[FFmpeg-devel] configure: disable vulkan if min version insufficient

Fixes: https://trac.ffmpeg.org/ticket/10596
---
 configure | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

On 29.09.2023 15:52, Tristan Matthews wrote:
> Fixes: https://trac.ffmpeg.org/ticket/10596
> ---
>   configure | 3 ++-
>   1 file changed, 2 insertions(+), 1 deletion(-)
> 
> diff --git a/configure b/configure
> index 20db1801ed..50ba6f772f 100755
> --- a/configure
> +++ b/configure
> @@ -7154,7 +7154,8 @@ enabled crystalhd && check_lib crystalhd "stdint.h libcrystalhd/libcrystalhd_if.
>   
>   if enabled vulkan; then
>       check_pkg_config_header_only vulkan "vulkan >= 1.3.255" "vulkan/vulkan.h" "defined VK_VERSION_1_3" ||
> -        check_cpp_condition vulkan "vulkan/vulkan.h" "defined(VK_VERSION_1_4) || (defined(VK_VERSION_1_3) && VK_HEADER_VERSION >= 255)"
> +        check_cpp_condition vulkan "vulkan/vulkan.h" "defined(VK_VERSION_1_4) || (defined(VK_VERSION_1_3) && VK_HEADER_VERSION >= 255)" ||
> +        warn "Disabling vulkan" && disable vulkan
>   fi

Doesn't that just always disable vulkan if any of the previous checks 
succeed? The logic looks weird to me.

Also, shouldn't all of those check_* calls disable vulkan already? 
That's what the first argument is for.

On Fri, Sep 29, 2023 at 1:37 PM Timo Rothenpieler <timo@rothenpieler.org> wrote:
>
> On 29.09.2023 15:52, Tristan Matthews wrote:
> > Fixes: https://trac.ffmpeg.org/ticket/10596
> > ---
> >   configure | 3 ++-
> >   1 file changed, 2 insertions(+), 1 deletion(-)
> >
> > diff --git a/configure b/configure
> > index 20db1801ed..50ba6f772f 100755
> > --- a/configure
> > +++ b/configure
> > @@ -7154,7 +7154,8 @@ enabled crystalhd && check_lib crystalhd "stdint.h libcrystalhd/libcrystalhd_if.
> >
> >   if enabled vulkan; then
> >       check_pkg_config_header_only vulkan "vulkan >= 1.3.255" "vulkan/vulkan.h" "defined VK_VERSION_1_3" ||
> > -        check_cpp_condition vulkan "vulkan/vulkan.h" "defined(VK_VERSION_1_4) || (defined(VK_VERSION_1_3) && VK_HEADER_VERSION >= 255)"
> > +        check_cpp_condition vulkan "vulkan/vulkan.h" "defined(VK_VERSION_1_4) || (defined(VK_VERSION_1_3) && VK_HEADER_VERSION >= 255)" ||
> > +        warn "Disabling vulkan" && disable vulkan
> >   fi
>
> Doesn't that just always disable vulkan if any of the previous checks
> succeed? The logic looks weird to me.

No, it will only disable vulkan if all the previous checks fail as
this conditional shortcircuits (so to get to here it would have to be
false || false || false || false || false || warn "Disable vulkan" &&
disable vulkan). If I hack the version number to match mine (changing
it to 1.3.239), vulkan won't be disabled (that's why I added the
warning message) and you'll hit the build breakage I and others are
hitting currently.

>
> Also, shouldn't all of those check_* calls disable vulkan already?
> That's what the first argument is for.

They don't, so my first thought was that since this is the only place
where check_pkg_config_header_only is called, it seemed likely that
there was a bug in that function. However, I couldn't spot anything
obviously broken in it and it is returning the correct value (hence
this patch behaving as expected).

Best,
Tristan

> _______________________________________________
> ffmpeg-devel mailing list
> ffmpeg-devel@ffmpeg.org
> https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
>
> To unsubscribe, visit link above, or email
> ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".

On Fri, Sep 29, 2023 at 3:55 PM Tristan Matthews <tmatth@videolan.org> wrote:
>
> Fixes: https://trac.ffmpeg.org/ticket/10596
> ---
>  configure | 3 ++-
>  1 file changed, 2 insertions(+), 1 deletion(-)
>
> diff --git a/configure b/configure
> index 20db1801ed..50ba6f772f 100755
> --- a/configure
> +++ b/configure
> @@ -7154,7 +7154,8 @@ enabled crystalhd && check_lib crystalhd "stdint.h libcrystalhd/libcrystalhd_if.
>
>  if enabled vulkan; then
>      check_pkg_config_header_only vulkan "vulkan >= 1.3.255" "vulkan/vulkan.h" "defined VK_VERSION_1_3" ||
> -        check_cpp_condition vulkan "vulkan/vulkan.h" "defined(VK_VERSION_1_4) || (defined(VK_VERSION_1_3) && VK_HEADER_VERSION >= 255)"
> +        check_cpp_condition vulkan "vulkan/vulkan.h" "defined(VK_VERSION_1_4) || (defined(VK_VERSION_1_3) && VK_HEADER_VERSION >= 255)" ||
> +        warn "Disabling vulkan" && disable vulkan
>  fi
>

This change doesn't seem right. If a feature is explicitly requested,
we generally fail the build and don't just disable the feature
(afterall the user wanted it to be on). If the feature is not
explicitly requested, then it should not print a message.

- Hendrik

On Fri, Sep 29, 2023 at 2:32 PM Hendrik Leppkes <h.leppkes@gmail.com> wrote:
>
> On Fri, Sep 29, 2023 at 3:55 PM Tristan Matthews <tmatth@videolan.org> wrote:
> >
> > Fixes: https://trac.ffmpeg.org/ticket/10596
> > ---
> >  configure | 3 ++-
> >  1 file changed, 2 insertions(+), 1 deletion(-)
> >
> > diff --git a/configure b/configure
> > index 20db1801ed..50ba6f772f 100755
> > --- a/configure
> > +++ b/configure
> > @@ -7154,7 +7154,8 @@ enabled crystalhd && check_lib crystalhd "stdint.h libcrystalhd/libcrystalhd_if.
> >
> >  if enabled vulkan; then
> >      check_pkg_config_header_only vulkan "vulkan >= 1.3.255" "vulkan/vulkan.h" "defined VK_VERSION_1_3" ||
> > -        check_cpp_condition vulkan "vulkan/vulkan.h" "defined(VK_VERSION_1_4) || (defined(VK_VERSION_1_3) && VK_HEADER_VERSION >= 255)"
> > +        check_cpp_condition vulkan "vulkan/vulkan.h" "defined(VK_VERSION_1_4) || (defined(VK_VERSION_1_3) && VK_HEADER_VERSION >= 255)" ||
> > +        warn "Disabling vulkan" && disable vulkan
> >  fi
> >
>
> This change doesn't seem right. If a feature is explicitly requested,
> we generally fail the build and don't just disable the feature
> (afterall the user wanted it to be on).

That is the case here, with or without this patch, on my system
--enable-vulkan will fail as expected on:
> ERROR: vulkan requested but not found

The bug I'm trying to address is the autodetect case. I'm happy to
drop the warning, I just wanted it to be obvious what was happening
(but one could infer it from the list of modules that will be built).

>  If the feature is not
> explicitly requested, then it should not print a message.
>

On 29.09.2023 20:10, Tristan Matthews wrote:
> On Fri, Sep 29, 2023 at 1:37 PM Timo Rothenpieler <timo@rothenpieler.org> wrote:
>>
>> On 29.09.2023 15:52, Tristan Matthews wrote:
>>> Fixes: https://trac.ffmpeg.org/ticket/10596
>>> ---
>>>    configure | 3 ++-
>>>    1 file changed, 2 insertions(+), 1 deletion(-)
>>>
>>> diff --git a/configure b/configure
>>> index 20db1801ed..50ba6f772f 100755
>>> --- a/configure
>>> +++ b/configure
>>> @@ -7154,7 +7154,8 @@ enabled crystalhd && check_lib crystalhd "stdint.h libcrystalhd/libcrystalhd_if.
>>>
>>>    if enabled vulkan; then
>>>        check_pkg_config_header_only vulkan "vulkan >= 1.3.255" "vulkan/vulkan.h" "defined VK_VERSION_1_3" ||
>>> -        check_cpp_condition vulkan "vulkan/vulkan.h" "defined(VK_VERSION_1_4) || (defined(VK_VERSION_1_3) && VK_HEADER_VERSION >= 255)"
>>> +        check_cpp_condition vulkan "vulkan/vulkan.h" "defined(VK_VERSION_1_4) || (defined(VK_VERSION_1_3) && VK_HEADER_VERSION >= 255)" ||
>>> +        warn "Disabling vulkan" && disable vulkan
>>>    fi
>>
>> Doesn't that just always disable vulkan if any of the previous checks
>> succeed? The logic looks weird to me.
> 
> No, it will only disable vulkan if all the previous checks fail as
> this conditional shortcircuits (so to get to here it would have to be
> false || false || false || false || false || warn "Disable vulkan" &&
> disable vulkan). If I hack the version number to match mine (changing
> it to 1.3.239), vulkan won't be disabled (that's why I added the
> warning message) and you'll hit the build breakage I and others are
> hitting currently.

My shell disagrees:

$ function warn() { echo $1; }
$ true || false || warn "Disabling vulkan" && echo DISABLED
DISABLED
$ false || true || warn "Disabling vulkan" && echo DISABLED
DISABLED
$ false || false || warn "Disabling vulkan" && echo DISABLED
Disabling vulkan
DISABLED
$ true || true || warn "Disabling vulkan" && echo DISABLED
DISABLED

No matter the results of the checks, this will always disable Vulkan, 
since the chain before it will always have a positive outcome, so && 
continues.

On Fri, Sep 29, 2023 at 3:26 PM Timo Rothenpieler <timo@rothenpieler.org> wrote:
>
> On 29.09.2023 20:10, Tristan Matthews wrote:
> > On Fri, Sep 29, 2023 at 1:37 PM Timo Rothenpieler <timo@rothenpieler.org> wrote:
> >>
> >> On 29.09.2023 15:52, Tristan Matthews wrote:
> >>> Fixes: https://trac.ffmpeg.org/ticket/10596
> >>> ---
> >>>    configure | 3 ++-
> >>>    1 file changed, 2 insertions(+), 1 deletion(-)
> >>>
> >>> diff --git a/configure b/configure
> >>> index 20db1801ed..50ba6f772f 100755
> >>> --- a/configure
> >>> +++ b/configure
> >>> @@ -7154,7 +7154,8 @@ enabled crystalhd && check_lib crystalhd "stdint.h libcrystalhd/libcrystalhd_if.
> >>>
> >>>    if enabled vulkan; then
> >>>        check_pkg_config_header_only vulkan "vulkan >= 1.3.255" "vulkan/vulkan.h" "defined VK_VERSION_1_3" ||
> >>> -        check_cpp_condition vulkan "vulkan/vulkan.h" "defined(VK_VERSION_1_4) || (defined(VK_VERSION_1_3) && VK_HEADER_VERSION >= 255)"
> >>> +        check_cpp_condition vulkan "vulkan/vulkan.h" "defined(VK_VERSION_1_4) || (defined(VK_VERSION_1_3) && VK_HEADER_VERSION >= 255)" ||
> >>> +        warn "Disabling vulkan" && disable vulkan
> >>>    fi
> >>
> >> Doesn't that just always disable vulkan if any of the previous checks
> >> succeed? The logic looks weird to me.
> >
> > No, it will only disable vulkan if all the previous checks fail as
> > this conditional shortcircuits (so to get to here it would have to be
> > false || false || false || false || false || warn "Disable vulkan" &&
> > disable vulkan). If I hack the version number to match mine (changing
> > it to 1.3.239), vulkan won't be disabled (that's why I added the
> > warning message) and you'll hit the build breakage I and others are
> > hitting currently.
>
> My shell disagrees:
>
> $ function warn() { echo $1; }
> $ true || false || warn "Disabling vulkan" && echo DISABLED
> DISABLED
> $ false || true || warn "Disabling vulkan" && echo DISABLED
> DISABLED
> $ false || false || warn "Disabling vulkan" && echo DISABLED
> Disabling vulkan
> DISABLED
> $ true || true || warn "Disabling vulkan" && echo DISABLED
> DISABLED
>
> No matter the results of the checks, this will always disable Vulkan,
> since the chain before it will always have a positive outcome, so &&
> continues.

Oh I see, I should have parens around the last bit, e.g.:
false || true || (warn "Disabling vulkan" && echo DISABLED)

but as Hendrik mentioned, the warning can be dropped in which case
this would become:
check_pkg_config_header_only vulkan ... || check_cpp_condition || disable vulkan

As mentioned earlier in the thread if someone has a better idea of how
to avoid explicitly calling disable vulkan like this (i.e. why isn't
this already working) that would of course be preferable.

Best,
-t