{
    /* We assume that min computed val is 0, contiguous up to max_computed_val,
       then zero or more other absolute values, then smaller_val then larger_val. */
    unsigned const min_computed_val = 0;
    unsigned const larger_val = smaller_val + 1;
    g_return_if_fail(min_computed_val < max_computed_val);
    g_return_if_fail(max_computed_val < smaller_val);

    if (parent.set && !parent.inherit) {
        if (!child.set || child.inherit) {
            child.value = parent.value;
            child.set = parent.set;  // i.e. true
            child.inherit = parent.inherit;  // i.e. false
        } else if (child.value < smaller_val) {
            /* Child has absolute value: leave as is. */
        } else if ( ( child.value == smaller_val
                      && parent.value == larger_val )
                    || ( parent.value == smaller_val
                         && child.value == larger_val ) )
        {
            child.set = false;
            /*
             * Note that this can result in a change in computed value in the
             * rare case that the parent's setting was a no-op (i.e. if the
             * parent's parent's computed value was already ultra-condensed or
             * ultra-expanded).  However, I'd guess that the change is for the
             * better: I'd guess that if the properties were specified
             * relatively, then the intent is to counteract parent's effect.
             * I.e. I believe this is the best code even in that case.
             */
        } else if (child.value == parent.value) {
            /* Leave as is. */
            /** \todo
             * It's unclear what to do if style and parent specify the same
             * relative directive (narrower or wider).  We can either convert
             * to absolute specification or coalesce to a single relative
             * request (of half the strength of the original pair).
             *
             * Converting to a single level of relative specification is a
             * better choice if the newly-unlinked clone is itself cloned to
             * other contexts (inheriting different font stretchiness): it
             * would preserve the effect that it will be narrower than
             * the inherited context in each case.  The disadvantage is that
             * it will ~certainly affect the computed value of the
             * newly-unlinked clone.
             */
        } else {
            unsigned const parent_val = parent.computed;
            child.value = ( child.value == smaller_val
                            ? ( parent_val == min_computed_val
                                ? parent_val
                                : parent_val - 1 )
                            : ( parent_val == max_computed_val
                                ? parent_val
                                : parent_val + 1 ) );
            g_assert(child.value <= max_computed_val);
            child.inherit = false;
            g_assert(child.set);
        }
    }
}