#2157 Regex bug for replace all '\<' with a word

[Zufu Liu]

Regex replace all \< with a word (add prefix for all word) not works as expected:
Open SciTE type the word "hello", after replace all with "With", "hello" become "WithhWitheWithlWithlWitho".

however add suffix for all word by replace \> works.

[Zufu Liu]

Another bug: ^\< can find words at line start, but \> or \>$ can't find words at line end.

[Zufu Liu]

The search bug was fixed, maybe in 5.1.1 with 8873 (80d3b0c2b86a) Extract word edge detection to prepare for fixing bug..

[Zufu Liu]

Sorry, not yet fixed. however, \w\> or \w\>$ does find the last letter.

[Neil Hodgson]

A problem here is that RESearch::Execute treats the start of search as a line start and thus word start. It should probably be passed the line start value as a parameter or be able to discover it.

I have changed the title and description from /< to \< as it didn't make sense otherwise. please advise if /< was meant.

[Zufu Liu]

how about change BOL and EOL to only match real line start and end like MatchFlags:

std::regex_constants::match_flag_type MatchFlags(const Document *doc, Sci::Position startPos, Sci::Position endPos) noexcept {
    std::regex_constants::match_flag_type flagsMatch = std::regex_constants::match_default;
    if (!doc->IsLineStartPosition(startPos))
        flagsMatch |= std::regex_constants::match_not_bol;
    if (!doc->IsLineEndPosition(endPos))
        flagsMatch |= std::regex_constants::match_not_eol;
    return flagsMatch;
}

[Zufu Liu]

Patch to sync match_not_bol and match_not_eol, this fixed original problem except finding \> or \>$ at line end (will needs another patch similar to RESearch-EOW-atLineEnd-1013.diff).

reverse search test cases for std::regex are omitted due to current MatchOnLines() treats empty match as failure.

[Neil Hodgson]

The code for line end and line start seem intertwined so should be in a combined patch to better understand the scope.

[Zufu Liu]

Just following after switch in RESearch::Execute():

if (ep == NOTFOUND) {
    /* similar to EOL, match EOW at line end */
    if (endp == lineEndPos && *ap == EOW) {
        if ((ap[1] == END || ((ap[1] == EOL && ap[2] == END))) && iswordc(ci.CharAt(lp - 1))) {
            lp = endp;
            ep = lp;
        } else {
            return 0;
        }
    } else {
        return 0;
    }
}

so it's independent of other changes except test cases.

[Zufu Liu]

The combined patch:

[Neil Hodgson]

I think there is a duplicated test for findingEOW on lines 600-603.

[Zufu Liu]

Removed the duplicated test.

[Zufu Liu]

first patch to fix finding \> or \>$ at line end.

fix \< will needs another patch to set line start/end position for BOL and EOL, see
https://github.com/zufuliu/notepad2/commit/1bfaf31d6e05b99a1dd441854cfb82c3cf6f601e

[Neil Hodgson]

Committed as [60a4cc] with extra tests for \>$.