Michel Juillard michel.juillard@mjui.fr writes:
Maybe it would be better to split the installer in 2 so as to limit inconvenience when a slow link is lost in the middle of a 200MB download.
Not sure if this is possible with NSIS installer, will check. But in my opinion it defeats the purpose since it makes the installation more complicated (you cannot just click "run" on the download as a normal installer).
We could also offer other download options than HTTP as we currently do (and indeed HTTP is not very good at resuming a download). We could also offer FTP and Bittorrent, which are much better at resuming downloads.
Concerning the concern of Stéphane about storage space, I forgot to mention that we can also host our packages on Sedna: I already do that for my Debian for Economists (which is 2.5Gb times 4 configurations!).
Also, I am open to the possibility of having 2 installers, with and without Octave, if you think this is necessary. This way, everybody will be happy since the current installer would still be available.
On 03/07/2012 11:12 AM, Sébastien Villemot wrote:
Houtan Bastani <houtan.bastani@ens.fr> writes: I guess, if you want to offer a 200mb package, you should offer a 20mb package as well. 200 mb is a lot to download if you are on a slow internet connection, which maybe in advanced countries may not be a problem but in other countries it can be. If you are in a developing country with a slow connection and not much money, you will likely not have MATLAB. So you will have to download Octave. The total download size is therefore the same whether we ship Octave or not in our installer. On 7 Mar 2012, at 11:01, Sébastien Villemot wrote: Stéphane Adjemian <stepan@dynare.org> writes: I strongly support solution 2. I am sure that even a windows user is able to unzip an archive. So except if there is another issue with this version of octave, I don't see why we should distribute octave (and pay space on kirikou for that). We may alternatively distribute a python script that download and unpack octave for the user... But I don't think this is necessary. I don't think it is a matter of monetary cost: currently we are far from using all our allocated space on kirikou (the hard disk is almost full but we could expand it by 20Go for free). Also note that it is a little more complicated than just unpacking a zip. You have to do manual renaming of the libraries if you want an optimized BLAS, you have to manually activate the Octave toolboxes, and manually create the desktop shortcut. See http://www.octave.org/wiki/index.php?title=Octave_for_Windows Of course this is not impossible for a user, but in my opinion it is sufficiently difficult to limit adoption of Dynare/Octave by our users. For me the real tradeoff is between download size and ease of installation. Since 200Mb is still a reasonably small download time, I would favor ease of installation. On 07/03/2012 09:47, Sébastien Villemot wrote: Hi, I intend to release Dynare 4.2.5 by the end of the week. This release will in particular incorporate bugfixes to sensitivity and identification, so that Marco can use it in its upcoming workshop. The 4.2.5 release will probably be the last in the 4.2 branch, since we aim at releasing 4.3.0 in June. Please backport to the 4.2 branch any bugfix that you made in master and that is relevant for 4.2. Your window of opportunity is only of a few days :) Also, I would like to get your feedback concerning the Windows package of Dynare, in relation to Octave. As you may know, Octave 3.6.1 has been released recently, and two Windows packages have been created for it (see [1]): - one compiled with Microsoft Visual C++ (MSVC), which comes with a neat installer and an experimental preview of the Octave GUI. The problem is that it is impossible to fully cross-compile Dynare for Octave/MSVC from GNU/Linux: I seem to be possible to do it for MEX files, but it is clearly impossible for Oct files (which are "ordschur" and "qzcomplex"). - one compiled with MinGW: no problem for cross-compiling Dynare for Octave/MinGW from GNU/Linux. But at this time there is no automated installer of Octave/MinGW: it comes only in the form of ZIP files that one has to uncompress manually. We have to decide which Windows version of Octave we support (supporting both seems an unnecessary burden). I see the following possible solutions: 1) Support the MinGW version. Cross-compile everything from GNU/Linux. Bundle Octave/MinGW in the Dynare installer (leaving the choice to our user of installing it or not). The big advantage is that the Dynare installer will be totally self-contained. The drawback is that it makes the installer bigger: it would go from 20Mb to about 200Mb. 2) Support the MinGW version. Cross-compile everything from GNU/Linux. Ask our users to manually download and unzip Octave/MinGW. 3) Support the MSVC version. Cross-compile the MEX files from GNU/Linux (seems possible). For the Oct files, compile them manually from time to time in a Windows Virtual Machine, and smuggle them in the nightly cross-compiled snapshot. Ask our users to use the Octave/MSVC installer (also note that before installing Octave/MSVC, they have to download and install the MSVC runtime from Microsoft website). My ranking is 1 > 2 > 3. If you have any opinion on this, please raise your voice. Otherwhise I’ll implement solution 1 for the 4.2.5 package. [1] http://www.octave.org/wiki/index.php?title=Octave_for_Windows _______________________________________________ Dev mailing list Dev@dynare.org https://www.dynare.org/cgi-bin/mailman/listinfo/dev -- Sébastien Villemot Researcher in Economics & Debian Maintainer http://www.dynare.org/sebastien Phone: +33-1-40-77-84-04 - GPG Key: 4096R/381A7594 _______________________________________________ Dev mailing list Dev@dynare.org https://www.dynare.org/cgi-bin/mailman/listinfo/dev _______________________________________________ Dev mailing list Dev@dynare.org https://www.dynare.org/cgi-bin/mailman/listinfo/dev _______________________________________________ Dev mailing list Dev@dynare.org https://www.dynare.org/cgi-bin/mailman/listinfo/dev
Dev mailing list Dev@dynare.org https://www.dynare.org/cgi-bin/mailman/listinfo/dev