On Wednesday 15 February 2006 18:23, NextGen$ wrote: > * freenetwork at web.de <freenetwork at web.de> [2006-02-15 17:08:05]: > > > i don't think it is wise if all the binary libraries have the same name! > > > > cpu identification has not to be only done by the installer to fetch the > > correct lib, but also when running the node. > > people (me) tend to copy data and even java programs from windows to linux > > and vice versa, or having multiple operating systems on the same computer > > and run java programs from one single location. > > now the best that can happen if by windows box runs a debian binary (or > > even extremely more worse: a sse3 lib from work on my athlon thunderbird at > > home - that's sure to blow) is that it refuses to work, the words that it > > crashes. to prevent this a cpu detection has to be > > done to keep the node from executing wrong library machine code, to the > > detection has to be in the node. and calling the correct library by its > > filename is FAR easier than to guess if the available lib is the corerect > > one for the platform. > > i think it is also very unintuitive to have different libs share the same > > name as you can never know if the current library is the correct for the > > used machine enviroment, in fact there is IMHO no reason to actually LET > > them have the same name! > > At least an obvious one : Simplicity : getting rid of CPUID code in the node. > > > also, for multi-os machines it would make sense to have the, e.g. windows > > and linux, library both available and in the same folder without having to > > swap them by a rename script which is extra hassle > > > > so, in short: > > separation of a huge native lib into smaller libpacks - yes > > let them have the same name - many, some small and some large, NO!s > > > > Well, it's basicaly a tradeoff : Code duplication against ease of use for > some power users (I don't think that many people are moving around their > node on different CPUs/OSes). > > What do Higher Gods think ?
I understand how and why you want to do this. I just do not see it as a win. You are introducing a more complex install and a less flexible runtime. In the case where the installer downloads the cpuid based -ext library you have increased the complexity of the install for a small bandwidth saving. If -ext contains all the code and you just set a env variable there is no win - just complexity. In my mind simpler is better. As you say 'What do Higher Gods think ?' Thanks Ed
