Messages in this thread |  | | Date | Fri, 14 Jun 1996 09:22:33 -0500 | From | (Andrew C. Esh) |
| |
>>>>> "Olaf" == Olaf Titz <olaf@bigred.inka.de> writes:
Aaron Tiensivu <tiensivu@pilot.msu.edu> wrote: >> o Ability to nuke source code that is not specific to your >> configuration. Not necessarily a default option, but an >> option, none-the-less.
Olaf> Split the architecture dependent parts off into separate Olaf> distributions. Okay, we've had this before, but with the Olaf> current kernel source in tar and feathers at way over 5 MB Olaf> and the guarantee that you'll _never_ need a substantial Olaf> portion of it, please reconsider.
Hi everyone!
If we split the source into separate distributions, then we stand the chance that someone will mismatch them. It would also be nice to have one central point which acts as root for the kernel mirroring tree, rather then one set of mirrors for the non-architecture source, and another (possibly less reliable) tree for each architecture portion. This may be fine for development groups, who are in-the-know, but it works against our popularity with the search-clueless general public.
The solution used by the X version of Netscape (and other Unix multi-platform products) is to create a directory for each version, and a tar.gz file for each platform. The file contains the entire distribution for that architecture. (I know: "Duh!"). What this does is use more space on the mirror machines, but it cuts down on the bandwidth needed to get a full source distribution which contains all the source for your architecture in one file. It's a no brainer for the user, as long as they know what machine they have.
The problem still to be solved with this is patches. I say we should continue to make patches non-architecture oriented, since they are small anyway, and have a pre-patch script which strips out the portions of the patch which are meant for each architecture you don't have. This keeps the regular patch program from emitting all sorts of errors when it tries to patch architecture files which are not present. It also keeps us from having to mirror 16-20 different patchfiles on a per-version, (multiplied by) per-architecture basis.
The script to split the main source tree into the different architecture .tar.gz files is trivial. I wouldn't mind switching into one extra directory and picking the linux-i386.tar.gz file from among the Sparc, HP, SGI, Mac, Alpha, AS400, Cray, and other files, especially if it means I only have to download a couple of megs, instead of 8 or 10.
What do you think?
(BTW: I would love nothing better than showing IBM what freedom can do by putting Linux on the AS400, and watching it become a significant portion of the OS market.)
--- Andrew C. Esh mailto:andrew_esh@cnt.com http://www.mtn.org/~andrewes - ACE Home Page
|  |