123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109 |
- import std/re
- proc testAll() =
- doAssert match("(a b c)", rex"\( .* \)")
- doAssert match("WHiLe", re("while", {reIgnoreCase}))
- doAssert "0158787".match(re"\d+")
- doAssert "ABC 0232".match(re"\w+\s+\d+")
- doAssert "ABC".match(rex"\d+ | \w+")
- {.push warnings:off.}
- doAssert matchLen("key", re"\b[a-zA-Z_]+[a-zA-Z_0-9]*\b") == 3
- {.pop.}
- var pattern = re"[a-z0-9]+\s*=\s*[a-z0-9]+"
- doAssert matchLen("key1= cal9", pattern) == 11
- doAssert find("_____abc_______", re"abc") == 5
- doAssert findBounds("_____abc_______", re"abc") == (5,7)
- var matches: array[6, string]
- if match("abcdefg", re"c(d)ef(g)", matches, 2):
- doAssert matches[0] == "d"
- doAssert matches[1] == "g"
- else:
- doAssert false
- if "abc" =~ re"(a)bcxyz|(\w+)":
- doAssert matches[1] == "abc"
- else:
- doAssert false
- if "abc" =~ re"(cba)?.*":
- doAssert matches[0] == ""
- else: doAssert false
- if "abc" =~ re"().*":
- doAssert matches[0] == ""
- else: doAssert false
- doAssert "var1=key; var2=key2".endsWith(re"\w+=\w+")
- doAssert("var1=key; var2=key2".replacef(re"(\w+)=(\w+)", "$1<-$2$2") ==
- "var1<-keykey; var2<-key2key2")
- doAssert("var1=key; var2=key2".replace(re"(\w+)=(\w+)", "$1<-$2$2") ==
- "$1<-$2$2; $1<-$2$2")
- var accum: seq[string] = @[]
- for word in split("00232this02939is39an22example111", re"\d+"):
- accum.add(word)
- doAssert(accum == @["", "this", "is", "an", "example", ""])
- accum = @[]
- for word in split("00232this02939is39an22example111", re"\d+", maxsplit=2):
- accum.add(word)
- doAssert(accum == @["", "this", "is39an22example111"])
- accum = @[]
- for word in split("AAA : : BBB", re"\s*:\s*"):
- accum.add(word)
- doAssert(accum == @["AAA", "", "BBB"])
- doAssert(split("abc", re"") == @["a", "b", "c"])
- doAssert(split("", re"") == @[])
- doAssert(split("a;b;c", re";") == @["a", "b", "c"])
- doAssert(split(";a;b;c", re";") == @["", "a", "b", "c"])
- doAssert(split(";a;b;c;", re";") == @["", "a", "b", "c", ""])
- doAssert(split("a;b;c;", re";") == @["a", "b", "c", ""])
- doAssert(split("00232this02939is39an22example111", re"\d+", maxsplit=2) == @["", "this", "is39an22example111"])
- for x in findAll("abcdef", re"^{.}", 3):
- doAssert x == "d"
- accum = @[]
- for x in findAll("abcdef", re".", 3):
- accum.add(x)
- doAssert(accum == @["d", "e", "f"])
- doAssert("XYZ".find(re"^\d*") == 0)
- doAssert("XYZ".match(re"^\d*") == true)
- block:
- var matches: array[16, string]
- if match("abcdefghijklmnop", re"(a)(b)(c)(d)(e)(f)(g)(h)(i)(j)(k)(l)(m)(n)(o)(p)", matches):
- for i in 0..matches.high:
- doAssert matches[i] == $chr(i + 'a'.ord)
- else:
- doAssert false
- block: # Buffer based RE
- var cs: cstring = "_____abc_______"
- doAssert(cs.find(re"abc", bufSize=15) == 5)
- doAssert(cs.matchLen(re"_*abc", bufSize=15) == 8)
- doAssert(cs.matchLen(re"abc", start=5, bufSize=15) == 3)
- doAssert(cs.matchLen(re"abc", start=5, bufSize=7) == -1)
- doAssert(cs.matchLen(re"abc_*", start=5, bufSize=10) == 5)
- var accum: seq[string] = @[]
- for x in cs.findAll(re"[a-z]", start=3, bufSize=15):
- accum.add($x)
- doAssert(accum == @["a","b","c"])
- block:
- # bug #9306
- doAssert replace("bar", re"^", "foo") == "foobar"
- doAssert replace("foo", re"", "-") == "-foo"
- doAssert replace("foo", re"$", "bar") == "foobar"
- testAll()
|