5c shouldn't discard the cast, which implies a mask, but you should also
rewrite the code as you suggest.
(it's not just 5c, since it's the same for several other RISCy ones.)


On Thu, Jul 25, 2019 at 12:20 PM Steve Simon <steve@quintile.net> wrote:

> Hi,
>
> I traced a long standing cifs issue to 5c the compiler.
>
> The code in lib9p to validate wstat calls
> uses a neat bit of casting which fails when using 5c.
>
> /sys/src/lib9p/srv.c:658-669
>
> These following lines missfire reporting type or qid change
> when trying to rename files over cifs.
>
>         if((ushort)~r->d.type){
>         ...
>         if((uchar)~r->d.qid.type || .. ){
>
> but rewriting them like this makes the code work as expected:
>
>         if(r->d.type != (ushort)~0){
>         ...
>         if(r->d.qid.type != (uchar)~0 || ... )
>
> Personally I am happy to hack lib9p but the worry is that
> this bug might be affecting other things.
>
> I have no knowledge of the internals of the compilers so
> I am afraid I have made no attempt to dig deeper.
>
> -Steve
>
>