On 7/10/19 12:12 PM, Hans Verkuil wrote:
> Hi Dafna,
>
> On 7/1/19 12:30 PM, Dafna Hirschfeld wrote:
>> commit 84219e2b5d013709ee5259621715966c46eec746
>> Author: Hans Verkuil <hverkuil-cisco@xxxxxxxxx>
>> Date: Sat Mar 30 14:16:25 2019 +0100
>>
>> v4l2-ctl: get fmt after source change
>>
>> When a source change event arrives during decoding get the new
>> format at that point instead of after restarting streaming.
>>
>> If there is another source change queued up, then when you call
>> streamon for CAPTURE again it might send the new source change
>> event and update the format for that one, so reading the format
>> after streamon might give the wrong format.
>>
>> Signed-off-by: Hans Verkuil <hverkuil-cisco@xxxxxxxxx>
>>
>> diff --git a/utils/v4l2-ctl/v4l2-ctl-streaming.cpp
>> b/utils/v4l2-ctl/v4l2-ctl-streaming.cpp
>> index bb656584..3695a027 100644
>> --- a/utils/v4l2-ctl/v4l2-ctl-streaming.cpp
>> +++ b/utils/v4l2-ctl/v4l2-ctl-streaming.cpp
>> @@ -2363,12 +2363,11 @@ static void stateful_m2m(cv4l_fd &fd,
>> cv4l_queue &in, cv4l_queue &out,
>> if (in_source_change_event) {
>> in_source_change_event = false;
>> last_buffer = false;
>> + fd.g_fmt(fmt_in, in.g_type());
>> if (capture_setup(fd, in, exp_fd_p))
>> return;
>> fps_ts[CAP].reset();
>> fps_ts[OUT].reset();
>> - fd.g_fmt(fmt_out, out.g_type());
>> - fd.g_fmt(fmt_in, in.g_type());
>> Removing those lines cause inconsistency when the user send the wanted
>> capture pixel format when decoding with the `v4l2-ctl` command. In
>> this case the value of `-v pixelformat=...` is updated in the kernel
>> with the capture_setup function but it is not updated in the fmt_in
>> variable and so the command will try to save the decoded video in a
>> different format from what is configured in the kernel.
>>
>> cap_streaming = true;
>> } else {
>> break;
>>
>>
>> Dafna
>>
>
> It looks like you are right. Can you try this v4l2-ctl patch?
>
> Regards,
>
> Hans
>
> [PATCH] v4l2-ctl: let capture_setup return the updated format
>
> Just getting the new format after a source change event isn't
> enough in the case that -v option was specified by the user.
>
> Instead let capture_setup optionally return the new format after
> any '-v' changes are applied.
>
> Signed-off-by: Hans Verkuil <hverkuil-cisco@xxxxxxxxx>
> Reported-by: Dafna Hirschfeld <dafna3@xxxxxxxxx>
I pushed this patch since your test with 'blog.sh' now works with this
patch applied.
Regards,
Hans
> ---
> diff --git a/utils/v4l2-ctl/v4l2-ctl-streaming.cpp b/utils/v4l2-ctl/v4l2-ctl-streaming.cpp
> index 28b2b3b9..470fa766 100644
> --- a/utils/v4l2-ctl/v4l2-ctl-streaming.cpp
> +++ b/utils/v4l2-ctl/v4l2-ctl-streaming.cpp
> @@ -2146,7 +2146,7 @@ enum stream_type {
> OUT,
> };
>
> -static int capture_setup(cv4l_fd &fd, cv4l_queue &in, cv4l_fd *exp_fd)
> +static int capture_setup(cv4l_fd &fd, cv4l_queue &in, cv4l_fd *exp_fd, cv4l_fmt *new_fmt = NULL)
> {
> if (fd.streamoff(in.g_type())) {
> fprintf(stderr, "%s: fd.streamoff error\n", __func__);
> @@ -2168,6 +2168,10 @@ static int capture_setup(cv4l_fd &fd, cv4l_queue &in, cv4l_fd *exp_fd)
> return -1;
> }
> fd.s_fmt(fmt, in.g_type());
> + if (new_fmt)
> + *new_fmt = fmt;
> + } else if (new_fmt) {
> + fd.g_fmt(*new_fmt, in.g_type());
> }
> get_cap_compose_rect(fd);
>
> @@ -2367,8 +2371,7 @@ static void stateful_m2m(cv4l_fd &fd, cv4l_queue &in, cv4l_queue &out,
> if (in_source_change_event) {
> in_source_change_event = false;
> last_buffer = false;
> - fd.g_fmt(fmt_in, in.g_type());
> - if (capture_setup(fd, in, exp_fd_p))
> + if (capture_setup(fd, in, exp_fd_p, &fmt_in))
> return;
> fps_ts[CAP].reset();
> fps_ts[OUT].reset();
>
