On 26/02/17 11:14, Nicolas George wrote:
> L'octidi 8 ventôse, an CCXXV, Mark Thompson a écrit :
>> And what would the user conclude from that message?  Have they done
>> something incorrectly?  If so, what?  Or is it a bug in the filter?
>> All of these messages pretty much require reading the source code and
>> looking at the documentation for the relevant libmfx function to
>> determine what they mean.  (Well, maybe not "out of memory".)
> 
>> I disagree.  For normal output, yes, this sort of thing is unhelpful.
>> For fatal error cases, precise information such as error codes should
>> be provided so that someone reading the source code can determine
>> exactly what the error was and where it happened.
> 
> Anything above the verbose level is meant for users, not just
> developers. If reading the source code is necessary to understand it,
> then you are doing something seriously wrong.

I suspect that the thing I am doing seriously wrong is attempting to use an 
opaque proprietary library which has more complex semantics than we can 
sensibly express.

> If the error results from invalid parameters set by the user, then the
> error message should say so, possibly stating what parameter and its
> acceptable values.

For this sort of thing the set of acceptable inputs is not well-defined, 
unfortunately.  (My original problem - seemingly some apparently-equivalent 
dxva2/qsv surfaces are more equal than others, in a way I am as-yet-unable to 
determine.)

> If the error results from external (hardware or system) problems, then
> the error message should say so, pointing to the cause of the problem.
> 
> Otherwise, the error should clearly state that it is a BUG, and suggest
> the user to report it, along with diagnostics relevant for the
> developers obtained at loglevel debug.

Would you prefer something like "external error: MFXVideoVPP_RunFrameVPPAsync 
returned error code %d", which would emphasise the origin of the problem?
_______________________________________________
libav-devel mailing list
[email protected]
https://lists.libav.org/mailman/listinfo/libav-devel

Reply via email to