Re: [Libusb-win32-devel] Everything but usb_bulk_read works...

[Bertrik S. <be...@si...>]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Adam Kropelin wrote:
> On Mon, Jun 11, 2007 at 03:47:31PM +0100, Dave Higton wrote:
>>> -----Original Message-----
>>> From: lib...@li... 
>>> [mailto:lib...@li...] On 
>>> Behalf Of Adam Kropelin
>>> Sent: 2007 June 11 15:43
>>> To: lib...@li...
>>> Subject: Re: [Libusb-win32-devel] Everything but 
>>> usb_bulk_read works...
>>>
>>> On Mon, Jun 11, 2007 at 03:23:38PM +0100, Dave Higton wrote:
>>>> Error 22 is an invalid parameter.  Input endpoint addresses
>>>> have to have the top bit set - this is incompatible with the
>>>> Linux version.
>>> It's not incompatible at all. Not setting that bit is a bug on any
>>> platform. It just so happens that Linux lets you get away with it.
>> It depends on your POV.  When you talk to somebody, do you refer to
>> endpoint 0x81 or endpoint 1?  I refer to 1, and I bet you do too.
>> Linux has explicit code to set the top bit so that either is equally
>> valid.
> 
> There is an argument to made for the API to be forgiving; you're
> performing a read, so obviously you must have meant to provide an input
> endpoint address. Since you didn't, the library will fix it up for
> you. That type of API makes me uncomfortable; I don't like an API that
> thinks it knows more about what I want to do than me...but some 
> people like to program that way. There's also an argument to be made 
> for consistency; not every API call can easily do such a fixup for you. 
> See usb_resetep(), for example. That one is even explicitly documented.
> 
> Personally I prefer consistency and exactness: Supply the actual
> endpoint address, as specified in bEndpointAddress. Since we have to do
> that for some API calls already, its best to do it for all of them.

I agree completely.

The libusb API is not exact on what is meant with the 'ep' parameter.
The USB2.0 specification differentiates between an *endpoint number*
and an *endpoint address*, where the endpoint number is a simple 4-bit
number and the endpoint address is the endpoint number combined with
the direction bit.

USB2.0 specification suggests that it is possible for a device to
have both a 0x01 and 0x81 endpoint address (see for example paragraph
5.3.1.2: "Full-speed devices can have additional endpoints only limited
by the protocol definition (i.e., a maximum of 15 additional input
endpoints and 15 additional output endpoints).")
In that case, having libusb setting or clearing the top bit
would amount to converting the endpoint into a different and
possibly unrelated endpoint address, which is not acceptable.

IMO, the Linux implementation is wrong because it allows people to
be sloppy about this.

Regards,
Bertrik
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.2 (MingW32)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org

iD8DBQFGbtyYETD6mlrWxPURAkD1AJ4qwJaQxIcOQ/xB7sa9rpeG4DmZ7wCgvMN0
Bf7kLCiO9cMY8bPJBlbjvZs=
=dh9I
-----END PGP SIGNATURE-----