Invalid JS produced when more than 26 variables are auto-generated

[hen-x]

Input:

x for x in y
x for x in y
x for x in y
x for x in y
x for x in y
x for x in y
x for x in y
x for x in y
x for x in y
x for x in y
x for x in y
x for x in y
x for x in y
x for x in y
x for x in y

Output:

var __a, __b, __c, __d, __e, __f, __g, __h, __i, __j, __k, __l, __m, __n, __o, __p, __q, __r, __s, __t, __u, __v, __w, __x, __y, __z, __{, __|, __}, __~, __�, __�, x;
__a = y;
for (__b = 0; __b < __a.length; __b++) {
  x = __a[__b];
  x;
}

//----SNIP----

__} = y;
for (__~ = 0; __~ < __}.length; __~++) {
  x = __}[__~];
  x;
}

It looks like the names of auto-generated JS variables (eg. __a) are not currently restricted to alphanumeric characters, or checked for legality. When more than 26 such variables are emitted in a single scope, after __z has already been used, the compiler seems to continue up the ascii chart, using variable names like __{, __|, and __}, which are not syntactically valid.

A better option would switching to longer names after __z, and emitting __aa, __ab, ... __zy, _zz.

[jashkenas]

Thanks for reporting this -- I was just using Prototype.js' succ() without thinking about it. On the latest master, bin/node_coffee will now generate shorter numeric ids instead of bothering with alphabetic ones:

_1, _2, _3 ...

Closing the ticket.