Ironic, considering the current refactoring (not yet committed) of MemberBuilder to split validation logic into MemberBuilderValidator was partially to be able to easily override the fallback logic. It's a useful refactoring nonetheless, but it could have waited.
As mentioned in a prior commit blog-like entry, many of the tests evolved into more of an integration or system-level type of test. Let's get away from that.
This is the first test case to use the new basic xUnit-style system. This
system is likely to evolve over time. Right now it's purely for
setUp, organizational and output purposes.
The one year anniversary of the beginning of the ease.js project is quickly
approaching. I find myself to be not quite where I had expected many months ago,
but find that the project has evolved so much further than I had event
originally anticipated. My main motivation behind the project continues to be
making my life at work easier, while providing an excellent library that others
can hopefully benefit from. If anything, it's a fascinating experiment and
clever hack around JavaScript.
Now I find myself with a newborn child (nearly four weeks old), who demands my
constant attention (and indeed, it is difficult to find the desire to put my
attention elsewhere). Still - I am a hacker. Software is my passion. So the
project must move forward.
I also find myself unwilling to create a blog for ease.js. I feel it's
inappropriate for a project that's in its (relative) infancy and does not have
much popularity (it has never been announced to anyone). As such, I feel that
commit messages will serve my purpose as useful journal entries regarding the
status of the project. They will also be interesting easter eggs for those who
would wish to seek them out for additional perspective on the project. (Granted,
one could easy script the discovery of such entries by examining the absurd
length of the commit message...perhaps the git log manpages would be useful).
So. Let's get back to the project.
ease.js is currently going through a strong refactoring in order to address
design issues that have begun to creep up as the project grew. The initial
design was a very simple one - a "series of modules", as it was originally
described in a CommonJS sense, that would provide features of a classical
Object-Oriented system. It would seem ironic that, having a focus on
classical Object-Oriented development, one would avoid developing the project in
such a paradigm. Instead, I wished to keep the design simple (because the
project seemed simple), more natural to JS developers (prototypal) and
performant (object literals do not have the overhead of instantiation). Well,
unfortunately, the project scope has increased drastically due to the success of
the implementation (and my playfulness), the chosen paradigm has become awkward
in itself and the performance benefit is indeed a micro-optimization when
compared with the performance of both the rest of the system and the system that
will implement ease.js as a framework.
You can only put off refactoring for so long before the system begins to trip
over itself and stop being a pleasure to work with. In fact, it's a slap in the
face. You develop this intricate and beautiful system (speaking collectively and
generally, of course) and it begins to feel tainted. In order to prevent it from
developing into a ball of mud - a truly unmaintainable mess - the act of
refactoring is inevitable, especially if we want to ensure that the project
survives and is actively developed for any length of time.
In this case, the glaring problem is that each of the modules are terribly,
tightly coupled. This reduces the flexibility of the system and forces us to
resort to a system riddled with conditionals. This becomes increasingly apparent
when we need to provide slightly different implementations between environments
(e.g. ES5/pre-ES5, production/development, etc and every combination).
Therefore, we need to decouple the modules in order to take advantage of
composition in order to provide more flexible feature sets depending on
environment.
What does this mean?
We need to move from object literals for the modules to prototypes (class-like,
but remember that ease.js exists because JS does not have "classes"). A number
of other prototypes can be extracted from the existing modules and abstracted to
the point where they can be appropriately injected where necessary. Rather than
using conditions for features such as fallbacks, we can encapsulate the entire
system in a facade that contains the features relevant to that particular
environment. This will also have the consequence that we can once again test
individual units rather than systems.
At the point of this commit (this entry was written before any work was done),
the major hurdle is refactoring the test cases so that they do not depend on
fallback logic and instead simply test specific units and skip the test if the
unit (the prototype) is not supported by the environment (e.g. proxies in a
pre-ES5 environment). This will allow us to finish refactoring the fallback and
environment-specific logic. It will also allow us to cleanly specify a fallback
implementation (through composition) in an ES5 environment while keeping ES5
detection mechanisms separate.
The remaining refactorings will likely be progressive. This all stemmed out of
the desire to add the method hiding feature, whose implementation varies
depending on environment. I want to get back to developing that feature so I can
get the first release (v0.1.0) out. Refactoring can continue after that point.
This project needs a version number so it can be used reliably.
- This commit was originally many. Unfortunately, certain Git objects became
corrupt shortly after my 500th commit due to HDD issues. Due to the scope, I
was unable to recover the set of commits I needed (after an hour of trying
every method).
- Fortunately, vim's swap files came to the rescue. Had I been able to
properly shut down my PC, I would have been rather frustrated.
- The ctor must be instantiated for use in the prototype chain
- This was working in the past, but apparently no test existed for it and
refactoring broke it
- In doing so, abandoned the super identifier (sid) for a more elegant solution with class ids (cid's)
- This permits fast and easy private member swapping via getMethodInstance() in class.js
- This incurs a performance hit for accessing protected members, and even further for public, internally
- But speeds up access to private members, likely due to there being less members
Mike Gerwitz