gist-logs: truncate log files to be gist-friendly (#279)

Truncates each log file to about 1 MB. This avoids upload and
download errors with the gists, and respects GitHub's suggested
size limits.

Library/Homebrew/cmd/gist-logs.rb

@@ -91,6 +91,9 @@ module Homebrew
     logs = {}
     dir.children.sort.each do |file|
       contents = file.size? ? file.read : "empty log"
+      # small enough to avoid GitHub "unicorn" page-load-timeout errors
+      max_file_size = 1_000_000
+      contents = truncate_text_to_approximate_size(contents, max_file_size, :front_weight => 0.2)
       logs[file.basename.to_s] = { :content => contents }
     end if dir.exist?
     raise "No logs." if logs.empty?

Library/Homebrew/dev-cmd/test-bot.rb

@@ -45,10 +45,6 @@ module Homebrew
 
   HOMEBREW_TAP_REGEX = %r{^([\w-]+)/homebrew-([\w-]+)$}
 
-  def ruby_has_encoding?
-    String.method_defined?(:force_encoding)
-  end
-
   if ruby_has_encoding?
     def fix_encoding!(str)
       # Assume we are starting from a "mostly" UTF-8 string
@@ -1032,13 +1028,7 @@ module Homebrew
 
       # Truncate to 1MB to avoid hitting CI limits
       if output.bytesize > MAX_STEP_OUTPUT_SIZE
-        if ruby_has_encoding?
+        output = truncate_text_to_approximate_size(output, MAX_STEP_OUTPUT_SIZE, :front_weight => 0.0)
-          binary_output = output.force_encoding("BINARY")
-          output = binary_output.slice(-MAX_STEP_OUTPUT_SIZE, MAX_STEP_OUTPUT_SIZE)
-          fix_encoding!(output)
-        else
-          output = output.slice(-MAX_STEP_OUTPUT_SIZE, MAX_STEP_OUTPUT_SIZE)
-        end
         output = "truncated output to 1MB:\n" + output
       end
     end

Library/Homebrew/test/test_utils.rb

@@ -210,4 +210,17 @@ class UtilTests < Homebrew::TestCase
     assert_equal "1,000", number_readable(1_000)
     assert_equal "1,000,000", number_readable(1_000_000)
   end
+
+  def test_truncate_text_to_approximate_size
+    glue = "\n[...snip...]\n" # hard-coded copy from truncate_text_to_approximate_size
+    n = 20
+    long_s = "x" * 40
+    s = truncate_text_to_approximate_size(long_s, n)
+    assert_equal n, s.length
+    assert_match(/^x+#{Regexp.escape(glue)}x+$/, s)
+    s = truncate_text_to_approximate_size(long_s, n, :front_weight => 0.0)
+    assert_equal glue + ("x" * (n - glue.length)), s
+    s = truncate_text_to_approximate_size(long_s, n, :front_weight => 1.0)
+    assert_equal(("x" * (n - glue.length)) + glue, s)
+  end
 end

Library/Homebrew/utils.rb

@@ -534,3 +534,49 @@ def number_readable(number)
   (numstr.size - 3).step(1, -3) { |i| numstr.insert(i, ",") }
   numstr
 end
+
+# True if this version of Ruby supports text encodings in its strings
+def ruby_has_encoding?
+  String.method_defined?(:force_encoding)
+end
+
+# Truncates a text string to fit within a byte size constraint,
+# preserving character encoding validity. The returned string will
+# be not much longer than the specified max_bytes, though the exact
+# shortfall or overrun may vary.
+def truncate_text_to_approximate_size(s, max_bytes, options = {})
+  front_weight = options.fetch(:front_weight, 0.5)
+  if front_weight < 0.0 || front_weight > 1.0
+    raise "opts[:front_weight] must be between 0.0 and 1.0"
+  end
+  return s if s.bytesize <= max_bytes
+
+  glue = "\n[...snip...]\n"
+  max_bytes_in = [max_bytes - glue.bytesize, 1].max
+  if ruby_has_encoding?
+    bytes = s.dup.force_encoding("BINARY")
+    glue_bytes = glue.encode("BINARY")
+  else
+    bytes = s
+    glue_bytes = glue
+  end
+  n_front_bytes = (max_bytes_in * front_weight).floor
+  n_back_bytes = max_bytes_in - n_front_bytes
+  if n_front_bytes == 0
+    front = bytes[1..0]
+    back = bytes[-max_bytes_in..-1]
+  elsif n_back_bytes == 0
+    front = bytes[0..(max_bytes_in - 1)]
+    back = bytes[1..0]
+  else
+    front = bytes[0..(n_front_bytes - 1)]
+    back = bytes[-n_back_bytes..-1]
+  end
+  out = front + glue_bytes + back
+  if ruby_has_encoding?
+    out.force_encoding("UTF-8")
+    out.encode!("UTF-16", :invalid => :replace)
+    out.encode!("UTF-8")
+  end
+  out
+end