- micro-benchmarks are really good at telling you the story you want to hear.
- profiling erector revealed that it placed a lot of pressure on the garbage collector. This was back when it too was the "fastest templating library." Micro-benchmarks won't really tell that story.
Other issues:
- blocks are a good abstraction until they aren't. No doubt backtraces are better than with erector. This was seriously traumatic to debug, templates were a gigantic ball of mud.
- "_I_ don't need caching" isn't a counter-argument. The reality of using blocks to nest elements is that the boundary between "cacheable, static content" and "dynamic stuff" is indistinguishable.
It's not like caching is turned off in erector (or fortitude). People were constantly making changes to templates, shipping them, and having changes not show up because memcache isn't running in development.
You COULD try to argue that it's the same problem in erb or slim, except it mostly isn't. When ALL your code is ruby code, seeing the "cache below this line" is a lot harder.