'. '

Excel

From APIDesign

Revision as of 21:45, 27 October 2010 by JaroslavTulach (Talk | contribs)
(diff) ←Older revision | Current revision (diff) | Newer revision→ (diff)
Jump to: navigation, search

Excel is almost a synonymum for any spreadsheet program. An example where a product name became de-facto name for the whole technology. I have never learned how to use Excel (or rather OpenOffice Calc) effectively, but I know that non-programmers can do quite amazing things with it.

Cluelessness is always a goal for any technology vendor. Everyone wants to provide good technology which is easy to learn, use and easy to get productive with. Everyone seeks such holy grail. Excel got quite close. It managed to allow non-programmers to be productive with computers without learning what are the if or while or for statements.

DSL or Library?

Rich Unger mentioned Excel few times during our DSL presentation at JavaOne2010 (btw. you can get slides from here: Image:Domain-library-shootout.pdf). Rich said that his company had to design Apex in order to attract people that are familiar with Excel, but are intimidated with Java. Actually when we (and also our audience) tried to summarize when to use DSLs (Rich's side of the shootout) and when to write libraries (my advice), we agreed that it basically depends on whether you are targeting (Java) programmers or not. For people with Excel knowledge the Apex was said to be a better fit.

Our talk was supposed to be a shootout and I used strong trash talk while preparing with Rich, but during the talk we treated each other with mutual respect. I somehow started to believe that Rich's excel argument is true. Apex was the language for Excel audience and it deserved its right to be a DSL, not a library (as the TheAPIBook would advocate). We ended up our talk in complete reunion.

Unit Test Excel Formulas!

However later, when we were celebrating success of our presentation at The Chieftains irish pub we got to the topic of testing. Apex was still on my mind and I mentioned that obviusly, Apex is not good enough for writing unit tests. In contrary to my feeling Rich explained thta people do write unit tests in Apex! I was shocked. I could not have imagined someone using Excel to write unit tests. Was Apex not supposed to be the language for Excel people, Rich?

It turned out that in fact, if you want to use Apex and deploy your application, the system measures the percentage of unit tests written by you and it rejects any changes until you reach 75% of test coverage. That indeed implies that Apex is a language suitable for writing unit tests. But that also means that Apex is not a language for Excel people. Excel guys don't even know what a test is!

Yikes, pretty harsh Jarda! Is there no middle ground, for those who have reached the limit of being able to run a business solely in excel, to take a step towards codifying their business practices without learning a general purpose language? I never claimed that there was no learning curve whatsoever. --rich

Somehow I feel that learning to write tests is as hard as learning some subset of Java to code against good API framework. The attempt to be as easy for clueless excel users is tempting. But in the light of the necessity to write tests to deploy anything, it sounds more a marketing goal or a vision than a reality.

Good API First

During the whole talk about DSLs vs. Library design I tried to argue that the important thing is to create a library to wrap your technology, the DSL is (at most) an additional wrapper to support more cluelessness. Rich tried to argue, that the library is not needed at all, that a DSL is enough. But in the light of those 75% unit test coverage, this is just a marketing dream, not a reality. Rich's company made an incorrect (technically, not from a marketing perspective) decision in the past to create their own DSL instead of producing a library in Java to access the same functionality. They have to stick to that decision (and even send Rich to talk about that at conferences), but to support other bussiness needs they give up on original level of cluelessness. Their users have to write tests (for good reasons which I agree with). However the whole story leaves quite a negative smell of doublethink, don't you think?

In spite of all my trash talk, my goal was never to say that DSLs are bad. My goal was to show that Java has amusing power too (see the LiveDB demo to agree). However if I knew all I know now, I would not let it be a tie! All the Rich's excel like arguments worth not a penny! I should have smashed Rich out of the stage. But too late, the only thing I can do about that now is to blog about that.


I agree that you showed that internal DSLs can be a nice addition to a library design, to help support cluelessness. However, in the case of Java that argument is somewhat hampered, since internal DSLs are only supported through the annotation mechanism, which itself has a learning curve for the end user, as the syntax is limited to certain constructs that can live in the existing Java AST. It is still possible, for certain use cases and for certain audiences, that an external DSL can provide a cleaner syntax, a cleaner evolution path, and hence better adoption. Apex has been extraordinarily successful, for the target audience it was originally intended for. Attempts to grow that into the general developer market would be a tough sell, to be sure. But the choice of an external DSL captured an important demographic. --rich

Personal tools
buy