This is the first step in migrating the tool’s translations to
translatewiki.net. The py2mw function added to translations.py converts
our custom Python syntax to MediaWiki syntax. This conversion is lossy
in four aspects:

1. the Python syntax has variable names, the MediaWiki syntax only
   numeric indices;
2. MediaWiki has no equivalent of our !l (list) conversion;
3. the MediaWiki syntax does not include the plural tag names (“one”,
   “many”, “other”, etc.); and
4. the MediaWiki syntax does not include the gender markers (“m”, “f”,
   “n”).

The first two aspects can be mitigated by recording the variable names
and sets of list-type variables during the conversion and printing them
out at the end; they will later be hard-coded:

    {'duplicates_warning': ['lexemes'],
     'duplicates_instructions': ['lexemes'],
     'description_with_forms_and_senses': ['description', 'forms', 'senses'],
     'edit_ambiguous_warning': ['forms'],
     'edit_unmatched_warning': ['forms'],
     'edit_form_list_item': ['form_link',
                             'grammatical_feature_labels',
                             'statements']}

    {'edit_form_list_item': {'grammatical_feature_labels'}}

As for the plural tag names, MediaWiki takes them from the underlying
CLDR data and uses their “natural” order. In Python, the Babel library
gives us the tag names of a language as well – but it doesn’t give us
their order: we get an unordered set of tag names. However, it turns out
that order is consistent across all languages, as shown by this script
which extracts the order from the MediaWiki source files:

    xq -r '
      .supplementalData.plurals.pluralRules
      | .[]
      | .pluralRule
      | if type == "array" then . else [.] end
      | map(."@count")
      | join(" ")
    ' languages/data/plurals*.xml \
        | while read -ra plurals; do
        for i in ${!plurals[@]}; do
            for ((j=i+1; j<${#plurals[@]}; j++)); do
                printf '%s %s\n' "${plurals[i]}" "${plurals[j]}";
            done;
        done;
    done \
        | tsort

So we can hard-code this order later and have that covered as well.

Finally, the gender markers are similar to the plural tag names in that
they’re always “m”, “f”, and “n”, in that order. (MediaWiki only three
gender distinctions.)