What motivated this article is generating "rich" protobufs for model graphs about 3 levels deep, and we render long lists at $work. Since there were some opport

What motivated this article is generating "rich" protobufs for model graphs about 3 levels deep, and we render long lists at $work. Since there were some opportunities for caching I was wondering whether this could be optimised in some way. Interestingly enough, the "use the branch of templates as part of the cache key" made things harder for us, not easier (as we were rendering very similar proto messages over and over)