New Features for Profiles
zhouhui at wam.umd.edu
Thu Nov 25 22:24:28 PST 2004
On Thu, Nov 25, 2004 at 11:51:20PM -0500, Thomas Pegg wrote:
>On Thu, 25 Nov 2004 20:23:37 -0500, Jeremy Huntwork <jeremy at jenacon.net> wrote:
>> Hello all,
>> Just wanted to put forward a few new features I'd like to see
>> implemented into the current nALFS profiles, if there are no objections.
>> 1) I'd like to move all the packages that the profiles expect to use in
>> the config_separate dir to bz2 format. First of all, it makes it
>> consistent and limits the possibility of mismatch between what you've
>> actually downloaded and what the profiles expect. Also, since we have at
>> least 2 fast and reliable ftp servers that house all our packages (in
>> bz2 format) it makes sense to me to use those as primary download
>> locations and make our profiles match. If there are any files that
>> aren't released as bz2, of course we can re-compress them and produce
>> our own new md5 sums, as I believe the ftp mirrors already do.
>That would be great. I know that Kevin proposed to start using the FTP
>mirrors a while back (on the website list I think), his biggest
>concern was bandwidth usage I believe. But really the amount of
>traffic ALFS would generate I think would be small, compared to LFS.
How about give the nALFS a little intelligence so it can figure out
what is the packagename in pkg.tar.gz or pkg.tar.bz2 and what is the
tar format and figure out how to unpack it upon seeing such archive?
>> 2) There are many I've talked to that *always* remove the md5sum feature
>> from the profiles. While I think it's good to leave them in by default,
>> I'd like to see at the very least a note in the README that provides a
>> sed command to remove the <digest> tags from the profile, and at most, a
>> shell script included to do that. In this way I hope to make it
>> *slightly* more optional to make use of the md5sums
>That one will be easy, I know I've got an old sed lying around here
>somewhere that did just that, a shell script would be even better.
Instead of modifing the profile, how about give nALFS an option so it
can optionally skip checking the checksum?
>> 3) I'd like the profiles to support parallel building, that is, passing
>> the -j flag to make. We could make a new entity in general.ent that
>> specifies the parallel build level and have it set by default to "".
>> Then, for all packages that are known to build sucessfully with a
>> reasonable parallel level (2-3 x the number of processors in the
>> machine) we can include param tags encasing the entity.
>I think I'll let you handle this one, I don't know a whole lot about
>parallel building. But it otherwise sounds like a good idea.
nALFS should be able to improvise on this option also, IMHO.
More information about the alfs-discuss