understand now what would be expected of my code for it to be
> accepted.  I will open a Bugzilla bug to track my progress.
> 
> You NEVER EVER want to do the following:
> 
> if ( condition )
>  assert( must_be_true );
> 
> following_line();
> 
> If NDEBUG is defined as for a release or profile build, you'll get the
> following in your preprocessed code, which will be very hard to debug:
> 
> if ( condition )
>   following_line();
> 
> If you really must insist on putting an assert as the only line within
> an if statement, you MUST place it in a Basic Block:
> 
> if ( condition ){
>  assert( must_be_true );
> }
> 
> Even better is to make the if statement part of the assert:
> 
> assert( condition && must_be_true );
> 
> Back in the day I wrote a new library to replace my client's old one.
> To test that my new library got the same results as his old one, he
> asked me to link his executable with both his old and my new library,
> and to compare the results of all the calls to each library.  I don't
> recall why anymore but I actually made the mistake, then spent all day
> long tracking down my bug, as the executables and both libraries were
> very complex.
> 
> I'm running Mac OS X Snow Leopard 10.6.8 on a MacBook Pro, Model
> Identifier MacBookPro1,1 (32-bit Core Duo, NOT 64-bit Core 2 Duo).
> 
> $ clang --version
> Apple clang version 2.1 (tags/Apple/clang-163.7.1) (based on LLVM 3.0svn)
> Target: i386-apple-darwin10.8.0
> Thread model: posix
> 
> $ gcc --version
> i686-apple-darwin10-llvm-gcc-4.2 (GCC) 4.2.1 (Based on Apple Inc.
> build 5658) (LLVM build 2335.15.00)
> 
> $ gcc --pedantic -Wall -c assert_within_if.c
> 
> This compile succeeds without any warnings.
> 
> $ clang --analyze assert_within_if.c
> 
> CLang doesn't print anything to the terminal.  assert_within_if.plist
> doesn't contain any diagnostics.
> 
> ====== Cut Here ======
> /* assert_within_if.c */
> #include <stdlib.h>
> #include <assert.h>
> 
> int assert_within_if( int flag, int  *pointer );
> 
> int assert_within_if( int flag, int  *pointer )
> {
> 	int result = 0;
> 
> 	if ( flag )
> 		assert( NULL != pointer );
> 
> 	result = *pointer;
> 
> 	return result;
> }
> 
> ====== Cut Here =====
> 
> Ever Faithful,
> 
> Don Quixote
> -- 
> Don Quixote de la Mancha
> Dulcinea Technologies Corporation
> Software of Elegance and Beauty
> http://www.dulcineatech.com
> quixote at dulcineatech.com
> _______________________________________________
> cfe-dev mailing list
> cfe-dev at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/cfe-dev