Instead, I tweaked the rest of the matching function, so that a special
check isn't necessary. If we are trying to match at the end of a string,
we skip any of the actual matching and proceed straight to finding
0-length matches.
This change was made because, with the special case, capturing groups
weren't getting updated if we had an end-of-string match.
I made findAllMatchesHelper a non-recursive function. It now only
returns the first match it finds in the string (so I should probably
rename it).
These indices are collected by findAllMatches and pruned (to
remove overlaps). The overlap function has also been rewritten, to make
it (I believe) less than O(n^2). I also used the uniq_arr type to make
checking for uniqueness O(1) instaed of O(n) (as it was with
unique_append()). This has resulted in massive performance gains.
There's been a lot of changes here, and I probably haven't documented
all of them.
I added support for transitions. I wrote a function to determine if
a given state has transitions for a character at a given point in the
string. This helps me check if the current state has an assertion, and
take actions based on that.
I also fixed zero-length matching (almost, see todo.txt). It works for
nearly all cases I could think of, although I still need to write more
tests. I wrote a function to check if zero-length matches are possible
with a given state.
I also changed the way recursive calls work. Rather than passing a
modified string, the function stores the location in the input string.
This location is updated with each call to the function.
Finally, the function now increments the offset by 1 instead of
incrementing by the length of the longest match. This leads to a bit of
overhead eg. if a regex matches index 1-5, then 1-5, 2-5, 3-5, 4-5 are
all stored. To fix this, I wrote (and used) a function to check if
a match overlaps with any matches in a slice.