Fixed prettyIndex for one-line files, use binarySearch

[hauserx]

Pretty index was failing with following exception for files with one line and no \n at the end. The lines array was empty:

java.lang.ArrayIndexOutOfBoundsException: Index -1 out of bounds for length 0

There are actually some generated files in our code based that follow this pattern.

Also, for expressions in the very first line it was creating values with negative columns, like:

x.libsonnet:2:-63

Fixed the issue and added tests, also for edge case with completely empty file.

While changing it also optimized searching for line - previously indexWhere was used which is linear search. Replaced with binary search. It does not matter to much usually, but actually was causing slowdown of profiler code that gets locations of all evaluated expressions - this will be added in a subsequent PR.

[He-Pin]

@hauserx Nice, seems you need run scalafmt in the shell.

[hauserx]

Just noticed that the current pretty index implementation is same as in fastparse: https://github.com/com-lihaoyi/fastparse/blob/385431ec271b411505a9458bf8f85d489a9b001b/fastparse/src/fastparse/ParserInput.scala#L123

Also, sjsonnet Utils already has binary search:

sjsonnet/sjsonnet/src/sjsonnet/Util.scala

Line 5 in abbcf33

def prettyIndex(lineStarts: Array[Int], index: Int): String = {

Let me change this code a little.

[hauserx]

@He-Pin @stephenamar-db PTAL

[He-Pin]

Why change it to 0 , it will always later be resiezd then