stri_locate_all_fixed(), stri_locate_first_fixed(), stri_locate_last_fixed()

[gagolews]

Find all/first/last position of a occurence of substr in str (vectorized over str and substr)

[bartektartanus]

to determine which one do we expect
stri_locate_last_fixed("111","11") = c(1,2) or stri_locate_last_fixed("111","11") = c(2,3) ?

In other words, do first and last return first and last row of matrix returned by all?

[gagolews]

Very good question.
I think that an option should be added to look for overlapping patterns.

Maybe we should use the ICU API for all stri_____fixed - http://www.icu-project.org/apiref/icu4c/usearch_8h.html#details

There are many useful options, also locale-dependent

The algorithm you use has the worst-time complexity of O(nk) (n - str len, k - patt len), modified Boyer-Moore (http://icu-project.org/docs/papers/efficient_text_searching_in_java.html) has the same complexity, but may yet be faster in practice.

What do you think? Well, we should probably rely on ICU wherever it's possible

[bartektartanus]

> one <- stri_flatten(c(stri_dup(1,10000),2))
> pat <- stri_flatten(c(stri_dup(1,1000),2))
> print(microbenchmark(stri_locate_all_fixed(one,pat),str_locate_all(one,fixed(pat))))
Unit: milliseconds
                             expr       min        lq    median        uq       max
1 stri_locate_all_fixed(one, pat) 18.151333 18.277081 18.347167 18.616544 19.761872
2 str_locate_all(one, fixed(pat))  1.545308  1.583163  1.651747  1.730458  1.876635

Indeed, sometimes it might be much more slower.

[gagolews]

Something's wrong:

> stri_locate_first_fixed(c('AaaaaaaA', 'aaa', 'AAA'), 'a')
[[1]]
     start end
[1,]    NA  NA

[[2]]
     start end
[1,]    NA  NA

[[3]]
     start end
[1,]    NA  NA

Moreover, I think that stri_locate_first_fixed() and stri_locate_last_fixed() should return ONE matrix, not a list of matrices...

[bartektartanus]

done in commit 524f4e4

[gagolews]

TO DO: use collator_opts

[bartektartanus]

all functions DONE :) Closing.

[gagolews]

Nope, sorry, stri__locate_*_fixed_byte should use StriContainerUTF8, just like *_detect_byte does... 🎱

[bartektartanus]

DONE