1 files changed, 85 insertions, 0 deletions
diff --git a/sample/trick2015/kinaba/remarks.markdown b/sample/trick2015/kinaba/remarks.markdown
new file mode 100644
index 0000000000..6316c51fb6
--- /dev/null
+++ b/sample/trick2015/kinaba/remarks.markdown
@@ -0,0 +1,85 @@
+### Remarks
+
+Just run it with no argument:
+
+    $ ruby entry.rb
+
+I confirmed the following implementation/platform:
+
+- ruby 2.2.3p173 (2015-08-18 revision 51636) [x64-mingw32]
+
+
+### Description
+
+The program is a [Piphilology](https://en.wikipedia.org/wiki/Piphilology#Examples_in_English)
+suitable for Rubyists to memorize the digits of [Pi](https://en.wikipedia.org/wiki/Pi).
+
+In English, the poems for memorizing Pi start with a word consisting of 3-letters,
+1-letter, 4-letters, 1-letter, 5-letters, ... and so on. 10-letter words are used for the
+digit `0`. In Ruby, the lengths of the lexical tokens tell you the number.
+
+    $ ruby -r ripper -e \
+        'puts Ripper.tokenize(STDIN).grep(/\S/).map{|t|t.size%10}.join' < entry.rb
+    31415926535897932384626433832795028841971693993751058209749445923078164062862...
+
+The program also tells you the first 10000 digits of Pi, by running.
+
+    $ ruby entry.rb
+    31415926535897932384626433832795028841971693993751058209749445923078164062862...
+
+
+### Internals
+
+Random notes on what you might think interesting:
+
+- The 10000 digits output of Pi is seriously computed with no cheets. It is calculated
+  by the formula `Pi/2 = 1 + 1/3 + 1/3*2/5 + 1/3*2/5*3/7 + 1/3*2/5*3/7*4/9 + ...`.
+
+- Lexical tokens are not just space-separated units. For instance, `a*b + cdef` does
+  not represent [3,1,4]; rather it's [1,1,1,1,4]. The token length
+  burden imposes hard constraints on what we can write.
+
+- That said, Pi is [believed](https://en.wikipedia.org/wiki/Normal_number) to contain
+  all digit sequences in it. If so, you can find any program inside Pi in theory.
+  In practice it isn't that easy particularly under the TRICK's 4096-char
+  limit rule. Suppose we want to embed `g += hij`. We have to find [1,2,3] from Pi.
+  Assuming uniform distribution, it occurs once in 1000 digits, which already consumes
+  5000 chars in average to reach the point. We need some TRICK.
+
+  - `alias` of global variables was useful. It allows me to access the same value from
+    different token-length positions.
+
+  - `srand` was amazingly useful. Since it returns the "previous seed", the token-length
+    `5` essentially becomes a value-store that can be written without waiting for the
+    1-letter token `=`.
+
+- Combination of these techniques leads to a carefully chosen 77-token Pi computation
+  program (quoted below), which is embeddable to the first 242 tokens of Pi.
+  Though the remaining 165 tokens are just no-op fillers, it's not so bad compared to
+  the 1000/3 = 333x blowup mentioned above.
+
+
+    big, temp = Array 100000000**0x04e2
+    srand big
+    alias $curTerm $initTerm
+    big += big
+    init ||= big
+    $counter ||= 02
+    while 0x00012345 >= $counter
+      numbase = 0x0000
+      $initTerm ||= Integer srand * 0x00000002
+      srand $counter += 0x00000001
+      $sigmaTerm ||= init
+      $curTerm /= srand
+      pi, = Integer $sigmaTerm
+      $counter += 1
+      srand +big && $counter >> 0b1
+      num = numbase |= srand
+      $sigmaTerm += $curTerm
+      pi += 3_3_1_3_8
+      $curTerm *= num
+    end
+    print pi
+
+- By the way, what's the blowup ratio of the final code, then?
+  It's 242/77, whose first three digits are, of course, 3.14.