We use 'versioned' user directories on OSX and Windows: By default, the major version number is used as part of the directory name. This leads, on Windows, to hassles for users: When you upgrade from 2.2 to 2.3, you have to manually copy the old user directory to the new location if you want to preserve your preferences, etc.
I've more or less got a patch for configure.py that mimics what already happens on OSX on Windows, but now I've started to wonder about it. The logic behind the patch is this. (1) When we configure, we first make sure we are in the 'default' user directory, by checking the pathname. If not, we do nothing special. (2) Next, we look for lyxrc.defaults. If it's there, we rightly figure it's not a new directory that we're creating but we're just reconfiguring, and so we do nothing special. (3) Next, we look for previous user directories LyX2.2, LyX2.1, etc. If one is found, we copy its contents into the current directory. But what if the user has purposely deleted their user directory so as to get a fresh one? Then, when they restart, if we find an old one, we copy its contents again. But what if there was some problem with the old preferences? Something seems wrong about this to me. I wonder if, all things considered, a better option would be to stop using versioned user directories on Windows, at least. Then no special copying is needed, etc. If we did this for 2.4, then we could tell people *as usual* that they needed to copy over their old preferences, but that would be the last time. I take it that the reason to have versioned user directories is so people can have parallel installations of 2.2 and 2.3, say. But you can still do that, of course, by manually setting the user directory. How many people do that anyway on Windows? Riki
