Wrong 'Include region' mask appliance to equirectangular/cylindrical image

[Rumpel]

I noticed this bug when tried to fix nadir/zenith errors by blending already stitched 360x180 panorama and retouched patch. Instead of applying 'Exclude region' mask to the pano I applied 'Include region' to the patch (which is simpler due to equirectangular distortion of poles) and get completely wrong exclusion on other image (look "0 - initial occurrence.pto").
First I thought it is related with poles on output projection, but after playing with shape of the mask, FoV of images and projections I found that it is not entirely the case. And occurs only on equirectangular/cylindrical projection of input files - other projections are ok.

I think it maybe somehow connected with transition through right/left borders of equirectangular/cylindrical input image, maybe poles somehow involved too, but I'm not sure. After playing with different parameters I failed to define clear reason why it happens. Moreover I found a case when images intersect but mask in one image do not cover another at all but causes error ("3 - weird case.pto"), this I cannot explain at all.

I have attached ZIP archive with pto files, images and screenshots to show what I'm talking about.

--------

Operating System: Windows NT 6.1 (build 7600)
Architecture: 64 bit
Free memory: 7279636 kiB
Active Codepage: 1251 (Cyrillic Windows)

Hugin
Version: 2011.4.0.cf9be9344356 built by Matthew Petroff
Path to resources: C:\Programs\Hugin/share/hugin/xrc/
Path to data: C:\Programs\Hugin/share/hugin/data/

Libraries
wxWidgets: 2.8.11.0
libpano13: 2.9.18
Boost: 1.44.0
Exiv2: 0.20.0

--------

Operating System: Windows XP (build 2600, Service Pack 3)
Architecture: 32 bit
Free memory: 2097151 kiB
Active Codepage: 1251 (Cyrillic Windows)

Hugin
Version: 2011.4.0.cf9be9344356 built by Matthew Petroff
Path to resources: C:\Programs\Hugin/share/hugin/xrc/
Path to data: C:\Programs\Hugin/share/hugin/data/

Libraries
wxWidgets: 2.9.2.0
libpano13: 2.9.18
Boost: 1.46.1
Exiv2: 0.21.1

[tmodes (tmodes)]

Confirmed.
The issue is that the mask is handled as a list of points. When the mask contains the nadir or zenith and the mask expands over the +-180 border the mask is incorrectly transformed.
I tried some approaches to fix it. But until now I have no solution which works for all use cases.