Svadev mailing list

[John Criswell <criswell AT illinois.edu>]

On 4/3/12 4:54 PM, Jason Newton wrote:
> On Tue, Apr 3, 2012 at 12:51 PM, John 
> Criswell<criswell AT illinois.edu>
>   wrote:
> > On 4/3/12 2:42 PM, Jason Newton wrote:
> > > Hi All,
> > > This email got blackhole'd yesterday, my test email a few minutes ago
> > > worked however.
> > >
> > > make[4]: *** No rule to make target
> > >
> > > `/home/jason/tmp/safecode/build/projects/poolalloc/Release+Asserts/lib/LLVMDataStructure.a',
> > > needed by
> > > `/home/jason/tmp/safecode/build/projects/safecode/Release+Asserts/lib/libLLVMDataStructure.a'.
> > >   Stop.
> > >
> > > I imagine this is a make issue, I'm using 3.82 on opensuse 12.1 x86_64
> > Did you compile poolalloc first per the directions?  If so, did you compile
> > it as a Release+Asserts build?
> Sorry, I did not follow the installation procedure from safecode, I
> assumed it built normally after looking over llvm's build page and how
> subprojects are built as part of the build infrastructure.  This was
> the issue and my mistake for not RTFMing the right manual.

The good news is that SAFECode does build like a regular sub-project for 
LLVM 3.0 (the Makefile in llvm/projects will even build poolalloc before 
safecode).

The problem is that SAFECode depends upon the Automatic Pool Allocation 
project, so just checking out SAFECode isn't enough.

We have also opted not to follow mainline LLVM because, in the past, the 
programming API would experience a lot of churn, and do we'd spend more 
time keeping up with LLVM than we did working on SAFECode.


> > > Btw is there any effort to sync back up with 3.1?  C++ support in 3.0
> > > is kinda lacking and can't really digest boost.  They removed the
> > > UnwindInst and changed the exception mehcanism but the rest may not be
> > > so bad..
> > Not at the moment.  At Illinois, we don't really have anyone who is free to
> > do an update, nor have we seen a pressing need internally yet.
> This is a bit concerning.  Perhaps I'm reacting a bit too early but
> I've seen many of these academic originating useful projects fall by
> the wayside.  Perhaps there should be an effort for inclusion into
> llvm mainline so this useful work doesn't.  Again apologies if I'm
> overreacting but I wish to voice my concern on this, I've seen it
> happen with projects like tcng, openmosix, lti-lib just to name a few.

We would love to get SAFECode (or parts of it) into mainline LLVM.  
However, to get SAFECode into LLVM mainline, we need people actively 
using it.  Without users, people with commit access to LLVM have no 
reason to spend the time reviewing the code on initial inclusion or 
helping maintain it once it goes in.

If you'd like to see SAFECode included in LLVM, please use it and let us 
know if it's helping you find bugs or protect programs.  If you're 
willing to lend a hand, we can always use more help improving the code.  
Many of the optimizations are currently disabled because they need more 
work to be sufficiently robust for users.


> > Is working with LLVM 3.0 problematic for any other SAFECode/SoftBound users
> > out there?
> llvm 3.0 means you cannot use boost c++ libraries which alone sucks...
>   boost 1.49.0  btw.  It seems its all rvalue reference stuff that
> breaks it.  These errors went away in llvm/clang svn.  No real c++11
> support on 3.0.

I see.

> > > ps.  Interesting this grew out of DOD / AFRL / NSF funding :-).
> > Just out of curiosity, why do you think it's interesting?  These are fairly
> > standard funding agencies for CS research.
> This is such a nice tool.  It's not too often one comes across an
> academic research based project that is practically usable.  This is
> government funding being used correctly.

Thank you.

Regards,

John Criswell

> Finally a challenger/alternative to valgrind.
>
> Best Regards,
> Jason
>
> > -- John T.
> > > _______________________________________________
> > > svadev mailing list
> > > svadev AT cs.uiuc.edu
> > > http://lists.cs.uiuc.edu/mailman/listinfo/svadev