Package download statistics

John Marino dragonflybsd at marino.st
Mon Jun 3 14:41:28 PDT 2013


On 6/3/2013 23:04, Rumko wrote:
> John Marino wrote:
>
>> Could it be that they had lots of issues with that
>> in the past?  Who wouldn't want something fast and quick if it were
>> trustworthy?  The "indication" alone is telling.
>
> Well, as above, the machine (admin?) specific options are to blame for at least
> a few (absolutely no idea how many in %), even fbsd ports/dports (IIRC) only
> build packages for the default options, so when you play around you can
> quickly doom yourself to source-only.

Yes, I think it's implicit that pre-built binary users are accepting the 
default set of options and extending that thought, they are trusting the 
pkgsrc/ports developers to select the defaults wisely.

> If I understand correctly, obsd ports seem to handle this way better than
> pkgsrc and (d/freebsd)ports, they seem to have these so-called flavors which
> actually tell what options are turned on/off. And IIRC ('twas quite a few
> years ago), the actual flavor was part of the package name, so theoretically,
> you could precompile all possible combinations of packages and their flavors.
>
> It would be quite interesting to see sth like that being put into
> pkgsrc/dports.

That would be an exponential explosion unfortunately.  A port with only 
4 options could be built 15 different ways.  Some ports have 20 options. 
  A port like mongodb takes 1 hour to build and is a few hundred Mb in a 
package.  So the number of packages would literally be in the millions 
and the amount of data required would fill up some Terrabyte drives I 
would think.  It's just not practical to do it automatically.

That leaves hand-selection.  To some extent ports and pkgsrc already 
does this (e.g. vim, vim-lite, emacs, emacs-nox11, etc).

The 1 flavor with good options seems to the best compromise as far as I 
can tell.

John




More information about the Users mailing list