Toxic Elephant

Don't bury it in your back yard!

Private Toolbox: An Anti-Pattern

Posted by matijs 10/04/2016 at 09h21

This is an anti-pattern that has bitten me several times.

Suppose you have an object hierarchy, with a superclass Animal, and several subclasses, Worm, Snake, Dog, Centipede. The superclass defines the abstract concept move, which is realized in the subclasses in different ways, i.e., by slithering or walking. Suppose that due to other considerations, it makes no sense to derive Worm and Snake from a SlitheringAnimal, nor Dog and Centipede from a WalkingAnimal. Yet, the implementation of Worm#move and Snake#move have a lot in common, as do Dog#move and Centipede#move.

One way to solve this is to provide methods walk and slither in the superclass that can be used by the subclasses that need them. Because it makes no sense for all animals be able to walk and slither, these methods would need to be accessible only to subclasses (e.g., private in Ruby).

Thus, the superclass provides a toolbox of methods that can only be used by its subclasses to mix and match as they see fit: a Private Toolbox.

This may seem an attractive course of action, but in my experience, this becomes a terrible mess in practice.

Let’s examine what is wrong with this in more detail. I see four concrete problems:

  • It is not always clear at the point of method definition what a method’s purpose is.
  • Each subclass carries with it the baggage of extra private methods that neither it nor its subclasses actually use.
  • The superclass’ interface is effectively extended to its non-public methods,
  • New subclasses may need to share methods that are not available in the superclass.

The Animal superclass shouldn’t be responsible for the ability to slither and to move. If we need more modes, we may not always be able to add them to the superclass.

We could extract the modes of movement into separate helper classes, but in Ruby, it is more natural to create a module. Thus, there would be modules Walker and Slitherer, each included by the relevant subclasses of Animal. These modules could either define move directly, or define walk and slither. Because the methods added in the latter case would actually makes sense for the including classes, there is less need to make them private: Once could make a instance of Dog walk, either by calling move, or by calling walk directly.

This solves all four of Private Toolbox’ problems:

  • The module names reveal the purpose of the defined methods.
  • Subclasses that do not need a particular module’s methods do not include it.
  • The implementor of Animal is free to change its private methods.
  • If a new mode of transportation is needed, no changes to Animal are needed. Instead, a new module can be created that provides the relevant functionality.

Tags , , no comments no trackbacks

Minimally Intrusive SimpleCov Loading

Posted by matijs 02/04/2016 at 16h55

I always like extra developer tooling to be minimally intrusive, to avoid forcing it on others working with the same code. There are several aspects to this: Presence of extra gems in the bundle, presence and visibility of extra files in the repository, and presence of extra code in the project.

For this reason, I’ve been reluctant to introduce tools like guard or some of the Rails preloaders that came before Spring. On the other hand, no-one would be bothered by my occasional running of RuboCop, Reek or pronto.

In this light, I’ve always found SimpleCov a little too intrusive: It needs to be part of the bundle, and the normal way to set things up makes it rather prominently visible in your test or spec helper. Nothing too terrible, but I’d like to just come to a project, run something like simplecov rake spec, and have my coverage data.

I haven’t reached that blissful state of casual SimpleCov use yet, but I’m quite pleased with what we achieved for Reek.

Here’s what we did:

  • Add simplecov to the Gemfile
  • Add a .simplecov file with configuration:
SimpleCov.start do
  track_files 'lib/**/*.rb'
  # version.rb is loaded too early to test
  add_filter 'lib/reek/version.rb'
end

SimpleCov.at_exit do SimpleCov.result.format! SimpleCov.minimum_coverage 98.9 SimpleCov.minimum_coverage_by_file 81.4 end

  • Add -rsimplecov to the ruby_opts for our spec task:
RSpec::Core::RakeTask.new('spec') do |t|
  t.pattern = 'spec/reek/**/*_spec.rb'
  t.ruby_opts = ['-rsimplecov -Ilib -w']
end

This has several nice features:

First, there are no changes to spec_helper.rb. That file can get pretty cluttered, so the less has to be in there, the better.

Second, it only calculates coverage when running the full suite with rake spec. This means running just one spec file while developing won’t clobber your coverage data, and it makes running single specs a little faster since it doesn’t need to update the coverage reports.

Third, it enforces a minimum coverage per file and for the whole suite. The second point helps a lot in making this practical: Otherwise, running individual specs would almost always fail due to low coverage.

no comments no trackbacks

Repo size

Posted by matijs 25/09/2015 at 08h11

I just realized one important factor for attracting casual open source contributions is code/repo size. A huge repo is a barrier. So, it’s hugely important to either use off-the-shelf libraries, or split off parts of your code into their own components. These components need to live in their own repository, so no monorepo’s.

Of course, a high-status, high-visibility project can get away with more. Rails, for example, has all its components in one repository and does not seem to be lacking in contributions. On the other hand, for a long time Gnome required the full source for everything to be checked out and built together. This requires a serious commitment for even the most trivial bug fixes.

Why the sudden insight? A project I’m involved in has problems with wkhtmltopdf: The version that used to work crashes after a server upgrade, and the version that works has problems with fonts and images. A simple solution could be to just recompile the old version on the new server. However, because it essentially forks all of Qt, checking out the source will require 1GB of disk space, while building will require another 2.5GB (and a commensurate amount of time). This is not undertaken lightly.

no comments no trackbacks

Try to avoid try

Posted by matijs 28/07/2015 at 10h52

Because of a pull request I was working on, I had cause to benchmark activesupport’s #try. Here’s the code:

require 'benchmark'
require 'active_support/core_ext/object/try'

class Bar def foo

end end

class Foo

end

bar = Bar.new foo = Foo.new

n = 1000000 Benchmark.bmbm(15) do |x| x.report(straight) { n.times { bar.foo } } x.report(try - success) { n.times { bar.try(:foo) } } x.report(try - failure) { n.times { foo.try(:foo) } } x.report(try on nil) { n.times { nil.try(:foo) } } end

Here is a sample run:

Rehearsal ---------------------------------------------------
straight          0.150000   0.000000   0.150000 (  0.147271)
try - success     0.760000   0.000000   0.760000 (  0.762529)
try - failure     0.410000   0.000000   0.410000 (  0.413914)
try on nil        0.210000   0.000000   0.210000 (  0.207706)
------------------------------------------ total: 1.530000sec
                  user     system      total        real

straight 0.140000 0.000000 0.140000 ( 0.143235) try - success 0.740000 0.000000 0.740000 ( 0.742058) try - failure 0.380000 0.000000 0.380000 ( 0.379819) try on nil 0.210000 0.000000 0.210000 ( 0.207489)

Obviously, calling the method directly is much faster. I often see #try used defensively, without any reason warrented by the logic of the application. This makes the code harder to follow, and now this benchmark shows that this kind of cargo-culting can actually harm performance of the application in the long run.

Some more odd things stand out:

  • Succesful #try is slower than failed try plus a straight call. This is because #try actually does some checks and then calls #try! which does one of the checks all over again.
  • Calling #try on nil is slower than calling a nearly identical empty method on foo. I don’t really have an explanation for this, but it may have something to do with the fact that nil is a special built-in class that may have different logic for method lookup.

Bottom line: #try is pretty slow because it needs to do a lot of checking before actually calling the tried method. Try to avoid it if possible.

Tags , , no comments no trackbacks

In Ruby, negation is a method

Posted by matijs 30/01/2014 at 06h16

These past few days, I’ve been busy updating RipperRubyParser to make it compatible with RubyParser 3. This morning, I discovered that one thing that was changed from RubyParser 2 is the parsing of negations.

Before, !foo was parsed like this:

s(:not, s(:call, nil, :foo))

Now, !foo is parsed like this:

s(:call, s(:call, nil, :foo), :!)

That looks a lot like a method call. Could it be that in fact, it is a method call? Let’s see.

Tags , no comments no trackbacks

Things: A classification

Posted by matijs 19/01/2014 at 11h33

  • Things needed every day
  • Things needed every week
  • Things needed only during a certain season
  • Things needed for administrative purposes
  • Things kept for sentimental reasons
  • Thinks kept for beauty

Tags no comments no trackbacks

Some thoughts on Ruby's speed

Posted by matijs 02/03/2013 at 16h42

Yesterday, I read Alex Gaynor’s slides on dynamic language speed. It’s an interesting argument, but I’m not totally convinced.

At a high level, the argument is as follows, it seems:

  • For a comparable algorithm, Ruby et al. do much more work behind the scenes than ‘fast’ languages such as C.
  • In particular, they do a lot of memory allocation.
  • Therefore, we should add tools to those languages that allow us to do memory allocation more efficiently.

Tags , , , no comments no trackbacks

How I found a bug in GirFFI using Travis and Git

Posted by matijs 17/02/2013 at 20h15

I love Travis CI. I love git bisect. I used both recently to track down a bug in GirFFI.

Suddenly, builds were failing on JRuby. The problem did not occur on my own, 64 bit, machine, so it seemed hard to debug. I tried making Travis use different JVMs, but that didn’t help, apart from crashing in a different way (faster, too, which was nice).

Building a Travis box

Using the travis-boxes repository, I created a VM as used by Travis. This is currently not documented well in the READMEs, so I’m writing it down here, slightly out of order of actual events.

I cloned the following three repositories:

travis-cookbooks travis-boxes veewee-definitions

First, I created a base box in veewee-definitions, according to its README. In this case, I created a precise32 box, since that’s the box Travis uses for the builds. The final, export, stage creates a precise32.box file.

Then, I moved the precise32.box file to travis-boxes/boxes, making a base box available there. There is a Thor task to create just such a base box right there, but it doesn’t work, and seems to be deprecated anyway, since veewee is no longer supposed to be used in that repository.

So, a base box being available in travis-boxes, I used the following to create a fully functional box for testing Rubies:

bundle exec thor travis:box:build -b precise32 ruby

Oddly, this didn’t produce a box travis-ruby, but it did produce travis-development, which I could then manipulate using vagrant.

Hunting down the bug

I ssh’d into my fresh travis box using vagrant ssh. After a couple of minutes getting to know rvm (I use rbenv myself), I was able to confirm the crash on JRuby. After some initial poking around trying to pin down the problem to one particular test case and failing, I decided to use git bisect. As my check I used the test:introspection task, which reliably crashed when the problem was present.

While it’s possible to automate git bisect, I like to use it manually, since a particular test used may fail for unrelated reasons. Also, since git bisect is a really fast process, there is a pleasent lack of tedium.

Anyway, after a couple of iterations, I was able to locate the problematic commit. By checking the different bits of the commit I then found the culprit: I accidentally broke the code that creates layout definitions, in particular the one used by GValue. Going back to master, I added a simple test and fix. I will have to revisit the code later to clean it up and make it more robust.

Tags , , , , no comments no trackbacks

How many s-expression formats are there for Ruby?

Posted by matijs 04/11/2012 at 13h34

Once upon a time, there was only UnifiedRuby, a cleaned up representation of the Ruby AST.

Now, what do we have?

  • RubyParser before version 3; this is the UnifiedRuby format:

    RubyParser.new.parse "foobar(1, 2, 3)"
    # => s(:call, nil, :foobar, s(:arglist, s(:lit, 1), s(:lit, 2), s(:lit, 3)))
    
  • RubyParser version 3:

    Ruby18Parser.new.parse "foobar(1, 2, 3)"
    # => s(:call, nil, :foobar, s(:lit, 1), s(:lit, 2), s(:lit, 3))
    
    Ruby19Parser.new.parse "foobar(1, 2, 3)"
    # => s(:call, nil, :foobar, s(:lit, 1), s(:lit, 2), s(:lit, 3))
    
  • Rubinius; this is basically the UnifiedRuby format, but using Arrays.

      "foobar(1,2,3)".to_sexp
      # => [:call, nil, :foobar, [:arglist, [:lit, 1], [:lit, 2], [:lit, 3]]]
    
  • RipperRubyParser; a wrapper around Ripper producing UnifiedRuby:

      RipperRubyParser::Parser.new.parse "foobar(1,2,3)"
      # => s(:call, nil, :foobar, s(:arglist, s(:lit, 1), s(:lit, 2), s(:lit, 3)))
    

How do these fare with new Ruby 1.9 syntax? Let’s try hashes. RubyParser before version 3 and Rubinius (even in 1.9 mode) can’t handle this.

  • RubyParser 3:

      Ruby19Parser.new.parse "{a: 1}"
      # => s(:hash, s(:lit, :a), s(:lit, 1))
    
  • RipperRubyParser:

      RipperRubyParser::Parser.new.parse "{a: 1}"
      # => s(:hash, s(:lit, :a), s(:lit, 1))
    

And what about stabby lambda’s?

  • RubyParser 3:

      Ruby19Parser.new.parse "->{}"
      # => s(:iter, s(:call, nil, :lambda), 0, nil)
    
  • RipperRubyParser:

      RipperRubyParser::Parser.new.parse "->{}"
      # => s(:iter, s(:call, nil, :lambda, s(:arglist)),
      #      s(:masgn, s(:array)), s(:void_stmt))
    

That looks like a big difference, but this is just the degenerate case. When the lambda has some arguments and a body, the difference is minor:

  • RubyParser 3:

      Ruby19Parser.new.parse "->(a){foo}"
      # => s(:iter, s(:call, nil, :lambda),
      #      s(:lasgn, :a), s(:call, nil, :foo))
    
  • RipperRubyParser:

      RipperRubyParser::Parser.new.parse "->(a){foo}"
      # => s(:iter, s(:call, nil, :lambda, s(:arglist)),
      #      s(:lasgn, :a), s(:call, nil, :foo, s(:arglist)))
    

So, what’s the conclusion? For parsing Ruby 1.9 syntax, there are really only two options: RubyParser and RipperRubyParser. The latter stays closer to the UnifiedRuby format, but the difference is small.

RubyParser’s results are a little neater, so RipperRubyParser should probably conform to the same format. Reek can then be updated to use the cleaner format, and use either library for parsing.

Tags , , , no comments no trackbacks

Building a Simple Markdown Viewer with GirFFI

Posted by matijs 17/04/2012 at 07h41

This morning, I found myself looking for a simple markdown previewer that would run on the desktop. Using GirFFI, it was ridiculously easy to create it myself.

The simple version, based on the Webkit example in the GirFFI repository, goes something like this:

require 'ffi-gtk3'
require 'github/markup'

GirFFI.setup :WebKit, 3.0 Gtk.init win = Gtk::Window.new :toplevel scr = Gtk::ScrolledWindow.new nil, nil wv = WebKit::WebView.new win.add scr scr.add wv win.set_default_geometry 700, 500 win.show_all

file = ARGV[0] fullpath = File.expand_path(file, Dir.pwd) html = GitHub::Markup.render fullpath wv.load_string html, nil, nil, "file://#{fullpath}"

GObject.signal_connect(win, "destroy") { Gtk.main_quit } Gtk.main

I got the basic version working in about 10 minutes. The more complex version adds a keyboard handler to allow reloading the viewed file.

Tags , 2 comments no trackbacks