-1 on any changes to rfind(). Maybe someone can consider it later, let's leave it off of this PR.

Also, how should I read your tables? I'm confused by-131.2%, since that implies negative time to me.

-1 on any changes to rfind(). Maybe someone can consider it later, let's leave it off of this PR.

It looks a big mess now, but I am half-way through making functions that handle both directions with the same code. Initially it slowed down code and looked complex, but it is slowly recovering in both aspects. I should have a presentable version in a day or so. You can take a look then and if you're not happy I can always revert to the first commit of this PR.

Also, how should I read your tables? I'm confused by-131.2%, since that implies negative time to me.

The formula is as follows:

current_time=`time of current python version`new_time=`time of this PR`result= (new_time-current_time)/min(new_time,current_time)

I did minimum to make it fair. E.g.:

new_time=2current_time=1(new_time-current_time)/current_time=100%(new_time-current_time)/min(new_time,current_time)=100%new_time=1current_time=2(new_time-current_time)/current_time=-50%(new_time-current_time)/min(new_time,current_time)=-100%

This way it is symmetric to both positive and negative.

So-131.2% means old version is131.2% slower than the new.

Could you keep the variable names clear? I don't like 1 letter abbreviations. (i &j for loop counters are generally fine though.)

Changing variable names is easy. However, there were several functions defining same variables with different names:
a)default_find and friends was usings/n/p/m, which is a default naming for string search problems. Seehttps://en.wikipedia.org/wiki/Boyer–Moore_string-search_algorithm. Generally, 90% of search algorithms use this notation.
b)two_way_find was using full names ofhaystack &needle.

I took liberty to synchronize them and picked a) for the following reasons:
a) To me, it is easier to ingest the code visually when variable names are short. High level code, where there are no complex loops, probably long names are better. But when code gets complex, such as algorithms, then bottleneck becomes the logic, not readability, because one needs to look a certain amount of time before starting to grasp something. 2% into the work variable names are learnt by heart (whatever they might be) and visual inspection of the code is priority for the rest 98% of the time. I think there is a reason why libraries of algorithms use short names.
b) haystack and needle are definitions that are created for analogy (and probably fun). I prefer more formal names for variables or the ones that match equations. Fun variable names can be mentioned in documentation.

However, this is my take. If there are strong objections, it is no problem to revert them back.

Nevertheless, I think they should be synchronized across all the functions in the module. Naming discrepancy has cost me a fair bit of time due to having to readjust myself every time I changed focus.

UPDATE:
AlsoObjects/stringlib/find.h usesstr/str_len/sub/sub_len and it contains higher level functions. Generally, variable names get shorter as it gets to lower level, so I would at least like to keep the same length. But given complexity of the code would still try to insist on 1-letter names that match most of resources on such problems (especially code).

Also, please document your code (especially the structures) and explain what is what when you use short variables. Even though it is standard in such algorithms, it does not hurt to remind ourselves thatp is the pattern,m is its length, etc. But there are other variables with short names that have no explanation.

I eventually changed back argument names toneedle andhaystack. Also did the same withpre-work structure. It is easier to figure things out this way, when there is no need to work with low details of algorithm.

However, inside algorithm logics/n/p/m are used and it is documented. But happy to change them to full names if needed.

Why was the force push necessary?

It wasn't necessary, but preferable, given the situation I was in.

cc@tim-one

@sweeneyde, are you still around? You're obviously the best person to look at this, and it appears to be substantial work with real value. I can't make time to climb this learning curve.