Re: [Mlt-devel] Image sequence question

[Dan D. <da...@de...>]

On Tue, Feb 7, 2012 at 2:06 PM, j-b-m <j-...@us...> wrote:
> On Tuesday 07 February 2012 10:32:24 Dan Dennedy wrote:
>> On Tue, Feb 7, 2012 at 10:08 AM, j-b-m <j-...@us...> wrote:
>> > Hello.
>> >
>> > Trying to improve image sequence support in Kdenlive, I discovered that
>> > MLT
>> > cannot currently load a sequence with files like this:
>
> (...)
>
>> We could simply increase the hard-coded 100 to something much higher
>> like INT_MAX. If you can do that and see how long it takes to fail on
>> an invalid input, that would be great. Of course, it will actually
>> vary depending on the machine, but just to get an idea...
>
> I tried increasing the max limit to INT_MAX but it then takes several minutes
> on a modern laptop (core I7).

yeah, not good :)

> Setting a max limit of 1 million, it takes between 1 and 8 seconds to process
> all possible file names, depending on the computer load.

Was that for a single producer init?

> However I was surprised to discover that the image producers are called twice.
>
> For example if I try:
>
> melt image%d.jpg
>
> the pixbuf producer is initialized, then closed and initialized again, which
> means that the filename processing routine is called twice.

Not surprising to me at all. Melt is trying to do a safer approach to
auto-profile by re-opening the composition instead of just asking all
producers in the composition for the first frame again. Maybe it is
not really necessary. Do you think I should change melt now to not do
that? Maybe it will have a bad effect on kdenlive rendering. Or, do
you always specify a -profile to melt, which bypasses the auto-profile
step?

> Maybe that is necessary for MLT. But worse, when there is no matching
> filename, it does the following:
>
> Create pixbuf producer
> Create qimage producer
> Create pixbuf producer
> Create qimage producer
>
> And then it aborts... Seems a bit too much. When a producer returns NULL, I
> don't see a reason to try it again.
>
> regards
> jb