Re: [PATCHSET] FUSE: extend FUSE to support more operations, take #2

[Tejun Heo <tj@kernel.org>]

Miklos Szeredi wrote:
> On Fri, 21 Nov 2008, Tejun Heo wrote:
>> Miklos Szeredi wrote:
>>> On Fri, 21 Nov 2008, Tejun Heo wrote:
>>>>> I removed ->unrestricted_ioctl() and associated code because it really
>>>>> doesn't make any sense: the high level lib won't be used for CUSE
>>>>> stuff, otherwise unrestrited ioctls are not allowed (and the interface
>>>>> is rather horrible anyway).
>>>> Well, CUSE highlevel interface piggy backs on FUSE so it requires
>>>> unrestricted_ioctl() there for it and ossp does use it.
>>> I thought it uses the lowlevel interface.  Why doesn't it do that?
>> Well, because it's simpler that way and people would be more used to it?
>>  It's just easier when you implement a method which returns something
>> and looks similar to the respective file operation.
> 
> Ah, that.  Yeah, it's more intuitive, but that comes at a price.  I'm
> not sure that for CUSE it's worth it.  As I said the biggest feature
> is having paths, the others are not that important (like allocating a
> buffer for read, that's really not too complex to do in each CUSE
> driver).
> 
>>> For CUSE there's really no point in going through high level
>>> interface, since there's just one file involved, so the path name
>>> generation (the main feature of the highlevel lib) doesn't make any
>>> sense.
>> Well, the choice was mostly for convenience as there also are a few
>> places where high level interface wraps things better a bit.  Converting
>> wouldn't be difficult.  Do you think it's important?  I think keeping
>> things as parallel to FUSE as possible is more important.
> 
> I wouldn't care very much, if it weren't for that horrid
> unrestricted_ioctl().  Not your fault, the interface is just not well
> suited to that.

If you want drop highlevel CUSE interface, that's fine with me.  After
all, the complexity difference isn't that big for CUSE anyway.

Thanks.

-- 
tejun