TDD is not the Failure, our Culture of Development Is

In the article TDD is dead.  Long live testing.  and a subsequent response The pitfalls of Test-Driven Development, both authors, in my opinion, are missing the mark by a mile. In the real world, walking into an existing code base, the reason you need TDD (but can’t use it) is because programmers didn’t spend sufficient time on good architecture practices, resulting in code that is an entangled morass of intertwined concerns, and yes, as one of the author’s points out, because the code itself was never intended to be tested except through, at best, acceptance test procedures (most likely the pen and paper variety — with the developer watching in the background hoping he rolls a 20.)  There is an overall lack of attention paid to designing a component with sufficient abstraction that it can accommodate changing requirements, and little or no attention paid to separating (decoupling) the behaviors between components such that component A shouldn’t ever care about what component B is doing. We have a philosophy of refactoring with “just get the minimum to work” to blame for that — the days of thinking about architecture and abstraction are long gone.

The metaphor of building a sky-scraper is inaccurate because anyone building a sky-scraper would know that you can’t make the walls of the first floor so weak that they can’t support a second floor. Except it is accurate because, not paying attention to the requirements and living in a “do the minimum work” philosophy, promoted by the likes of Kent Beck’s Agile Programming and Martin Fowler’s refactoring philosophies (along with their mutant child Extreme Programming), this actually is exactly what ends up happening, and thus TDD is an absolute necessary fallout of a broken coding paradigm. A more accurate metaphor would have been, the requirements called for a single story building, then the requirements changed. Again, with sufficient abstraction up front, the straw-bale walls could be replaced with titanium reinforced hay quite easily.

As for Rails (or rather Ruby, or rather any duck-typed language), TDD is again essential because duck-typing allows for variances in the behavior at runtime, both of type and function calls. The non-strictness of duck-typing is leveraged in lieu of good object-oriented design–why create sub-classes when I can just pass in an instance that quacks just like the other “ducks.” While object-oriented programming cannot be done well without object-oriented design (and yes, I’ve seen both the “P” and the “D” done horribly and have done it horribly myself) a duck-type language allows the programmer to completely eliminate the “D” — class, method, quack, quack. Perhaps we should take a clue from the name, “duck-typing”, that it is actually quackery, and like medical quackery, promises rapid “feel good” development that you end up paying for in little fragments of time running unit and integration tests because you didn’t do the necessary up front design, you haven’t clearly abstracted how the rules are handled, you haven’t clearly decoupled the behaviors of complex systems to identify the dependencies (which all complex systems will have.) If you add up all the time spent running those tests (which of course spawned whole new technologies to run those tests faster and faster) you will discover that over the lifetime of the product, you spent far more time watching little green bars (or red ones) than you would have spent on solid up-front architecture, particularly in the areas of abstraction. But nay, it quacks, and it’s fast. At first.

As for “the industry’s sorry lack of automated, regression testing”, here, let’s not blame the programmer directly even though they consistently over-promise and under-estimate, but rather, let’s blame a culture, yes, starting with “the geek” but also placing responsibility firmly on the business practices of management and the continually demonstrated lack of understanding of the importance of regression testing, and the time & cost that developing regression tests and even more costly, maintaining those tests, requires. As with essentially all other aspects of our society, we are living in a constant tension between short-term gains and long-term investment (TDD can be considered an investment) and we all know which side is winning. We have a culture that rewards quick results and punishes the methodical and (seen as) slow thinker. Some of this may be justifiable due to market pressures and real budgetary constraints, but what is lacking is the consciousness to balance planning and activity.  So what we have instead is a knee-jerk culture oriented to quick results (Agile Programming and Refactoring) which, to support a broken development paradigm, demands TDD as the “fix”, but nobody seems sees that.

When one of the author’s writes “I have yet to see a concrete illustration of how to use Test-Driven Development to test a back-end system interacting with a 20-year-old mainframe validating credit card transactions” I laugh because I have done just that — CC validation systems all have the means of simulating transactions, it’s actually trivial to write TDD’s against such systems. Yes, the author does have a point that much of the “…source code [encountered in legacy systems]…was never designed to be tested in the first place…”, but again, that’s missing the point — TDD is clearly the wrong tool for those systems. TDD works best in an environment:

  • lacking architecture,
  • most likely using duck-typing languages,
  • and, most importantly, one that has started from ground zero with testing as one of the coding requirements.

This is independent of whether it’s a 200 line gem (as in Ruby library as opposed to a “great thing”) or a 100 million line application. If those 100 million lines were written with the intention of being unit / feature tested, then there is no problem. Except that it’s TDD and probably not well architected out of the gate.  And let me be clear that TDD, when applied to a well architected application, is a perfectly valid and beneficial practice, but then TDD is also simplified because it ends up testing behaviors, not architectural flaws and geese trying to pretend to be ducks.

So, is the failure TDD? No. The failure is in a culture entrenched at all levels of software development that says that Agile Programming and Refactoring can replace thinking and it’s brothers “design” and “planning.”  We have Kent Beck and Martin Fowler (I commit sacrilege in criticizing the gods) to squarely blame for that “regression”, excuse me, “story.”  And it’s those two aspects (pun intended) of programming that should be given the boot, not TDD, which has a validity in and of itself under the right conditions.  However, even Agile & refactoring are merely symptoms  (or victims) of a cultural disease that demotes thinking (we need only look at our K-12 education systems and Common Core for proof) and promotes the short-term gain (as demonstrated by our economic, medical, agricultural, etc. practices.)

Beware of Ruby’s string split function

Observe this behavior:

split

 

 

 

And the description in the documentation:

“If the limit parameter is omitted, trailing null fields are suppressed. If limit is a positive number, at most that number of fields will be returned (if limit is 1, the entire string is returned as the only entry in an array). If negative, there is no limit to the number of fields returned, and trailing null fields are not suppressed.”

Now, while I can somewhat understand this behavior, it is certainly in opposition with regards to the behavior of the split function in other languages that I’ve used.  Ruby’s default behavior can cause serious problems when parsing CSV files and auto-populating fields where you expect empty strings rather than nils.

Which brings up the next point:

split2

 

 

When I index outside of the array length, no exception is thrown.  Come on, Ruby!  That’s just bad form.  Again, the Ruby documentation for array says:

To raise an error for indices outside of the array bounds or else to provide a default value when that happens, you can use fetch.

Sigh.

 

LibXML — empty nodes (and the libxml-rails gem)

I was going to title this post “LibXML — how to fail right out of the box” but then thought a more accurately description of the problem might be better.  There is something I don’t understand about the open source community: its tolerance.  I encountered this problem right out of the gate:

The XML:

<?xml version="1.0" encoding="UTF-8"?>
<root_node>
<elem1 attr1="val1" attr2="val2"/>
<elem2 attr1="val1" attr2="val2"/>
<elem3 attr="baz">
  <elem4/>
  <elem5>
    <elem6>Content for element 6</elem6>
  </elem5>
</elem3>
</root_node>

The resulting root children:

libxml-1

 

 

 

 

Do tell me why whitespace and CRLF’s are seen as empty nodes?  Do tell me why this is absurd behavior is tolerated as the default?  I had to google for some indication as to what’s going on.  This fixes the problem (Ruby code):

doc = XML::Document.file('foo.xml', :
  encoding => XML::Encoding::UTF_8, 
  :options => LibXML::XML::Parser::Options::NOBLANKS)

Other than that, it looks like a decent enough package, though I haven’t explored it further.

The libxml-ruby gem

The libxml-ruby gem worked fine, but I did have to, as the documentation says, copy three binaries into the one of the directories in the Windows path.  Happily, the gem comes with the precompiled binaries – that’s a real help and kudos to the gem authors for providing the MinGW32 binaries.

Add routes in Rails dynamically at runtime

After poking around some more, I answered my question in my previous post — “when I add a route at runtime, do they add to the route list or replace existing ones?”  The answer is that they are added to the route list.  So here’s the code to add the route once by checking to see if it already exists.  If not, the route is added, if it does exist, we verify that the controller, method, and verb are identical.

  # Add the route if it doesn't exist.
  # Example: add_route('post', 'sign_in', 'users', 'sign_in') 
  def add_route(verb, url, controller, method)
    route = controller + '_' + method
    if Rails.application.routes.named_routes.routes[route.to_sym].nil?
      Rails.application.routes.disable_clear_and_finalize = true
      Rails.application.routes.draw do
        meth = method(verb.downcase)
        meth.call("/#{url}" => "#{controller}##{method}", :as => "#{controller}_#{method}")
      end
    else
      # Check that we're adding the same verb and route.
      # TODO: Verify the URL is the same too.
      route_verb = Rails.application.routes.named_routes.routes[route.to_sym].verb
      unless route_verb.to_s.upcase =~ /#{verb.upcase}/
        raise "Attempting to use a different verb #{verb} for route #{controller}_#{method}."
      end
    end
  end

 

 

Routes, form_for, the HTML form tag, and dynamic routes

Part 4 of n of the series “Never write HTML, Javascript, or CSS again”

Let’s look at a simple form_for:

<%= form_for @user do |f| %>
  <%= f.label :name %>:
  <%= f.text_field :name %><br />
  <%= f.submit %>
<% end %>

Interestingly, the default behavior does not require a route for the
button (which by default is labelled “Create User”) to work.  When we click
on this button, Rails calls the “create” method:

class UsersController < ApplicationController
  def create
    # ... do something
  end
end

That’s pretty interesting.  Now let’s tweak the form_for a
bit by introducing a desired action:

<%= form_for @user, :url => { :action => "upload" } do |f| %>
  <%= f.label :name %>:
  <%= f.text_field :name %><br />
  <%= f.submit %>
<% end %>

Now that we’re specifying a URL (and what, you may ask, does this have to do
with the action?) we now need a route.  If we don’t provide a route, we get
this message:

No route matches {:action=>"upload", :controller=>"home"}

Interesting again, because Rails is also telling us the controller that it
expects to handle the route “/upload”.  Because I created the view for this
demo page in home and therefore we also have a HomeController defining the page
(I called it “test2”):

class HomeController < ApplicationController
  def test2
    @user = User.new()
  end
end

and, the accompanying route:

get "/test2" => "home#test2"

we can assume that Rails is making a stab at the controller because that’s
the controller that invoked the page.  Rails is making the assumption that
the controller should be HomeController.  Interestingly, it did NOT make
this assumption earlier.

So, let’s provide this route:

post "/upload" => "home#upload"

and the accompanying method (in HomeController):

def upload
  # do something
end

and all is well.

But we don’t want this route vectoring off to the HomeController.  We
want it to go to the UsersController like it was before we started finagling
with the URL.  So we provide an override:

<%= form_for @user, :url => { :controller => "users", :action => "upload" } do |f| %>
  <%= f.label :name %>:
  <%= f.text_field :name %><br />
  <%= f.submit %>
<% end %>

and yes, we would expect to get:

No route matches {:controller=>"users", :action=>"upload"}

so once again, we finagle the route:

post "/upload" => "users#upload"

and write the method in UsersController:

class UsersController < ApplicationController
  def upload
    # Do something
  end
end

and again all is well.

But let’s look at the generated HTML now.  It looks like this:

<form accept-charset="UTF-8" action="/upload" class="new_user" id="new_user" method="post">
  <label for="user_name">Name</label>:
  <input id="user_name" name="user[name]" size="30" type="text" /><br />
  <input name="commit" type="submit" value="Create User" />
</form>

Notice there is nothing that indicates the controller in the markup.

So, what Rails is doing, when it parses the form_for tag, is it is matching
what is on the right-hand side of the route action with what you specify in the
:url => { } hash.  Now, wait a minute.  This seems
unnecessarily invasive.  Why should form_for care what controller is
handling the action route?  It’s not like you can specify two different
controllers for the same route!  Ironically, if you do, Rails doesn’t
throw an exception
(at least 3.2.17, which is what I’m testing with) — in
fact, it will always use the first matching route’s controller and method
mapping!

Now, let’s make matters more interesting.  The Airity DSL isn’t going to
generate a form_for Rails tag, it will be generating the HTML form
tag, like in the HTML example above.  And since there’s no controller being
specified in the HTML, the controller that receives the “action” will be
determined by the route.  If I say home#upload, then the HomeController#upload
function is called, if I say "users#upload, then the UsersController#upload
function is called.

So why does Rails introduce this complexity of making an assumption of what
the default route is (and making an incorrect assumption) and secondly,
requiring a controller hash, which it then enforces that the route is mapped to
the same controller?  These conflicts that can occur between the route
definition and the form_for tag can make for (quite unnecessary, in my opinion)
heaps of confusion.  First, it shouldn’t make any assumptions, second, it
should require a controller hash, and third, it should just let the route
specify the controller.

However!

In my DSL, why not define the controller and method that should handle the
button click?  (Notice we’re dealing with the form tag, not the input
button — the page that receives the data is determined by the action
attribute in the
form tag.)  Why should I have to edit the routes.rb file
when I create forms in the DSL?  This adds complexity and unnecessary
decoupling of action with the controller and method that handles the action.
Unfortunately, the solutions are non-trivial.  The basic solution, the “catch
all
“, can have undesirable side-effects.  A

more robust solution
by Michael Lang is Rails 4 dependent — in other words, it looks like
that there is no common solution (other than the “catch all”, that is Rails
version independent.

Then I stumbled across

this post
which looks to me like it would be Rails agnostic, as it is adds a
route to the routes table with Rails.application.routes.draw do…end.
Looking at this implementation, it looks very similar to what Michael Lang’s
Rails4 dynamic router is doing.  So let’s give it a try.

In the DSL, I want to specify the Sign In action:

html_dsl.form("user", {action: 'sign_in'}) do

and after modifying the DSL code a bit to handle this new hash, we can see
the generated HTML:

<form id="new_user" action="/sign_in" class="new_user" method="post">

and indeed, clicking on the Sign In button gives us the expected routing
error:

No route matches [POST] "/sign_in"

Now we try to specify the handler, let’s hard-code it for the moment:

def sign_in_markup(html_dsl, fz_dsl, styles)
  Rails.application.routes.draw do
    post "/sign_in" => "users#sign_in", :as => "users_sign_in"
  end

  html_dsl.div({id: 'sign_in_page', styles: ['display: none']}) do
    html_dsl.form("user", {action: 'sign_in'}) do
      ....

what you’ll notice is that this initially works.  But do a refresh on
the home page, and you get:

No route matches [GET] "/"

Whoops!  What happened to the routes that were defined in
routes.rb?

If we add the mysterious
Rails.application.routes.disable_clear_and_finalize = true before the “do”
statement, lo-and-behold, it works!

That was remarkably simple, yet a few questions remain — how exactly is this
working, are there unintentional side-effects, does this work with Rails 4, and
are these route additions cumulative, or do they replace any existing
definition?

I’m sure these questions will be explored in more depth at some point!

Recaptcha, Foundation, Sass, and the “different prefix” error

Recaptcha and Foundation

After creating a test project, I finally connected the dots that my inclusion of foundation_and_overrides.scss was causing the problem with the rendering of the recaptcha block (see my previous post.)  A quick google search led me here where DinkoMiletic provides a good solution:

#recaptcha_area input[type="text"] {
  display: inline-block;
  height: auto;
}

Disturbingly, “thedeerchild” writes: ” It’s not really feasible to pull out a core part of our styling due to third party compatibility issues, and it’s fairly easy to overwrite the Foundation styles to fix reCaptcha.”

Now, personally, I think that’s a cop out.  If you’re going to write something as “foundational” (pun intended) as a presentation layer styler, then it is damn well your responsibility when some third party component has compatibility issues, especially one so prevalent as Google’s recaptcha.   Now, granted, I’m not using the most recent version of Foundation, so perhaps the issue has been fixed, but at a minimum, I think these kinds of compatibility issues should be easily found in the documentation, not buried in some forum post.

SASS and the “Different Prefix” Error

However, when I set up my new test project, I started getting the following error:

ActionView::Template::Error (different prefix: "C:/" and "E:/rails-projects/airity/app/assets/stylesheets"

I have never, until this morning, had a problem running rails projects on a different drive.  Now, this morning, this happens.  As others suggested, moving the project back to the C drive where I RailsInstaller did its thing is a solution.  While I came across a few different “solutions” (none of which worked and some of which required forking the sass-rails gem), the only solution that worked and allowed me to keep the project on the desired drive was to use a simlink.

Because this problem appears to be common with other gems (like bootstrap-sass, read here), I am definitely interested in a global solution rather than something that fixes the problem for just one issue.  The mklink command was for some reason undecipherable to me, so as per a suggestion here, I downloaded Directory Linker and filled in:

Link Location: the new folder name on the C drive that I want.  In my case “c:\rails-projects\airity”

Link To: The existing folder name on my E drive.  In my case “e:\rails-projects\airity”

Click Go!, it does it’s thing in a second or so, then verify that the project files and folders exist on the C drive.

After reloading the project in RubyMine from the C drive, lo-and-behold, no more problems with the prefix.

Now if I could only bill the 4 hours I spent this morning working around problems that someone else caused by some update (which I was completely unable to track down, even after setting specific versions of sass and sass-rails based on my laptop version numbers).

Then again, those people that have suffered through this and found solutions which I then used, well, they should be the ones getting the money.

So, now it’s almost 3PM and I’m finally able to get to what I wanted to at 9 this morning (yes, there was lunch and grocery run in between.)  And yes, the reCAPTCHA block now works with Foundation:

recap1