2011-11-07

Adventures with Newsqueak

I pushed Rob Pike's Newsqueak interpreter to my github page.

This might seem (and largely is) a senseless thing to do, as it is freely available on Rob's (old) homepage. And unfortunately it looks a bit like me trying to take credit for his work, which I really do not intend to do - so let me explain, why I did this.

Last week or so, I tried to build squint (that's the official name of this marvelous piece of software) on a machine running Ubuntu 10.10, which gave strange warnings about a gcc built-in being called with a wrong type.

This seemed odd, given that the version I tried to build was explicitly marked as the "Linux" version (which implies that using gcc is okay) and not too old, either (2000, okay that's eleven years, but still).

When I tried to run some test program through the interpreter, it died with a segmentation fault. Now, this was really odd. I hooked up gdb and it seemed that it died on a strlen() call somewhere deep in the lexer.

Now, it is pretty easy to write a sloppy C program that dies on a strlen() - just handle it a char array without a terminating NUL byte and off you go. But lexers are usually super-simple and, last but not least, it was friggin' Rob Pike who wrote it.

Clearly, there was something very wrong. And probably not the code itself.

So I though to myself "okay, let's debug this sometime, and upload the modern version to github" - thinking that gcc probably changed some default behavior over the years and I just needed to add something to CFLAGS to make up for that.

Fast forward a week including an update to the latest Ubuntu, a disgusted look, and a downgrade to 10.04 LTS (i.e. "long term support").

I had some time on my hands and tried to tackle this squint thing again.

I typed make all - and, lo and behold, no warning, but one error. The error was about some pretty strange #include for sys/wait.h but with an absolute path instead of, well, sys/wait.h.

So I removed the statement - and everything, including running the test programs, just worked. I thought "cool, fixed it" and uploaded it to github. Then I realized that the value of that one-line "fix" is, well, not too high, actually...

So now it looks like I took some software from one place, did absolutely nothing on it, and uploaded it to another place, but with my name added. Not that anybody cares, but I just wanted to explain the whole story so I can stop feeling bad about it.

What also helps is that Newsqueak is a pretty neat (and mostly forgotten) language. It has, among other nice things, basically typed named-pipe elements at the language level. And there's a language-level select for these things, too. So that's pretty cool and powerful, while still easy to reason about. But the real language overview will have to wait until the next post - I have to really learn it, first. :)

Happy Hacking!

(Update: Well, it was (and is) a problem with the code. Some of the plan9 libraries included with squint don't build for 64 bit targets (or well, they do build, but they die at runtime). I thought it was dependent on the gcc version, because I switched that and forgot that I switched to a 32 bit CPU at the same time. Word-size depend bugs never occured to me before, so this wasn't exactly my first though on what was going wrong.)



comments powered by Disqus