[PATCH] qcap: fix recent regression that limited Wine's v4l2 support to devices that read()

[Zebediah Figura]

On 4/20/19 10:56 AM, Damjan Jovanovic wrote:
> 
> 
> On Sat, Apr 20, 2019 at 5:44 PM Zebediah Figura <z.figura12 at gmail.com 
> <mailto:z.figura12 at gmail.com>> wrote:
> 
>     On 4/20/19 10:31 AM, Damjan Jovanovic wrote:
>      >
>      >
>      > On Sat, Apr 20, 2019 at 4:52 PM Zebediah Figura
>     <z.figura12 at gmail.com <mailto:z.figura12 at gmail.com>
>      > <mailto:z.figura12 at gmail.com <mailto:z.figura12 at gmail.com>>> wrote:
>      >
>      >     On 4/20/19 8:15 AM, Damjan Jovanovic wrote:
>      >      > Recent changes to qcap/v4l.c resulted in the removal of mmap()
>      >      > for v4l2 devices, but read() is optional for device
>     drivers and
>      >      > some don't support it. This isn't a major problem with the
>      >      > presense of libv4l2 though, as it can emulate read() on top of
>      >      > mmap(). However the code checks whether the device can read(),
>      >      > and if not, doesn't even try to use it, even though it can.
>      >      >
>      >      > Fix this by only warning that read() is being emulated, and
>      >      > continuing with libv4l2 if available.
>      >      >
>      >      > Also clarifies the logging.
>      >      >
>      >      > Signed-off-by: Damjan Jovanovic <damjan.jov at gmail.com
>     <mailto:damjan.jov at gmail.com>
>      >     <mailto:damjan.jov at gmail.com <mailto:damjan.jov at gmail.com>>>
>      >      > ---
>      >      >   dlls/qcap/v4l.c | 13 ++++++++-----
>      >      >   1 file changed, 8 insertions(+), 5 deletions(-)
>      >      >
>      >      >
>      >      >
>      >
>      >     This doesn't seem right; libv4l2 should massage the result of
>      >     VIDIOC_QUERYCAP to include V4L2_CAP_READWRITE. See
>      >   
>       <https://git.linuxtv.org/v4l-utils.git/tree/lib/libv4l2/libv4l2.c#n1210>.
>      >
>      >
>      > On FreeBSD 11.2 with libv4l-1.6.3_4, that is certainly NOT the
>     case. It
>      > could be a newer or not yet released feature.
>      >
>      >
>      >
> 
>     It's been the case since libv4l2 0.5:
>     <https://git.linuxtv.org/v4l-utils.git/commit/lib/libv4l2/libv4l2.c?id=0215f2ac08c3ad0dc66ad8036f4e186a5e8f56d6>
> 
>     If FreeBSD is using 1.6.3, then I think something else is wrong. Are
>     you
>     sure that libv4l2 is actually being loaded?
> 
> 
> int v4l2_ioctl(int fd, unsigned long int request, ...)
> {
> ...
>          if (devices[index].convert == NULL)
>                  goto no_capture_request;
> 
> <your code nippet from above>
> 
> no_capture_request:
> 
> 
> 
> So the code that would set that flag, is skipped if the .convert field 
> is NULL.
> 
> When is it NULL?
> 
> 
> 

If my reading of the code is correct, only when v4l2_fd_open() is called 
manually with V4L2_DISABLE_CONVERSION. Obviously we don't do that.

Perhaps setting LIBV4L2_LOG_FILENAME might give some hints as to what's 
going wrong?