On Fri, Oct 24, 2014 at 5:23 AM, Joerg Sonnenberger <[email protected] > wrote:
> On Thu, Oct 23, 2014 at 01:35:37PM -0700, Alexey Samsonov wrote: > > On Thu, Oct 23, 2014 at 12:04 PM, Joerg Sonnenberger < > > [email protected]> wrote: > > > > > On Thu, Oct 23, 2014 at 12:46:10AM -0000, Alexey Samsonov wrote: > > > > "-nodefaultlibs" is told to link only the libraries explicitly > provided > > > > by the user, and providing "-fsanitize=address" is a clear > indication of > > > > intention to link with ASan runtime. > > > > > > Is it? I would expect such use for free-standing for example and I can > > > think of use cases for ASAN that do not involve the normal userland > > > libraries. As such, I believe this needs more thought... > > > > > > > Please elaborate what example / use case you have in mind. > > Consider that you want to debug a boot loader in userland. You provide a > small glue library that emulates the normal hardware access via direct > system calls. You don't use libc at all. You still have to provide your > own ASAN implementation of course, but now keeping -fsanitize=address on > the command line breaks things badly. > In this case you should compile your code with -fsanitize=address (so that it's instrumented, and has callbacks into ASan runtime), and link it with your custom ASan implementation instead of relying on the driver behavior for -fsanitize=address linking. That is, you may have to split compilation and linking in two separate steps, but that doesn't seem terrible. > > Joerg > _______________________________________________ > cfe-commits mailing list > [email protected] > http://lists.cs.uiuc.edu/mailman/listinfo/cfe-commits > -- Alexey Samsonov [email protected]
_______________________________________________ cfe-commits mailing list [email protected] http://lists.cs.uiuc.edu/mailman/listinfo/cfe-commits
