This is the mail archive of the gcc-help@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: Installing gcc 4.7.2 on Solaris 10 - SPARC platform


> If running through testsuite is the acid test of certification, then I'd
> also not want to package and release unless thoroughly tested. Still I 
> have
> the build directory lying on my system. I haven't run a gcc testsuite
> earlier. Can you please let me know how to run a testsuite and post the
> result back? Is it downloadable from somewhere? How to run it? Apart from
> console log, does it produce any other log which I can post back?
> 

Pretty well documented here : 

    http://gcc.gnu.org/install/test.html

Essentially, and this is a belief in the absence of proof, software should pass
some sort of test harness before we trust it at any level. Even if a software
component may pass its own test suite with glorious flying colors we should
still distrust it and keep a watchful eye upon it.  Having said that, we know
as a fact that really exceptional software ( ISC Bind for example ) does exist
and it does pass a rigorous software test suite and yet in spite of this we 
still see horrendous bugs discovered that reveal weaknesses : 

http://web.nvd.nist.gov/view/vuln/detail?vulnId=CVE-2012-1667

Does this mean that we should not trust ISC Bind to run the entire name
server infrastructure of the planet?  In no way would I agree with such
a stance.  Furthermore should we truct a compiler that can not pass its
own testsuite? Well no, we should not. Even a perfect compiler can be 
used to cronstruct a security violation.  There is a famouse lecture on
this by Ken Thompson in his 1983 Turing Award speech named "Reflections
on Trusting Trust" which revealed how a virtually invisible back door security
hole could be inserted into a C compiler.  Please look that one up. 

What I am saying is that if you buy a car there should be some paper
somewhere that says you can trust it in at least some way to not blow
up nor take a sudden lurch to the left and kill people. We have safety 
standards ( these days ) to enforece this.  In the same way I feel strongly
that a critical piece of software used worldwide to produce other software
pieces should come with a statement of compliance to some tests. 

In short, don't release a compiler that doesn't have a publicly posted test
report somewhere. It bothers me to no end that I can not find any such 
piece of paper from Red Hat nor Oracle to tell me that their GCC compiler
in their distribution does anything correctly. God forbid we even glance
at Oracle Studio 12 ( which was Sun Studio which was Forte Studio ..)
as closed source stuff rarely comes with anything but the words "trust us".

Dennis 

ps: I recently posted changes to nearly every Makefile.am in the flex project
to get it to pass all but two of its own tests and yet still I trust flex because 
I know where it fails and why.  Sometimes you just have to look at the source.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]