bug#14917: Missing range check in fxcopy-bit can give SIGABRT

[Andy Wingo <wingo@pobox.com>]

Howdy :)

Three years later, this is now fixed and will be in 2.1.4.  I think
we'll cherry-pick it back to 2.0.12 too.

Cheers,

Andy

On Sat 20 Jul 2013 08:57, Göran Weinholt <goran@weinholt.se> writes:

> Hello schemers,
>
> the fxcopy-bit procedure from (rnrs) is missing some range checks. It
> can return a non-fixnum:
>
> scheme@(guile-user)> (import (rnrs))
> scheme@(guile-user)> (fxcopy-bit 0 (fixnum-width) 1)
> $1 = 9223372036854775808
>
> It can also crash the guile process, which is somewhat surprising for a
> fixnum procedure:
>
> scheme@(guile-user)> (import (rnrs))
> scheme@(guile-user)> (fxcopy-bit 0 100000000000 0)
> FATAL: memory error in realloc
> Aborted
>
> Here's an alternative error message:
>
> scheme@(guile-user)> (import (rnrs))
> scheme@(guile-user)> (fxcopy-bit 0 1000000000000 0)
> gmp: overflow in mpz type
> Aborted
>
> Other implementations of fxcopy-bit usually check that the third
> argument is 0 or 1, but I'm not sure that is required.
>
> There's also a bitwise-copy-bit procedure that is similary affected.
> Tested with Guile 2.0.9.40-824b-dirty on an amd64 system.
>
> Regards,