Slashdot Mirror
Why New Programming Languages Succeed Or Fail
snydeq writes "Fatal Exception's Neil McAllister discusses the proliferation of programming languages and what separates the successful ones from obscurity. 'Some people say we don't need any more programming languages at all. I disagree. But it seems clear that the mainstream won't accept just any language. To be successful, a new language has to be both familiar and innovative — and it shouldn't try to bite off more than it can chew. ... At least part of the formula for success seems to be pure luck, like a band getting its big break. But it also seems much easier for a language to shoot itself in the foot than to skyrocket to stardom.'"
Ada did not fail at all. It is used for exactly what it was designed for: mission critical defense applications.
Ada was not designed for intranet or web or mobile or desktop applications, although it can do those things really well.
Java succeeded because Sun 1) gave it away, and 2) threw money at giving it away. Remember "applets"? Java was supposed to be the programming language of the Web. That didn't work out. It ended up being the new COBOL, which was not Sun's intent.
Some languages fail, or get stuck, because the designer is in love with their own implementation. That happened to Pascal and Python. Wirth's own Pascal implementation was a cute little recursive-descent compiler that generated RPN byte codes, like a Java compiler. Wirth resisted changes to the language that would allow programming in the large. ISO Pascal reflects his biases. So Pascal became stuck in an educational niche. The original Macintosh software was all written in an extended Pascal, as was much '80s software. But everybody had a different dialect - there was Turbo Pascal, Clascal, and a few others. They never merged.
Modula, Wirth's second try, was also crippled in certain ways. Modula 2 was better. Modula 3 was good enough to be used to write an operating system kernel. Unfortunately, Modula 3 was only used with DEC, which died after being acquired by Compaq.
Python has some of the same problems. The feature set of Python reflects what it's easy to implement in a naive interpreter, like von Rossum's CPython. Internally, everything is an object, even integers and floats, and object access involves dictionary lookups. This makes CPython slow. Every attempt to speed up Python substantially has hit a wall, including Google's "Unladen Swallow" effort. (PyPy is making progress, but it's taken a decade and requires an incredibly complex internal combination of interpreters and compilers.)
The biggest disappointment to me has been that we're still stuck with C. C has two killer bad design decisions - the language doesn't know how big arrays are, and the "pointer=array" thing lies to the language. Both reflect how things are done in assembler, and the fact that the original compiler had to fit in a 128K PDP-11. Most of the millions of buffer overflows and crashes that occur daily can be traced to those two design decisions. (C++, as I point out occasional, tries to paper over these problems with collection classes. But the mold usually seeps through the wallpaper, since most operating system and library calls want raw C pointers.)