PixInsight 1.5.1

[Juan Conejero]

Hi all,

PixInsight 1.5.1 is a quick bug-fix release. It should be released during the next week. These are the main bug fixes:

- ImageIntegration process. Pixels that are completely clipped (which may happen with small sets of an even number of images) are now being set to zero in the integrated image, which is clearly incorrect. The correct behavior is to assign the median value of the corresponding pixel stack, so "black holes" can never occur (there are other criteria, but using the median seems the best option, IMO). This is a really stupid mistake on my part. Until version 1.5.1 is released, if you want to integrate a small even set of images, say 4 or 6, the workaround is simple: include one of them twice. You can also use the Statistics tool to ensure that no black pixel is present in the integrated image.

- A rare problem with the averaged sigma clipping algorithm (ImageIntegration), exclusively on Windows, leading to a crash sometimes. This seems a bug in the Visual C++ compiler (invalid compilation). I'm still trying to find a workaround.

- Instability in morphological transformations. Multithreaded morphological transformations can lead to crashes due to occasional race conditions (a small error in PCL code). This affects several processes, especially MorphologicalTransformation and StarMask, and I've been able to reproduce it on Linux only (the race condition probability seems to be extremely small on Windows and Mac OS X). This has been fixed.

- A strange "memory exhausting" bug in ProcessContainer (see this thread on the Bug Reports board). This also seems a Windows-exclusive issue. I've been unable to reproduce it on Linux, Mac OS X and Vista x64. It seems to happen only on 32-bit Windows XP.

- Several interface drawing issues.

Along with these, there are other reported bugs that I still haven't investigated.

[Nocturnal]

Thanks for being on top of things Juan!

[Juan Conejero]

Next version report:

* All open bugs have been fixed now, except:

- The "memory exhausting" ImageContainer error. I seem unable to reproduce this bug. Tomorrow I'll make a serious attempt to catch it on a box running native Windows XP 32-bit, since it doesn't happen on the other platforms I have for testing: several Linux distributions (Fedora 8, 9, 10, Ubuntu 8.10, 9.04), Mac OS X 10.5.7, Windows Vista x64 and Vista x86.

- The strange crash with ImageIntegration's averaged sigma clipping algorithm, due to erroneous compilation with MS Visual C++ 2008.

- 32-bit TIFF files generated by DeepSkyStacker are always loaded as pure black, also due to erroneous compilation with MS Visual C++ 2008.

Fixing the last two bugs involves rebuilding the corresponding modules with a different compiler under Windows. I've been waiting for MinGW to release a new version with the GCC 4.3 compiler - unfortunately it is still in the alpha phase... and it seems there is no 64-bit version. The only practical solution seems buying the Intel compiler for Windows, which certainly doesn't make me too happy 

* Integration of the Mozilla JavaScript engine (SpiderMonkey) version 1.8.0 RC1. This is the same engine that comes with Firefox 3. Despite the fact it's a release candidate, this version is extremely stable (of course, I have tested it thoroughly) and significantly faster than previous version 1.7 (Example: 3dplot script running on a 600x500 image, 1.7=18.1 seconds, 1.8=14.5 seconds).

* Many runtime JavaScript methods use now the JSFastNative calling convention, including all methods of the Math object.

* New properties of the JavaScript ImageWindow object to retrieve additional metadata: exposure time, ISO speed, aperture, CFA type, etc.

* I'm thinking on a way to improve the DBE interface so that it allows application of successive background corrections without the current undo/redo problems due to the active dynamic session. See this thread for more information. This is complex and probably requires a redesign of the DBE tool.

[Nocturnal]

Great progress Juan! I think the DBE thing can wait another release. The way I describe it in my article works quite well I think. It's a bit un-intuitive to new users so an eventual solution would be a good thing.

[Niall Saunders]

Juan,

How much is the Intel compiler that you are looking for?

Perhaps your loyal user base might be able to send you enough by way of donation to help you move forward?

I certainly can't afford to volunteer the whole donation myself, but I would be happy to at least try and reduce your costs.

Cheers,

[Juan Conejero]

Hi Niall,

Thank you. That's very nice of you, and I appreciate it a lot 

Each commercial user has paid his/her license, and that's all I ask. No need for donations, and if we arrive at that situation in the future --which I hope won't happen--, I'll re-think the entire project to find a good solution, but I'll never ask for donations. This is a big bet and I am prepared to lose since the beginning (I always prefer to win, but fortunately I'm accustomed to lose, so it would be no surprise ).

The Intel compiler isn't cheap:

http://software.intel.com/en-us/articles/buy-or-renew/

but the actual problem isn't the money (it would be 1300 Euros for the three platforms, at today's exchange rates). The problem is that I find illogical to use the Intel compiler, since I am extremely happy with the GCC compilers on Linux and Mac OS X (wait to see the performance improvements when I recompile the whole PI with GCC 4.4 and automatic vectorization enabled). I'll try to find a way to use GCC on Windows. Ultimately, I'll consider buying the Intel compiler if I can't find a better solution.

[David Serrano]

Isn't cygwin an option?

[Nocturnal]

Juan,

when I was messing with Fedora 10 and installing gcc it also offered a MingW windows cross compiler in the package manager. Is this the package you are talking about?

[georg.viehoever]

Juan,

I am working with the Intel compiler in my professional life. It has its own strengths and weaknesses. Getting something running that works on gcc/VC can be anything between "just recompile" and "what is this thing doing with my code?" . My recommendation: Stick to Visual C and find a workaround for its quirks.

Georg