import regex import rand /****************************************************************************** * * Test section * ******************************************************************************/ struct TestItem { src string q string s int e int } const( match_test_suite = [ // minus in CC TestItem{"d.def",r"abc.\.[\w\-]{,100}",-1,0}, TestItem{"abc12345.asd",r"abc.\.[\w\-]{,100}",-1,0}, TestItem{"abca.exe",r"abc.\.[\w\-]{,100}",0,8}, TestItem{"abc2.exe-test_12",r"abc.\.[\w\-]{,100}",0,16}, TestItem{"abcdefGHK",r"[a-f]+\A+",0,9}, TestItem{"ab-cd-efGHK",r"[a-f\-g]+\A+",0,11}, // base OR TestItem{"a",r"a|b",0,1}, TestItem{"a",r"b|a",0,1}, TestItem{"b",r"a|b",0,1}, TestItem{"b",r"b|a",0,1}, TestItem{"c",r"b|a",-1,0}, // test base TestItem{"[ciao]",r"(.)ciao(.)",0,6}, TestItem{"[ciao] da me",r"(.)ciao(.)",0,6}, // positive TestItem{"this is a good.",r"this",0,4}, TestItem{"this is a good.",r"good",10,14}, TestItem{"this is a good.",r"go+d",10,14}, TestItem{"this is a good.",r"g[oae]+d",10,14}, TestItem{"this is a goed.",r"g[oae]+d",10,14}, TestItem{"this is a good.",r"g[oae]*d",10,14}, TestItem{"this is a goaezd.",r"g[ea-cm-z]*d",10,16}, TestItem{"this is a good.",r"this (\w+) a",0,9}, TestItem{"this is a good.",r"this( \w+){2} g",0,11}, TestItem{"this is a good.",r"( ?\w+){,1}",0,4}, TestItem{"this is a good.",r"( ?\w+)+",0,14}, TestItem{"this is a good.",r"this( \w+)+",0,14}, TestItem{"this is a good sample.",r"( ?\w+){,2}",0,7}, TestItem{"this is a good sample.",r"( ?\w+){,3}",0,9}, TestItem{"this is a good sample.",r"( ?\w+){,4}",0,14}, TestItem{"this is a good sample.",r"( ?\w+){,5}",0,21}, TestItem{"this is a good sample.",r"( ?\w+){2,3}",0,9}, TestItem{"this is a good sample.",r"(\s?\w+){2,3}",0,9}, TestItem{"this these those.",r"(th[ei]se?\s|\.)+",0,11}, TestItem{"this these those ",r"(th[eio]se? ?)+",0,17}, TestItem{"this these those ",r"(th[eio]se? )+",0,17}, TestItem{"this,these,those. over",r"(th[eio]se?[,. ])+",0,17}, TestItem{"soday,this,these,those. over",r".+(th[eio]se?[,. ])+",0,23}, TestItem{"cpapaz",r"(c(pa)+z)",0,6}, TestItem{"this is a cpapaz over",r"(c(pa)+z)",10,16}, TestItem{"this is a cpapapez over",r"(c(p[ae])+z)",10,18}, TestItem{"test@post.pip.com",r"[a-z0-9_]+@([a-z0-9_]+\.?)+",0,17}, TestItem{"test1@post.pip.com, pera",r"[\w]+@([\w]+\.)+\w+",0,18}, TestItem{"pippo@pera.com ",r"[a-z0-9_]+@([a-z0-9_]+\.?)+",0,14}, TestItem{"adce aabe",r"(a(ab)+)|(a(dc)+)e",0,4}, TestItem{"zadce aabe",r"(a(ab)+)|(a(dc)+)e",1,5}, TestItem{"abbz accz addz.",r"c|(d)|e|(ab+)",0,3}, TestItem{"this those these ciao",r"((t[hieo]+se?)\s*)+",0,17}, TestItem{"this ciao",r"((t[hieo]+se?)\s*)+",0,5}, TestItem{"this cpapaz adce aabe",r"(c(pa)+z)(\s[\a]+){2}",5,21}, TestItem{"1234this cpapaz adce aabe",r"(c(pa)+z)(\s[\a]+){2}$",9,25}, TestItem{"this cpapaz adce aabe third",r"(c(pa)+z)(\s[\a]+){2}",5,21}, TestItem{"123cpapaz ole. pippo",r"(c(pa)+z)(\s+\a+[\.,]?)+",3,20}, TestItem{"this is a good sample.",r".*i(\w)+",0,4}, TestItem{"soday,this,these,those. over",r".*,(th[eio]se?[,. ])+",0,23}, TestItem{"soday,this,these,thesa.thesi over",r".*,(th[ei]se?[,. ])+(thes[ai][,. ])+",0,29}, TestItem{"cpapaz ole. pippo,",r".*(c(pa)+z)(\s+\a+[\.,]?)+",0,18}, TestItem{"cpapaz ole. pippo",r"(c(pa)+z)(\s+\a+[\.,]?)+",0,17}, TestItem{"cpapaz ole. pippo, 852",r".*(c(pa)+z)(\s+\a+[\.,]?)+",0,18}, TestItem{"123cpapaz ole. pippo",r".*(c(pa)+z)(\s+\a+[\.,]?)+",0,20}, TestItem{"...cpapaz ole. pippo",r".*(c(pa)+z)(\s+\a+[\.,]?)+",0,20}, TestItem{"cpapaz ole. pippo,",r".*c.+ole.*pi",0,14}, TestItem{"cpapaz ole. pipipo,",r".*c.+ole.*p([ip])+o",0,18}, TestItem{"cpapaz ole. pipipo",r"^.*c.+ol?e.*p([ip])+o$",0,18}, TestItem{"abbb",r"ab{2,3}?",0,3}, TestItem{" pippo pera",r"\s(.*)pe(.*)",0,11}, TestItem{" abb",r"\s(.*)",0,4}, TestItem{"/home/us_er/pippo/info-01.txt", r"(/?[-\w_]+)*\.txt$",0,29} // negative TestItem{"zthis ciao",r"((t[hieo]+se?)\s*)+",-1,0}, TestItem{"this is a good.",r"thes",-1,0}, TestItem{"test1post.pip.com, pera",r"[\w]+@([\w]+\.)+\w+",-1,0}, TestItem{"this cpapaz adce",r"(c(pa)+z)(\s[\a]+){2}",-1,0}, TestItem{"this cpapaz adce aabe third",r"(c(pa)+z)(\s[\a]+){2}$",-1,0}, TestItem{"1234this cpapaz adce aabe ter",r"(c(pa)+z)(\s[\a]+){2}$",-1,0}, TestItem{"cpapaz ole. pipipo,",r"^.*c.+ol?e.*p([ip])+o$",-1,0}, TestItem{"/home/us_er/pippo/info-01.jpeg", r"(/?[-\w_]+)*\.txt$",-1,0} // check unicode TestItem{"this is a Ⅰ Ⅱ Ⅲ Ⅳ Ⅴ Ⅵ test",r".*a [Ⅰ-Ⅵ ]+",0,34}, TestItem{"123Ⅰ Ⅱ Ⅲ Ⅳ Ⅴ Ⅵ test",r"[Ⅰ-Ⅴ\s]+",3,23}, // new edge cases TestItem{"12345678", r"[0-9][0-9][0-9][0-9][0-9][0-9][0-9][0-9][0-9]",-1,0}, TestItem{"12345678", r"[0-9][0-9][0-9][0-9][0-9][0-9][0-9][0-9]",0,8}, TestItem{"123456789", r"^[0-9][0-9][0-9][0-9][0-9][0-9][0-9][0-9][0-9]$",0,9} TestItem{"12345678", r"^\d{8}$",0,8}, TestItem{"12345678", r"^\d{7}$",-1,0}, TestItem{"12345678", r"^\d{9}$",-1,0}, TestItem{"eth", r"(oth)|(eth)",0,3}, TestItem{"et", r"(oth)|(eth)",-1,0}, TestItem{"et", r".*(oth)|(eth)",-1,0}, TestItem{"peoth", r".*(ith)|(eth)",-1,0}, TestItem{"poth", r"(eth)|(oth)",1,4}, TestItem{"poth", r"(oth)|(eth)",1,4}, TestItem{"poth", r".(oth)|(eth)$",0,4}, TestItem{"poth", r"^.(oth)|(eth)$",0,4}, TestItem{"poth", r"^\w+$",0,4}, // test dot_char TestItem{"8-11 l: qllllqllklhlvtl", r"^(\d+)-(\d+) ([a-z]): (.*)$",0,23}, TestItem{"accccb deer", r"^a(.*)b d(.+)r",0,11}, TestItem{"accccb deer", r"^a(.*)b d(.+)",0,11}, TestItem{"accccb deer", r"^(.*)$",0,11}, TestItem{"accccb deer", r"^a(.*)b d(.+)p",-1,0}, TestItem{"##.#....#.##.####...#.##", r".{18}[.#]",0,19}, TestItem{"#.#......##.#..#..##........##....###...##...######.......#.....#..#......#...#........###.#..#.", r'.*#[.#]{4}##[.#]{4}##[.#]{4}###',0,49}, // test bcksls chars TestItem{"[ an s. s! ]( wi4ki:something )", r"\[.*\]\( *(\w*:*\w+) *\)",0,31}, TestItem{"[ an s. s! ](wiki:something)", r"\[.*\]\( *(\w*:*\w+) *\)",0,28}, TestItem{"p_p", r"\w+",0,3}, TestItem{"p_é", r"\w+",0,2}, // Crazywulf tests (?:^|[()])(\d+)(*)(\d+)(?:$|[()]) TestItem{"1*1", r"(\d+)([*])(\d+)",0,3}, TestItem{"+1*1", r"^(\d+)([*])(\d+)",-1,0}, TestItem{"*1*1", r"(?:^|[*])(\d+)([*])(\d+)",0,4}, TestItem{"*1*1", r"(?:^|[*()])(\d+)([*])(\d+)",0,4}, TestItem{")1*1", r"(?:^|[*()])(\d+)([*])(\d+)",0,4}, TestItem{"(1*1", r"(?:^|[*()])(\d+)([*])(\d+)",0,4}, TestItem{"*1*1(", r"(?:^|[*()])(\d+)([*])(\d+)(?:$|[*()])",0,5}, TestItem{" 1*1(", r"(?:^|[*()])(\d+)([*])(\d+)(?:$|[*()])",-1,0}, TestItem{"1*1 ", r"(?:^|[*()])(\d+)([*])(\d+)(?:$|[*()])",-1,0}, // particular groups TestItem{"ababababac", r"ab(.*)(ac)",0,10}, // backslash on finish string TestItem{"a", r"\S+",0,1}, TestItem{"aaaa", r"\S+",0,4}, TestItem{"aaaa ", r"\S+",0,4}, ] ) struct TestItemRe { src string q string rep string r string } const ( match_test_suite_replace = [ // replace tests TestItemRe{ "oggi pibao è andato a casa di pbababao ed ha trovato pibabababao", r"(pi?(ba)+o)", "CIAO", "oggi CIAO è andato a casa di CIAO ed ha trovato CIAO" }, TestItemRe{ "Today is a good day and tomorrow will be for sure.", r"[Tt]o\w+", "CIAO", "CIAO is a good day and CIAO will be for sure." }, TestItemRe{ "Today is a good day and tomorrow will be for sure.", r"(a\w) ", r"[\0] ", "Tod[ay] is a good d[ay] and tomorrow will be for sure." }, TestItemRe{ "Today is a good day and tomorrow will be for sure.", r"(a\w) ", r"[\0_\0] ", "Tod[ay_ay] is a good d[ay_ay] and tomorrow will be for sure." }, TestItemRe{ "Today is a good day and tomorrow will be for sure.", r"(a\w) ", r"[\0\1] ", "Tod[ay] is a good d[ay] and tomorrow will be for sure." }, ] match_test_suite_replace_simple = [ // replace tests TestItemRe{ "oggi pibao è andato a casa di pbababao ed ha trovato pibabababao", r"(pi?(ba)+o)", "CIAO", "oggi CIAO è andato a casa di CIAO ed ha trovato CIAO" }, TestItemRe{ "Today is a good day and tomorrow will be for sure.", r"[Tt]o\w+", "CIAO", "CIAO is a good day and CIAO will be for sure." }, ] ) struct TestItemCGroup { src string q string s int e int cg []int // [number of items (3*# item), id_group_0, start_0, end_0, id_group_1, start1, start2,... ] cgn map[string]int } const ( cgroups_test_suite = [ TestItemCGroup{ "http://www.ciao.mondo/hello/pippo12_/pera.html", r"(?Phttps?)|(?:ftps?)://(?P[\w_]+[\.|/])+",0,42, [7, 0, 0, 4, 1, 7, 11, 1, 11, 16, 1, 16, 22, 1, 22, 28, 1, 28, 37, 1, 37, 42], {'format':int(0),'token':1} }, TestItemCGroup{ "http://www.ciao.mondo/hello/pippo12_/pera.html", r"(?Phttps?)|(?Pftps?)://(?P[\w_]+.)+",0,46, [8, 0, 0, 4, 1, 7, 11, 1, 11, 16, 1, 16, 22, 1, 22, 28, 1, 28, 37, 1, 37, 42, 1, 42, 46] //[8, 0, 0, 4, 1, 7, 10, 1, 11, 15, 1, 16, 21, 1, 22, 27, 1, 28, 36, 1, 37, 41, 1, 42, 46], {'format':int(0),'token':1} }, TestItemCGroup{ "http://www.ciao.mondo/hello/pippo12_/pera.html", r"(?Phttps?)|(?Pftps?)://([\w_]+\.)+",0,16, [3, 0, 0, 4, 1, 7, 11, 1, 11, 16], {'format':int(0)} }, TestItemCGroup{ "acc +13 pippo", r"(\w+)\s(.)([0-9]+) \w+",0,13, [0, 3, 4, 5, 5, 7], map[string]int{} }, TestItemCGroup{ "acc +13", r"(\w+)\s(.)([0-9]+)",0,7, [0, 3, 4, 5, 5, 7], map[string]int{} }, TestItemCGroup{ "ababababac", r"ab(.*)(ac)",0,10, [2, 8, 8, 10], map[string]int{} }, ] ) struct Test_find_all { src string q string res []int // [0,4,5,6...] res_str []string // ['find0','find1'...] } const ( find_all_test_suite = [ Test_find_all{ "abcd 1234 efgh 1234 ghkl1234 ab34546df", r"\d+", [5, 9, 15, 19, 24, 28, 31, 36], ['1234', '1234', '1234', '34546'] }, Test_find_all{ "abcd 1234 efgh 1234 ghkl1234 ab34546df", r"\a+", [0, 4, 10, 14, 20, 24, 29, 31, 36, 38], ['abcd', 'efgh', 'ghkl', 'ab', 'df'] }, Test_find_all{ "oggi pippo è andato a casa di pluto ed ha trovato pippo", r"p[iplut]+o", [5, 10, 31, 36, 51, 56], ['pippo', 'pluto', 'pippo'] }, Test_find_all{ "oggi pibao è andato a casa di pbababao ed ha trovato pibabababao", r"(pi?(ba)+o)", [5, 10, 31, 39, 54, 65], ['pibao', 'pbababao', 'pibabababao'] }, Test_find_all{ "Today is a good day and tomorrow will be for sure.", r"[Tt]o\w+", [0, 5, 24, 32], ['Today', 'tomorrow'] }, Test_find_all{ "pera\nurl = https://github.com/dario/pig.html\npippo", r"url *= *https?://[\w./]+", [5, 44], ['url = https://github.com/dario/pig.html'] }, Test_find_all{ "pera\nurl = https://github.com/dario/pig.html\npippo", r"url *= *https?://.*"+'\n', [5, 45], ['url = https://github.com/dario/pig.html\n'] }, Test_find_all{ "#.#......##.#..#..##........##....###...##...######.......#.....#..#......#...#........###.#..#.", r"#[.#]{4}##[.#]{4}##[.#]{4}###", [29, 49], ['#....###...##...####'] }, Test_find_all{ "#.#......##.#..#..##........##....###...##...######.......#.....#..#......#...#........###.#..#.", r".*#[.#]{4}##[.#]{4}##[.#]{4}###", [0, 49], ['#.#......##.#..#..##........##....###...##...####'] }, Test_find_all{ "1234 Aa dddd Aaf 12334 Aa opopo Aaf", r"Aa.+Aaf", [5, 16, 23, 35], ['Aa dddd Aaf', 'Aa opopo Aaf'] }, Test_find_all{ "@for something @endfor @for something else @endfor altro testo @for body @endfor uno due @for senza dire più @endfor pippo", r"@for.+@endfor", [0, 22, 23, 50, 63, 80, 89, 117], ['@for something @endfor', '@for something else @endfor', '@for body @endfor', '@for senza dire più @endfor'] }, Test_find_all{ "+++pippo+++\n elvo +++ pippo2 +++ +++ oggi+++", r"\+{3}.*\+{3}", [0, 11, 18, 32, 33, 44], ['+++pippo+++', '+++ pippo2 +++', '+++ oggi+++'] } ] ) const ( debug = true // true for debug println ) fn test_regex(){ // check capturing groups for c,to in cgroups_test_suite { // debug print if debug { println("$c [${to.src}] [q${to.q}] (${to.s}, ${to.e})") } mut re := regex.regex_opt(to.q) or { eprintln('err: $err') assert false continue } if to.cgn.len > 0 { re.group_csave_flag = true //re.group_csave = [-1].repeat(3*20+1) if debug { println("continuous save")} } else { if debug { println("NO continuous save")} } start, end := re.match_string(to.src) mut tmp_str := "" if start >= 0 && end > start{ tmp_str = to.src[start..end] } if start != to.s || end != to.e { //println("#$c [$to.src] q[$to.q] res[$tmp_str] $start, $end") eprintln("ERROR!") //C.printf("ERROR!! res:(%d, %d) refh:(%d, %d)\n",start, end, to.s, to.e) assert false continue } // check cgroups if to.cgn.len > 0 { if re.group_csave.len == 0 || re.group_csave[0] != to.cg[0] { eprintln("Capturing group len error! found: ${re.group_csave[0]} true ground: ${to.cg[0]}") assert false continue } // check captured groups mut ln := re.group_csave[0]*3 for ln > 0 { if re.group_csave[ln] != to.cg[ln] { eprintln("Capturing group failed on $ln item!") assert false } ln-- } // check named captured groups for k in to.cgn.keys() { if to.cgn[k] != (re.group_map[k]-1) { // we have -1 because the map not found is 0, in groups we start from 0 and we store using +1 eprintln("Named capturing group error! [$k]") assert false continue } } } else { // check normal captured groups if re.groups.len != to.cg.len { assert false } for ln:=0; ln < re.groups.len; ln++ { if re.groups[ln] != to.cg[ln] { eprintln("Capture group doesn't match:") eprintln("true ground: [${to.cg}]") eprintln("elaborated : [${re.groups}]") assert false } } } } // check find_all for c,to in find_all_test_suite { // debug print if debug { println("#$c [$to.src] q[$to.q] ($to.res, $to.res_str)") } mut re := regex.regex_opt(to.q) or { eprintln('err: $err') assert false continue } re.reset() res := re.find_all(to.src) if res != to.res { eprintln('err: find_all !!') if debug { println("#$c exp: $to.res calculated: $res") } assert false } res_str := re.find_all_str(to.src) if res_str != to.res_str { eprintln('err: find_all_str !!') if debug { println("#$c exp: $to.res_str calculated: $res_str") } assert false } } // check replace for c,to in match_test_suite_replace{ // debug print if debug { println("#$c [$to.src] q[$to.q] $to.r") } mut re := regex.regex_opt(to.q) or { eprintln('err: $err') assert false continue } res := re.replace(to.src,to.rep) if res != to.r { eprintln("ERROR: replace.") assert false continue } } // check replace simple for c,to in match_test_suite_replace_simple{ // debug print if debug { println("#$c [$to.src] q[$to.q] $to.r") } mut re := regex.regex_opt(to.q) or { eprintln('err: $err') assert false continue } res := re.replace_simple(to.src,to.rep) if res != to.r { eprintln("ERROR: replace.") assert false continue } } // check match and find for c,to in match_test_suite { // debug print if debug { println("#$c [$to.src] q[$to.q] $to.s $to.e") } // test the find if to.s > 0 { mut re := regex.regex_opt(to.q) or { eprintln('err: $err') assert false continue } // q_str := re.get_query() // eprintln("Query: $q_str") start,end := re.find(to.src) if start != to.s || end != to.e { err_str := re.get_parse_error_string(start) eprintln("ERROR : $err_str start: ${start} end: ${end}") assert false } else { //tmp_str := text[start..end] //println("found in [$start, $end] => [$tmp_str]") assert true } continue } // test the match mut re := regex.new() //re.debug = true re.compile_opt(to.q) or { eprintln('err: $err') assert false continue } //println("#$c [$to.src] q[$to.q]") start, end := re.match_string(to.src) mut tmp_str := "" if start >= 0 && end > start{ tmp_str = to.src[start..end] } if start != to.s || end != to.e { eprintln("#$c [$to.src] q[$to.q] res[$tmp_str] $start, $end") eprintln("ERROR!") //C.printf("ERROR!! res:(%d, %d) refh:(%d, %d)\n",start, end, to.s, to.e) assert false continue } // test the match predicate if to.s >= 0 { assert re.matches_string(to.src) } else { assert !re.matches_string(to.src) } // rerun to test consistency tmp_str1 := to.src.clone() start1, end1 := re.match_string(tmp_str1) if start1 != start || end1 != end { eprintln("two run ERROR!!") assert false continue } } if debug { println("DONE!") } } // test regex_base function fn test_regex_func(){ query := r"\d\dabcd" test_str := "78abcd" mut re, re_err, err_pos := regex.regex_base(query) if re_err == regex.compile_ok { start, end := re.match_string(test_str) assert (start == 0) && (end == 6) } else { eprintln("Error in query string in pos ${err_pos}") eprintln("Error: ${re.get_parse_error_string(re_err)}") assert false } } fn my_repl(re regex.RE, in_txt string, start int, end int) string { s0 := re.get_group_by_id(in_txt,0)[0..1] + "X" s1 := re.get_group_by_id(in_txt,1)[0..1] + "X" s2 := re.get_group_by_id(in_txt,2)[0..1] + "X" return "${s0}${s1}${s2}" } // test regex replace function fn test_regex_func_replace(){ filler := "E il primo dei tre regni dell'Oltretomba cristiano visitato da Dante nel corso del viaggio, con la guida di Virgilio." txt := r'"content": "They dont necessarily flag "you will be buying these shares on margin!"", "channel_id"' query := r'"(content":\s+")(.*)(, "channel_id")' mut re := regex.regex_opt(query) or { panic(err) } mut txt1 := "" mut txt2 := "" for _ in 0..3 { rnd := int(10+rand.u32() % 20) txt1 += txt + filler[0..rnd] + "\n" txt2 += "cXTX,X" + filler[0..rnd] + "\n" } result := re.replace_by_fn(txt1, my_repl) if debug { eprintln(result) eprintln(txt2) } assert result == txt2 } // test quantifier wrong sequences const( test_quantifier_sequences_list = [ r'+{3}.*+{3}', r'+{3}.*?{3}', r'+{3}.**{3}', r'+{3}.*\+{3}*', r'+{3}.*\+{3}+', r'+{3}.*\+{3}??', r'+{3}.*\+{3}{4}' ] ) fn test_quantifier_sequences(){ for pattern in test_quantifier_sequences_list { re, re_err, err_pos := regex.regex_base(pattern) if re_err != regex.err_syntax_error { eprintln("pattern: $pattern => $re_err") } assert re_err == regex.err_syntax_error } } // test group index in find struct Test_find_groups { src string q string s int // start index e int // end index res []int // groups indexes } const ( find_groups_test_suite = [ Test_find_groups{ "aabbbccccdd", r"(b+)(c+)", 2, 9, [2, 5, 5, 9], }, Test_find_groups{ "aabbbccccdd", r"(a+).*(c+)", 0, 9, [0, 2, 5, 9], }, Test_find_groups{ "aabbbccccdd", r"((b+).*)(d+)", 2, 11, [2, 9, 2, 5, 9, 11], }, ] ) fn test_groups_in_find(){ for test_obj in find_groups_test_suite { src_text := test_obj.src query := test_obj.q mut re := regex.regex_opt(query) or { panic(err) } start, end := re.find(src_text) // Debug print do not remove!! /* println("---------") println("src_text:[${src_text}]") println("query :[${query}]") println("[${start}, ${end}]") println(re.groups) mut gi := 0 for gi < re.groups.len { if re.groups[gi] >= 0 { println('${gi / 2} :[${src_text[re.groups[gi]..re.groups[gi + 1]]}]') } gi += 2 } */ // check assert start == test_obj.s assert end == test_obj.e assert re.groups == test_obj.res } }