ollie lho <[EMAIL PROTECTED]> writes: > On Fri, 2002-06-28 at 13:44, Eric W. Biederman wrote: > > > > So I am probably going to take a stab at it. But I have to admit > > the more I look at grub the less configurable it feels. The backend > > of grub feels quite flexible, but the rest I really don't know. > > > > I don't know what your are refering as "backend". IMHO, the only > reusable part is the filesystem code. But unfortunately, it is not > as reusable as you or anyone would like to use in his code, if > possible. And the OS image loader is very terrible (your elf_boot > is much better).
>From 10,000 feet grub looks like a great bootloader. Lots of operating systems, lots of filesystems etc are supported. Even the files the code is put in look fairly clean. It is only when you really get close that things start to look bad. If nothing else grub appears to be a great proof of concept with a large user base. Given the open source rule that if your user base is large enough, and your developers sane enough all manner of problems can be overcome. What I really expect to get out of working closely with grub is a very sane reasoned explanation why it is easier to write something else than to expand on grub. My current stock pile: - The first of my objections is that it doesn't support loading anything below 1MB. - Doing the BIOS calls for an OS (as the multiboot spec wants to) is the wrong thing. - If grub wants to be a dynamic linker it should use file formats that were made for dynamic linking. Instead of inventing a very bad one of it's own. Program loading is a very well understood problem, and there is no reason for bootloaders to go inventing their own file formats. On a related note I believe I have found a very cheap maintainable solution to the problem of building compact ELF images, with interesting content that ld won't mess up. I believe I can use as to actually generate the header as binary data, link everything together with ld. And do ``objcopy -O binary'' to get the final image. Eric
