[Support] Faster line and column lookup in SourceMgr (#195881)
Previously line and column lookup worked by finding the line number by
utilizing a cache of offsets of line ends. And finding the column number
by linear search of a previous newline. This is quite slow and caused
some issues after merging #174566 and was fixed by wrapping all calls of
the slow function in a `if (ParserContext)` in #180068. This is not
ideal, since when the parser context would be supplied the parsing of
files with long lines will take ages.
This PR implements a fix to the problem by utilizing the computed
offsets for line ends to calculate the position on the current line.
Thus improving the asymptotic complexity of the getLineAndColumn method
from O(log l + c) to O(log l) (l is number of lines, c is column
position of the pointer). While not adding any overhead.
The second commit in this PR just updates two comments that I thought
were misleading.
Since the getLineNumber methods are now unused I removed them in the
third commit. And that caused MLIR to not compile, so I fixed it, while
removing TODOs, that are implemented by this PR.
llvm-project
72af17eb - [Support] Faster line and column lookup in SourceMgr (#195881)
