KMP calculates a table of how many characters it can skip if some of the search string matches but not all of it. Its more useful for relatively large search strings since it can skip potentially a lot of compares. Since this use is always going to search the whole file, to mark all occurrances rather than stop at the first/next match, its probably worthwhile for any reasonable size search string despite the preparation needed.
But I would really like to see good benchmarks, since with modern CPUs and caches I would also suspect many of the comparisons on the naive algorithm will be hidden by the memory access time and its simple order of access might make better use of prefetches.
And of course there are the SIMD instructions in modern processors just waiting to be exploited by the adventurous :)
Cheers Lex
On 29 May 2015 at 20:49, Thomas Martitz kugel@rockbox.org wrote:
Am 29.05.2015 um 12:44 schrieb marius buzea:
Hello,
With KMP it is possible to search all occurrences of a m length string, into a n length string, using O(m+n) machine operations. Next page: http://www.inf.fh-flensburg.de/lang/algorithmen/pattern/kmpen.htm describes the algorithm.
The KMP works well with the utf-8 encoding of unicode. One property of utf8 is that the encoding one unicode symbol is not a substring of another utf8 substring. This property allows to take the utf-8 encoding of the string you wish to search, and to find this utf8 encoding string, in the utf8 encoding of the text string. Geany uses scintilla, and scintilla uses utf8 to encode the document it displays, and scintilla has a command that gives the raw utf8 byte array for a [start, end) range. So, KMP gives great speed for searching all occurrences, and may be used with the underlying text representation of scintilla used by geany. The utf-8 encoding of a unicode string of length n, is less than 6n, each utf8 encoding is at most 6 bytes.
I also think that including this functionality/feature into Geany core would be a good choice. It would be a small tradeoff between keeping the core small, and adding this new functionality, but this is your choice.
If you wish to extend automark, then this is good choice too. If you wish, and if it helps, please reuse any part of the implementation provided here:
http://sourceforge.net/p/geanyhighlightselectedword/code/HEAD/tree/trunk/Gea... If needed, I would help.
What should I do next? Should I not do the pull request for GeanyHighlightSelectedWord? It is okay with me. GeanyHighlightSelectedWord would then be still available at sourceforge until Geany provides this functionality from its core, or from automark.
I wonder if this algorithm should be applied to all searches, and thus be integrated into scintilla. Does it have any major drawbacks? I read it has to some kind "prefix table" prior to running the search, but I guess that's negligible for all reasonable search terms?
Best regards _______________________________________________ Devel mailing list Devel@lists.geany.org https://lists.geany.org/cgi-bin/mailman/listinfo/devel