[Lustre-discuss] [HPDD-discuss] Clang plugin for identifying un-used code.

Henwood, Richard richard.henwood at intel.com
Thu Jul 24 17:11:15 PDT 2014


Thanks for this feedback Chris, my comments in-line.:

On Thu, 2014-07-24 at 15:00 -0700, Christopher J. Morrone wrote:
> Richard,
> 
> Some notes for improvements needed:
> 
> The URL under the git clone line on the wiki page needs fixing.  It is 
> specific to you, Richard.
> 

fixed.

> Next, I would like to see this wiki page on the opensfs wiki instead of 
> (or in addition to) the Intel wiki.
> 

Sound good to me.

> The information about how to build and use lustre-static-analysis should 
> be added to the lustre-static-analysis code tree.
> 

Agreed. I personally favor input from users who use the tool to identify
unused code in Lustre to drive this process.

> 
> The _new_ code is GPLv2, but the part(s) taken from clang/llvm are under 
> the University of Illinois/NCSA Open Source License, which I think is 
> basically just the BSD.
> 
> Why did you choose GPLv2 when the tool you based it on is BSDish?
> 

I have considered the mixing of Licenses. We did the work under
contract. The contract specified GPL2 for code. GPL2 and Clang BSD
license are compatible. I sided with the contract.

> We need a bit more of documentation of exactly what the two plugins 
> actually do.
> 

My suggestion here is that we keep the build and usage instructions in a
single readily available format until more feedback arrives. At that
point we can consider a suitable README in the code or similar.

> I would like to see lustre-static-analysis given a bit fuller of a build 
> system.  Ideally it would use autotools.  At a minimum it would have a 
> top-level make file.
> 

As part of Removal of Dead Code, this plugin proved valuable -- but it
must be said that with the conclusion of that project, we cannot commit
to on-going maintenance. The license, of course, allows anyone to
modify, distribute and provide that role.

If there is general interest in this tool, I guess the best place is to
voice that here and at the TWG?

> Next, it seems to me like the method of invoking the tool is a little 
> strange.  I think I would prefer to have a script that behaves like a 
> compiler (it calls clang for analysis, and then passes the full command 
> line to gcc for the real compilation).  Right now we have a psuedo-make 
> command.
> 
> So I would like the lustre build commands to look like:
> 
> $ sh autogen.sh
> $ ./configure CC=my_fine_lustre_clang_wrapper
> $ make
> 
> That just seems a little more obvious to me then the way it is currently 
> invoked.  Granted, I might be missing something.
> 

Probably not. Our project goal was to delete unused code from Lustre
[1]. We developed this tool enough to get us to the goal. There is more
unused code, and I think further use of this tool will drive the next
enhancements.

best regards,
Richard

1.
http://wiki.opensfs.org/images/f/f4/RemovalOfDeadCode_ScopeStatement.pdf

-- 
Richard.Henwood at intel.com
Intel High Performance Data Division


More information about the lustre-discuss mailing list