Fix buf error on interrupt.

[samuel-williams-shopify]

I think this fixes #57.

[samuel-williams-shopify]

@ianks can you please take a look? You investigated this some years ago.

[samuel-williams-shopify]

If you add this line:

        if (err != Z_OK) fprintf(stderr, "[ZLIB] zstream_run_try: Z_BUF_ERROR with avail_out > 0, err=%d interrupt=%d\n", err, args->interrupt);

(instead of err = Z_OK) and run one of the reproductions, you can see the issue manifesting:

~/D/r/zlib (fix-buf-error-on-interrupt) [1]> ./test.rb
[SIGINT]
........................................................................................................[SIGINT]
[ZLIB] zstream_run_try: Z_BUF_ERROR with avail_out > 0, err=-5 interrupt=0
./test.rb:17:in 'Zlib::Deflate#deflate': buffer error (Zlib::BufError)
  from ./test.rb:17:in 'block (2 levels) in <main>'
  from ./test.rb:15:in 'Integer#times'
  from ./test.rb:15:in 'block in <main>'
  from ./test.rb:13:in 'Kernel#fork'
  from ./test.rb:13:in '<main>'
✗ Bug reproduced: #<Process::Status: pid 11347 exit 1>

~/D/r/zlib (fix-buf-error-on-interrupt) [1]> ./test2.rb
[ZLIB] zstream_run_try: Z_BUF_ERROR with avail_out > 0, err=-5 interrupt=0
#<Thread:0x0000000104f6c8e0 ./test2.rb:11 run> terminated with exception (report_on_exception is true):
./test2.rb:13:in 'Zlib::GzipReader#initialize': buffer error (Zlib::BufError)
  from ./test2.rb:13:in 'block (2 levels) in <main>'
  from ./test2.rb:12:in 'Kernel#loop'
  from ./test2.rb:12:in 'block in <main>'
./test2.rb:18:in 'Thread#wakeup': killed thread (ThreadError)
  from ./test2.rb:18:in 'block in <main>'
  from ./test2.rb:17:in 'Kernel#loop'
  from ./test2.rb:17:in '<main>'

test.rb

#!/usr/bin/env ruby
# frozen_string_literal: true

# Simple test to reproduce signal interrupt bug
# Run: ruby test.rb

require 'zlib'
require 'securerandom'

z = Zlib::Deflate.new
trap('INT') { $stderr.puts "[SIGINT]" }

pid = fork do
  # Run deflate calls:
  1000.times do |i|
    data = SecureRandom.random_bytes(100)
    z.deflate(data, Zlib::SYNC_FLUSH)
    $stdout.write(".")
  end
end

# Send signals
5.times { |i| Process.kill('INT', pid); sleep 0.001 }

status = Process.wait2(pid)[1]
if status.success?
  puts "✓ Test passed"
  exit 0
else
  puts "✗ Bug reproduced: #{status.inspect}"
  exit 1
end

# ./test.rb:19:in 'Zlib::Deflate#deflate': buffer error (Zlib::BufError)
#   from ./test.rb:19:in 'block (2 levels) in <main>'
#   from ./test.rb:17:in 'Integer#times'
#   from ./test.rb:17:in 'block in <main>'
#   from ./test.rb:12:in 'Kernel#fork'
#   from ./test.rb:12:in '<main>'
# ✗ Bug reproduced!

test2.rb

#!/usr/bin/env ruby
# frozen_string_literal: true

require "securerandom"
require "stringio"
require "zlib"

content = SecureRandom.base64(5000)
gzipped = Zlib.gzip(content)

thr = Thread.new do
  loop do
    Zlib::GzipReader.new(StringIO.new(gzipped)).read
  end
end

loop do
  thr.wakeup
end

[ioquatix]

Okay, so I came to the same conclusion as everyone else, basically we are retrying on interrupt when we shouldn't. So I tightened up the condition. Not sure if there are other cases, but for sure we should not retry unless there is available input and available output space.

[ianks]

LGTM