Just adding some data points, you might want to look at the recent efforts to adopt CLDR script names in glibc:
https://sourceware.org/ml/libc-locales/2016-q2/msg00034.html
and some *BSD variants:
https://wiki.freebsd.org/LocaleNewApproach

Now I understand what you mean.

BTW, I don't understand all the components of the locale identifiers as defined by http://git.savannah.gnu.org/cgit/gettext.git/tree/gettext-runtime/intl/finddomain.c#n71

For Unicode we have this definition: http://unicode.org/reports/tr35/#Unicode_language_identifier

Here's an example of an Unicode ID with all the fields:
it-Latn-IT-POSIX-NYNORSK

Where:
it: language subtag
Latn: script subtag
IT: region subtag
POSIX, NYNORSK: variant subtags

For gettext IDs, we have:
language[_territory][.codeset][@modifier]
What are the values of codeset and modifier? Do their meaning overlaps somehow with script/variant as defined by Unicode?

My code build alternative IDs looking for these chars: '.', '_', '-', '@'.
So, if locale is "foo@bar@baz", we have this list of IDs:
- "foo@bar@baz"
- "foo@bar"
- "foo"

Instead of storing copies of the string, I save the position of the stop chars. In the above case, we'll have these values:
- locale_lengths_count = 3
- locale_lengths[0] = 11
- locale_lengths[1] = 7
- locale_lengths[2] = 3

When looking for the plural rule, the code compares the IDs using each of these locale_lengths using xmlStrncmp and stops at the first match.

So, if someone specify "en@us", the code won't convert it to "en-us", but simply looks for "en@us" and "en".

Let's allow the CLDR plural rules lookup to fallback to "parent" locale ids.
For instance, looking for "it_IT.UTF-8" will also find "it_IT" and "it".