Legacy Loops Get Very Slow
When parsing a "standard" article, the legacy parser loop has time and time again (we kept canceling plans to remove it) proven to be fast compared to the BB-Code parser. However, since 2.2.0, nested templates have been permitted, and it has given rise to this reality. When using more than a single template per article, the internal (recursive) parsing that is necessary to reflect the intent of the nest substantially inflates the parsing time. In fact, the legacy parser did not really take these situations into account at all.
Keeping the tag list short for the noparse and template cycles really does make a difference, and they remain fairly efficient even for complex articles. Making this change has cut seconds from page creation time for these intricate articles.
The existing legacy parser has been completely removed for the next build. As discussed in a new feature thread, a more advanced legacy parser (that, this time, is truly ever called when turned on) is being developed.