I vote for carpentry. Because,

• You have to work within the constraints of the materials.

  The carpenter has to assess and work with the variations of the grain, strength, and imperfections of the wood.

  And he has to chose the tools and techniques he uses for any particular job to suit the requirements of the job. Mortice and tenon joints and glue may be perfect for building a cabinet, but would take far too long when erecting shuttering for a concrete pour, where nails are more cost effective.

  The programmer has to work with and around data that may be incomplete, badly formatted, or just damned inconvenient.

  Likewise, he has to choose his language and tools carefully, and work with the strengths and around the weaknesses and imperfections of them.

• It takes skills knowledge and experience to make the raw materials approximate the design.

  No matter how well designed and drawn the plans for a workpiece are done, translating the raw materials into the product is far more than just 'cut here', 'drill there', 'screw that'. How you cut, drill and screw means that two equally skilled practitioners of the art, or even the same guy on different days, can produce good and bad examples of the same thing.

  The same analogy holds for programmers. How you choose to perform each step is nearly as influencial to the outcome as performing all the steps and doing them in the right order.

• Carpenters, like programmers, are not machines.

  Where a sheet metal press, or autolathe or grinder can be set up to turn out consistantly identical parts, each workpiece or program is created anew.

  Even the best carpenter can have off-days. Cut in the wrong place or too deep; pick the wrong piece of stock; or misread the measurements.

• Mass production methods can go some way to speeding up production and alleviate variance, but the final product is never quite as satisfactory.

  Using engineered woods (plywood, MDF, veneers etc.), and automated machinery can allow for the production of cheaper products.

  But just as a set of mass produced kitchen cabinets will never quite fit any given kitchen as well as the bespoke output of master carpenter. So the results of a 4GL; the mix'n'match of generic libraries; or the filling-in-the-blanks of a framework; program will never be as fit for purpose as one written by a skilled programmer to satisfy a targetted specification.

Basically, just as with carpentry, there is just as much art and craft in software development as there is engineering. Perhaps more so. That will have to change unfortunately.