On Sunday 02 November 2003 00:08, Philipp Thomas wrote:
Even if they do, there are a lot of cases where you'd like to have more. Take for instance development kernels (i.e. those with uneven minor version) where lots of things can change which can lead to the nvidia driver not working anymore because symbols (i.e. functions or variables) they need don't exist anymore in the kernel. In that case it's normally rather easy to fix that yourself, given that you have the source code.
And before you mention it, I do know that the glue code is open source. But some experiences tell me that the code in the binary only module isn't encapsulated as good as it ought to be.
So whichever way you turn it, I uphold my POV that binary only modules are a bad thing that should be avoided as much as possible and so also hardware that needs them to fully use it.
Of course, I agree with the last statement that they should avoided as much as possible. But we must also understand the position of companies like Nvidia and try meet them somewhere in the middle. In my personal point of view, the problems with modules in the Linux kernel is not just a question that concerns graphics drivers. I think the modularity of the kernel could be improved, as these symbols are far too easily broken. And when its too easily broken, creating drivers for commercial systems becomes a big issue, as a company that is selling state of the art hardware is going to wait with releasing these for us, because maintaining these drivers for every version is too expensive, and we just have to understand their point of view, not to release specs to us until "later" in this hardwares life. Which is just hard luck for us ... so, to me, the question here would be a challenge to improve the modular aspects of the kernel, at al. Just my 2¢, Örn