bump version, describe zordoz changes

Author: bennn

info.rkt

@@ -3,7 +3,7 @@
 (define deps '("base" "typed-racket-lib" "typed-racket-more" "require-typed-check"))
 (define build-deps '("scribble-lib" "racket-doc" "rackunit-lib" "typed-racket-doc" "at-exp-lib" "gtp-util" "pict-lib" "scribble-abbrevs" "syntax-sloc" "with-cache"))
 (define pkg-desc "Gradual typing benchmark programs")
-(define version "2.0")
+(define version "3.0")
 (define pkg-authors '(ben))
 (define scribblings '(("scribblings/gtp-benchmarks.scrbl" () ("Benchmarks"))))
 (define compile-omit-paths '("benchmarks"))

scribblings/gtp-benchmarks.scrbl

@@ -73,6 +73,20 @@ To run a benchmark:
 @section[#:tag "gtp:history"]{Version Notes}
 
 @itemlist[
+  @item[
+    @history[#:changed "3.0"
+             @elem{Fixed an issue in the untyped @bm{zordoz} code.
+                   Before, the untyped code imported the typed version of the @tt{compiler/zo-structs} library.
+                   After, the untyped code avoids this unnecessary type boundary.
+                   This issue seriously affects the conclusion about @bm{zordoz}
+                    reached in @cite{JFP-2019}.
+                   That paper compares one version of @bm{zordoz} that does not
+                    suffer the issue (6.2) against two that do (6.3 and 6.4),
+                    and incorrectly concludes that changes to Typed Racket
+                    improved the overhead in the later versions.
+                   In fact, the overhead is better because the untyped code
+                    got unnecessarily slower.
+                   More at: @url{https://github.com/bennn/gtp-benchmarks/issues/16}.  }]]
   @item[
     @history[#:changed "2.0"
              @elem{Fixed a difference between the typed and untyped @bm{mbta} code.