Rope code appears to do a string overrun

[graydon]

In commit 956bc77 I had to disable one of the rope tests. The purpose of that commit was to fix up a bug in the way operator [] works on strings. In particular it used to let you index up to and including the trailing null; now it considers access to the null byte an index overrun and fails.

I tried to figure out exactly what was going on in the rope code for quite a while, and tried adjusting a number of obvious-looking things, but to no avail. I'm not really good with boundary-case arithmetic (are any of us?) and the rope module is a maze of such arithmetic.

Sharper minds requested to find the bug. Meanwhile I marked the rope test in question (char_at1) as ignored. The others seem to work.

[catamorphism]

Paging @Yoric

[Yoric]

I'll take a look.

[msullivan]

There is some discussion on IRC that maybe we should just pitch the rope code as it exists now.

[Aatch]

@msullivan agreed. The code is pretty outdated and the work required to modernize it would probably be similar to the work needed to just re-write it.

When I was looking through it, it seemed to be doing some very strange things that I don't think are even close to being safe, so there's that too.