Just as I was laying out the current status with regards to binary distributions, Allard Katan released a Windows binary installer of 2009.2.0 and uploaded it to the project’s Sourceforge page. I quickly fired up my ailing Windows XP partition to test it.
It is still based on the old 0.7.0 installer that I wrote more than two years ago. My verdict upfront: not official release quality if you ask me. The devil lies in the details.
The most serious issue is Enblend/Enfuse. Because Windows does not have a package manager, the Hugin installer must ship runtime dependencies with the actual Hugin binary. Enblend/Enfuse is a mandated runtime dependency, and a moving target. Compare the images on the right.
The first one are six layers, simulating six images to be blended into a 360°x180° full spherical panorama.
The second one is the blend by Enblend-3.2 as shipped with this installer. The third one is the blend by Enblend’s pre-release 4.0. All using default settings. This installer ships with an outdated version of Enblend.
Unlike Enblend/Enfuse which is a mandated dependency, a control point generator is not mandatory.
Ippei Ukai has introduced an elegant modular plug-in solution for control point detectors for 0.8.0 on OSX, and the Windows version should be updated to a similar mechanism. It’s only a few lines in the installer. But it requires understanding for the process.
There are good reasons not to include control point detectors with Hugin binary distributions: Patents. U.S.-based Linux distributions such as Fedora/Red Hat and Debian steer away from distributing Autopano-SIFT-C because of the patents. And the SURF algorithm, used by Panomatic is restricted by a European patent. I personally would have not uploaded a binary installer with control point detectors to Sourceforge.
Last but not least: the release notes are unclean. They are simply copy & pasted from the source code release notes. To me this is a warning about the quality of the installer. This shortcut may not be critical, but how many other corners have been cut? In my opinion this installer’s quality is closer to an unofficial snapshot than to an official, long term supported release. There are plenty of such snapshots around the net, some of them linked from the download page of this blog. It’s the beauty of the GPL: the code is set free and others can do almost all they want with it.
I do not endorse this installer. But if it makes other users happy, let it be. When I wrote the community charter, I meant it. My opinion on the Windows installer is unimportant. I am anyway not using Hugin on Windows. Even if I disagree with Allard, I gave him full access rights so that he can go ahead and do what he thinks is right with the Windows installer. And I don’t regret it. I wrote the installer when I had an interest to understand the process of building an installer for Windows, and even if I am the original author of the installer, it does not belong to me. I’m happy if you like it as-is. Just don’t complain that it does not match up to the binaries on OSX or on Linux, it is a known bug and limitation.
Filed under: hugin, windows | Tagged: Allard Katan, Ippei Ukai |
This week in panospace 
Hi Yuv,
I feel compelled to reply. I admire your effort to bring the development of Hugin to a higher level, so your opinion matters to me. Though I think we ‘agree to disagree’ I just want to make sure both you and your audience see my point of view too.
1 Yes, this installer is far from perfect. A better installer could have been produced by someone willing to invest the time. I hoped someone would pick it up but nobody did, so I felt obliged to at least do something with the limited time I had to spend on this.
2 I think the enblend version is a bit of a lame point as a) the 4.0 isn’t even an official enblend release so even enblend ships with an outdated version of enblend in your reasoning. b) As soon as the new enblend version is released anyone can replace the .exe in their hugin/bin directory and I can (and will) replace the installer easily. If you have a Windows binary of enblend 4 that has been sufficiently tested for a release just email it to me and I will upload a 2009.2.1 installer within a week.
3 All the points you make -except maybe release notes- are also valid for the previous default installer (0.7). If you had the choice between the two, would you prefer the 0.7 installer? Because that was basically my choice, to go with what I uploaded now or leave things as they are. I believe not uploading the current file would have shortchanged both new and existing users even more. To me, the lag between the binary release and the current status of the program was just far beyond what was acceptable.
Anyway, I really hope somebody else will stand up to take care of the windows binaries in a more systematic way because I simply don’t want to spend the time it takes me (because of my limited experience) to do it as it really should be done.
Allard
Dear Allard,
1. I was hoping too, for more than two years. After I published my unofficial installer prior to 0.7.0 Guido produced the 0.7.0 that is on SourceForge. Then there was nothing for a long while, until you and Ad contributed to improve the installer again. In between, there were many Windows users building Hugin for themselves (traces of their questions are on the mailing list), but for whatever reason none of them felt compelled to distribute. Unlike you, I don’t feel obliged to release something I believe is “far from perfect”; nor do I feel obliged to do something to please a group of users who can’t help themselves. I allocate my limited time differently: if I was to work on the Window version of Hugin, and I had limited time, I’d work on making the CP detector installer modular and committing it to SVN.
2. You’re using an unofficial version of Libpano from SVN, then why must Enblend be an official version? When I released my 0.7.0 series of installers, I used an unofficial version of Enblend. It turns out to be better than the official Enblend 3.2 (memory leaks!). Official version make sense for mature and slower moving code. For the rest, it takes time to identify the best combination, which happens every now and then, like eclipses.There were recently binaries of Enblend pre-release 4.0 posted on the mailing list IIRC. I would have put those in the installer. Now don’t call your next version of the installer 2009.2.1: the third version number is reserved for patches in the code. You can call it Hugin-2009.2.0a or Hugin-2009.2.0_win_2 (like there is Hugin-0.8.0_mac2.dmg). We have total chaos with the names of the files on SF, but one thing should be constant: the first part should always be “Hugin-VM.Vm.Vp” from the tarball that was used to build it (VM=major version, Vm = minor version Vp=patch version). I think you should slightly rename the current files to align them with this important convention.
3. I was not involved in the production of the 0.7 binary installer other than teaching Guido how to use the ISS files. What is on Sourceforge is not my work. I would have never put an installer with any of the autopano code on Sourceforge. The Sourceforge servers are in the US and this amounts IMO to importing patent-protected software to the US. If others want to expose themselves to the risk, it’s their business. A modular installer enables US citizens to abide by their patent laws, while leaving to others the option to fetch and install the CP generator from another location.
Last but not least, the lag between the current status of the program and the binary release is like a spring. Press it hard enough and it will spring back. By publishing your installer you have effectively reduced some of the pressure that may have motivated others with more time on their hand to go through the experience that it takes to become a good builder/distributor.