diff options
Diffstat (limited to 'tests/stdlib/nre/find.nim')
-rw-r--r-- | tests/stdlib/nre/find.nim | 12 |
1 files changed, 6 insertions, 6 deletions
diff --git a/tests/stdlib/nre/find.nim b/tests/stdlib/nre/find.nim index caa953ff4..7e7555d73 100644 --- a/tests/stdlib/nre/find.nim +++ b/tests/stdlib/nre/find.nim @@ -3,23 +3,23 @@ import nre except toSeq import optional_nonstrict import times, strutils -suite "find": - test "find text": +block: # find + block: # find text check("3213a".find(re"[a-z]").match == "a") check(toSeq(findIter("1 2 3 4 5 6 7 8 ", re" ")).map( proc (a: RegexMatch): string = a.match ) == @[" ", " ", " ", " ", " ", " ", " ", " "]) - test "find bounds": + block: # find bounds check(toSeq(findIter("1 2 3 4 5 ", re" ")).map( proc (a: RegexMatch): Slice[int] = a.matchBounds ) == @[1..1, 3..3, 5..5, 7..7, 9..9]) - test "overlapping find": + block: # overlapping find check("222".findAll(re"22") == @["22"]) check("2222".findAll(re"22") == @["22", "22"]) - test "len 0 find": + block: # len 0 find check("".findAll(re"\ ") == newSeq[string]()) check("".findAll(re"") == @[""]) check("abc".findAll(re"") == @["", "", "", ""]) @@ -27,7 +27,7 @@ suite "find": check("word\r\lword".findAll(re"(*ANYCRLF)(?m)$") == @["", ""]) check("слово слово".findAll(re"(*U)\b") == @["", "", "", ""]) - test "bail early": + block: # bail early ## we expect nothing to be found and we should be bailing out early which means that ## the timing difference between searching in small and large data should be well ## within a tolerance margin |