[cfe-dev] analyzer: invoking a single analyzer from the static analysis tools.

[Michael Katelman]

I've gone down a similar road within the last couple of days. I'm not sure
I'll be able to solve your problem -- I'm just familiarizing myself with
the tools and code myself -- but I did get what you're talking about to
work for the simple checker that I made. Would you mind posting the
entirety of the shell script you made?


On Wed, Feb 26, 2014 at 11:05 AM, Aitor San Juan <aitor.sj at opendeusto.es>wrote:

> Hello,
>
> I have developed a simple checker, just to become familiar with Clang's
> Checker API. I have sucessfully compiled it and installed in package
> alpha.security. For the sake of understanding and tracing the execution
> flow of the checker, I have added some "debug" lines like the following
> (I'm aware this is discouraged):
>
> llvm::outs() << "BEGIN: checkPreStmt" << "\n";
>
> I run scan-build or clang --analyze, as suggested in previous answers, but
> I don't see the messages printed out. The commands I'm running are:
>
> (1) clang --analyze my_file.c
>
> (2) scan-build clang -c my_file.c
>
> Reading the manual, I understand that the reason is that "my" checker is
> not enabled by default, only a few.
>
> So I tried the following command to enable it:
>
> (3) scan-build -enable-checker alpha.security.DCL41C clang -c my_file.c
>
> but again my "debugging" messages don't appear in the standard output. So
> enabling verbose info with "-v -v" in command (3), I can see the complete
> command line run behind the scenes, and scan-build is passing:
>
> '-analyzer-checker' 'alpha.security.DCL41C'
>
> but I've remarked that the checkers enabled by default are being enabled
> like this:
>
>  '-analyzer-checker=core' '-analyzer-checker=unix'
> '-analyzer-checker=deadcode'
> '-analyzer-checker=security.insecureAPI.UncheckedReturn'
>
> So I saved the complete command lie to a shell script, and changed the
> options '-analyzer-checker' 'alpha.security.DCL41C' to
> '-analyzer-checker=alpha.security.DCL41C'
>
> The same problem: I don't see the messages of llvm:outs, so I don't
> whether my checker is being called, or the output of llvm::outs is going to
> some other place I don't know.
>
> My Clang version is a build that dates back to late Dec. 2013. I hope it's
> not a bug :-(
>
> clang version 3.5 (trunk 197976) (llvm/trunk 197973)
> Target: x86_64-unknown-linux-gnu
> Thread model: posix
>
> Any help would he highly appreciated
>
> Date: Tue, 25 Feb 2014 11:37:17 -0800
>
>> From: Michael Katelman <katelman at gmail.com>
>> To: Ted Kremenek <kremenek at apple.com>
>> Cc: cfe-dev at cs.uiuc.edu
>> Subject: Re: [cfe-dev] analyzer: invoking a single analyzer from the
>>         static analysis tools.
>> Message-ID:
>>         <
>> CAAn2fBCGH69fAWP_kXGpwpqjAykj29ps9RfcPbNi9e4foqqKuQ at mail.gmail.com>
>> Content-Type: text/plain; charset="iso-8859-1"
>>
>>
>> Hi Ted,
>>
>> Thanks for the help! I actually have a follow-up question, though. This
>> question came up because I was toying around with developing my own simple
>> checker and wanted to test it out. My first inclination was to do this in
>> isolation from the other checkers which is why I attempted the invocation
>> from my original post, but perhaps you're saying that even for a dev
>> situation like this just run the whole platter of checkers? Or, did you
>> just mean for a general usage scenario where someone isn't adding new
>> checkers etc.?
>>
>> I just ended up using --analyze because the scan-build command I could
>> figure out was somewhat more verbose, needing --use-analyzer=... etc.;
>> but,
>> I was mostly fumbling around trying to get it to work, so I'm sure I'm
>> missing a bunch of different things.
>>
>> Thanks again.
>>
>> -Mike
>>
>>
>> On Tue, Feb 25, 2014 at 11:00 AM, Ted Kremenek <kremenek at apple.com>
>> wrote:
>>
>> > Hi Michael,
>> >
>> > It's not recommended to run the low-level driver like this.  You can
>> just
>> > use scan-build, for example:
>> >
>> >   $ scan-build clang -c /tmp/main.c
>> >
>> > or more generally
>> >
>> >   $ scan-build <compiler line>
>> >
>> > You can also use:
>> >
>> >   $ clang --analyze /tmp/main.c
>>
>> >
>> > directly.  That's somewhat discouraged because the long-term idea is
>> that
>> > the static analyzer supports global analysis.  The intention is that you
>> > can declare a set of files to analyze and they get analyzed together,
>> > whereas the latter line is clearly just analyzing a particular file
>> using
>> > clang.
>> >
>> > Cheers,
>> > Ted
>> >
>> > On Feb 21, 2014, at 2:22 PM, Michael Katelman <katelman at gmail.com>
>> wrote:
>> >
>> > > I was wondering if someone might be able to help me with cleanly
>> > invoking a single analyzer from the static analysis tools.
>> > >
>> > > I am not sure what I need to do (or, should be doing instead) in a
>> > situation like the one below where I've got a header like stdio.h
>> included
>> > (--analyze figures it out, but then it appears that I lose the ability
>> to
>> > apply a single checker) :
>> > >
>> > > %  ./Debug+Asserts/bin/clang -cc1 -analyze
>> > -analyzer-checker=core.DivideZero ./tmp/main.c
>> > >
>> > > ./tmp/main.c:1:10: fatal error: 'stdio.h' file not found
>> > > #include <stdio.h>
>> > >          ^
>> > > 1 error generated.
>> > >
>> > >  % cat ./tmp /main.c
>> > >
>> > > #include <stdio.h>
>> > >
>> > > int main( int argc, char** argv){
>> > >   int x = 1;
>> > >   int y = 0;
>> > >
>> > >   printf("%d\n", x / y);
>> > >
>> > >   return  0;
>> > > }
>> > >
>> > > Thanks!
>> > >
>> > > -Mike
>> > > _______________________________________________
>> > > cfe-dev mailing list
>> > > cfe-dev at cs.uiuc.edu
>> > > http://lists.cs.uiuc.edu/mailman/listinfo/cfe-dev
>> >
>> >
>>
>
> _______________________________________________
> cfe-dev mailing list
> cfe-dev at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/cfe-dev
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-dev/attachments/20140226/fbb3d15c/attachment.html>