Re: [Fwd: Re: Bug#196085: Device files are not accepted under the Hurd]

[Neal H. Walfield]

> > Under the Hurd, when device file is passed as an argument to "parted",
> > Segmentation fault occurs.  More detailed report can be found in
> > http://mail.gnu.org/archive/html/bug-hurd/2002-10/msg00036.html
> > The reason is using store_typed_open instead of store_open.  More
> > information can be found in $HURD/libstore/store.h (or
> > /include/hurd/store.h).  Basicly, store_typed_open expects something
> > like "device:hd0", while store_open expects something like "/dev/hd0".
> > Needless to say, the latter is what everyone would try.  The first style
> > is equivalent of (MAJOR,MINOR) pair in *nix, and the second style is
> > equivalent of block device file name.
> 
> It seems that the bug is in libstore, not in parted.  When
> store_typed_open doesn't find ":" character, it calls store_open.
> Unfortunately, it does it incorrectly, and, as a result, store_open is
> not called as intended.  There is second bug somewhere in
> store_find_class, called by store_typed_open.  Some dlsym magic is used
> to find store class.  I'm investigating what exactly goes wrong
> (resulting in SIGSEGV) when the store class is not found.  A patch will
> follow if I find out.

I found this bug once and came to the same conclusion as you have.  I
believe that I may have even attempted to fix it, however, that code
is long gone.

> I'm reassigning this bug to the hurd package.

This is correct.