While that tendency and experience are of course
valuable, it is still worth ?¬?ghting them down to a reasonable extent. If the modeling
language can be made 10% better to use, at the expense of making the creation of the
code generator 10% harder, many modelers will bene?¬?t over many applications, while
only one metamodeler needs to suffer once.
Still, as metamodelers we are all for reducing the suffering of fellowmetamodelers
where possible! One important way to do that when building generators is to have
something concrete to work with: a working mini-application in exactly the format
you want to generate. That is what we will look at in the next section, before moving
on to the generators themselves.
11.1 ???HERE??™S ONE I MADE EARLIER???
Before we can build a generator, we need to know what we want to generate. The best
way to do that is to have a working example of the output. Remember, DSM is about
automating what wewould otherwise be doing by hand. Unless we knowwhat wewant
and are able to do that by hand, there is little hope of teaching somebody else to do it??”
let alone teaching something as stupid as a computer to do it. The good news is that we
only need one example of each part of the output, and the computer learns quickly.
But wait! It is not enough to have just the output: since a generator is a
kind of transformation, we must also have the corresponding input.
Pages:
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493