complex masks slow down nona considerably

[KFJ]

When using masks with numbers of points ranging in the hundreds on images rendered by nona, rendering times increase to several times the time taken for unmasked images, or images with masks with only a few tens of points. I posted my observation on hugin-ptx and Bruno said it should be a bug report. Here's a copy of my initial post:

I'm currently working on a script that relies heavily on masking. The script generates the masks to arbitrary precision. Most of the time I was using spacings of about 100 pixels for the masks, and all was well. Today, while testing, I generated much more precise masks. I set the spacing to one pixel, and this threw a spanner in the works when I tried to process the data with nona. I didn't wait for the process to terminate and tried with 10 pixel stride. This increased processing time by roughly 200% compared to 100 pixel stride.

So I made the test and made a single-image pto and put a 125-point mask on the image manually, just to be sure it wasn't some strange effect of my script. It took nona over 11 seconds to render the image. when I removed the mask, it only took 4.4 seconds. This was with a 12MP 16bit TIFF in stereographic projection, on Pre-Release 2010.5.0.a11acbf8be1d. I wonder if anyone else has noticed this? Has it been like this all along? Maybe I'm the only one using masks with more than a couple of dozen points?

I've attached the .msk file which I used to slow down nona.

Kay

[Yuv (yuv)]

ok, so we have a confirmed scalability bottleneck. how many real-life user-entered masks are going to have 100+ points?