ryniker at alum.mit.edu
Thu Nov 26 03:55:45 UTC 2009
If the goal is to separate locale-dependent data from packages that now
contain data for multiple locales, it seems reasonable in many cases to
combine locale-dependent data from multiple packages into single, new,
If there are 5000 packages with locale-dependent data, and 100 locales,
creation of individual, new, locale-specific packages could produce
500,000 new packages - the explosion that rightly concerns Mr. Sundaram.
Of course, all those 5000 packages might not have data for all 100
locales, but any significant fraction of 500,000 is a nightmare.
A different approach would be more manageable: combine locale-dependent
data for many packages into one locale_data package. In the minimum
case, this would result in only 100 new packages, each one containing
data for one locale from all 5000 original packages.
A more practical scheme would probably group those 5000 packages into a
small number of categories (perhaps aligned with translation group,
distribution organization, etc.). There would then be a small_number *
100 new packages, instead of a Million Package March.
Organization of packages in this way makes addition of a new locale
straightforward: just install the package (or small number of packages)
for the desired locale.
Is this worth doing? I don't know. If enough people see the goal (a
system contains data only for the desired locales) as worthwhile, they
can do this for a small set of packages to learn what problems may
manifest and what techniques are efficient to manage the data. If a
standard, easy technology is developed, package authors or maintainers
may be willing to move locale-dependent data out of their packages.
More information about the fedora-test-list