[cfe-dev] Scan-Build question

[Artem Dergachev via cfe-dev]

Sounds pretty weird, i don't have any immediate ideas on what may cause 
this. If running scan-build over the same projects in different 
environments yields different results, then my first guess would be to 
try to see if the same files get compiled during build, and then if they 
do actually get analyzed after that.

Have a look at the standard output - are there compile errors? Only 
what's compiled can be analyzed. Make sure you're doing a clean rebuild: 
if things aren't built, they aren't analyzed.

Scan-build's method is very simple: it changes the CC= and CXX= 
environment variables so that they point to its own "compiler" 
(ccc-analyzer) which is a wrapper that launches both the original 
compiler and a clang instance that performs the analysis. If something 
messes with environment variables, it may be the problem. You can see 
which files were compiled/analyzed by adding a print at the beginning of 
the script (i don't think there's a more comfy way to debug that, but 
ccc-analyzer is a simple perl script, so it shouldn't be hard to modify).

Also see if you're using a different clang - it's unlikely, but it may 
cause differences in results simply due to updates in the Static Analyzer.

Additionally, debug vs. release builds of the project you're analyzing 
(and, similarly, builds with different ./configure flags) may yield 
different results due to different macro expansions (debug is preferred 
because it lets Static Analyzer to take advantage of assertions within 
the code).



On 2/15/19 12:58 PM, Bradley Jackson via cfe-dev wrote:
> Hello all,
>
> I am a current student at University of Pennsylvania and am 
> experimenting with scan-build as part of a senior project. I just had 
> a couple questions regarding some behavior I didn't quite understand. 
> Does the output of scan-build differ depending on the user 
> permissions, or whether it is created as a child process? Are there 
> any flags that determine how sensitive scan-build is to detecting 
> bugs? For reference, running scan build myself versus through 
> executing a shell command on a Node.js server on that same machine 
> yields different outputs, with the server outputting many more. I 
> cannot find any configuration file, and all the bugs reported belong 
> to the set of checkers enabled by default. I would love some insight 
> into why this could be happening.
>
> Best regards,
>
> Bradley Jackson
> _______________________________________________
> cfe-dev mailing list
> cfe-dev at lists.llvm.org
> https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-dev