discard """
targets: "c cpp js"
"""
import std/strutils
# xxx each instance of `disableVm` and `when not defined js:` should eventually be fixed
template rejectParse(e) =
try:
discard e
raise newException(AssertionDefect, "This was supposed to fail: $#!" % astToStr(e))
except ValueError: discard
template disableVm(body) =
when nimvm: discard
else: body
template main() =
block: # strip
doAssert strip(" ha ") == "ha"
doAssert strip(" foofoofoo ") == "foofoofoo"
doAssert strip("sfoofoofoos", chars = {'s'}) == "foofoofoo"
doAssert strip("barfoofoofoobar", chars = {'b', 'a', 'r'}) == "foofoofoo"
doAssert strip("stripme but don't strip this stripme",
chars = {'s', 't', 'r', 'i', 'p', 'm', 'e'}) ==
" but don't strip this "
doAssert strip("sfoofoofoos", leading = false, chars = {'s'}) == "sfoofoofoo"
doAssert strip("sfoofoofoos", trailing = false, chars = {'s'}) == "foofoofoos"
block:
let a = "xxxxxx"
doAssert a.strip(chars={'x'}).len == 0
doAssert "".strip(chars={'x'}).len == 0
doAssert " ".strip(chars={'x'}) == " "
doAssert "xxx xxx".strip(chars={'x'}) == " "
doAssert "xxx wind".strip(chars={'x'}) == " wind"
doAssert "xxx iii".strip(chars={'i'}) == "xxx "
doAssert "x".strip(leading = false, chars={'x'}).len == 0
doAssert "x".strip(trailing = false, chars={'x'}).len == 0
doAssert "x".strip(leading = false, trailing = false, chars={'x'}) == "x"
block: # split
var ret: seq[string] # or use `toSeq` or `collect`
for p in split("/home/a1:xyz:/usr/bin", {':'}): ret.add p
doAssert ret == @["/home/a1", "xyz", "/usr/bin"]
let s = " this is an example "
let s2 = ":this;is;an:example;;"
doAssert s.split() == @["", "this", "is", "an", "example", "", ""]
doAssert s2.split(seps = {':', ';'}) == @["", "this", "is", "an", "example",
"", ""]
doAssert s.split(maxsplit = 4) == @["", "this", "is", "an", "example "]
doAssert s.split(' ', maxsplit = 1) == @["", "this is an example "]
doAssert s.split(" ", maxsplit = 4) == @["", "this", "is", "an", "example "]
block: # splitLines
let fixture = "a\nb\rc\r\nd"
doAssert len(fixture.splitLines) == 4
doAssert splitLines(fixture) == @["a", "b", "c", "d"]
doAssert splitLines(fixture, keepEol=true) == @["a\n", "b\r", "c\r\n", "d"]
block: # rsplit
doAssert rsplit("foo bar", seps = Whitespace) == @["foo", "bar"]
doAssert rsplit(" foo bar", seps = Whitespace, maxsplit = 1) == @[" foo", "bar"]
doAssert rsplit(" foo bar ", seps = Whitespace, maxsplit = 1) == @[
" foo bar", ""]
doAssert rsplit(":foo:bar", sep = ':') == @["", "foo", "bar"]
doAssert rsplit(":foo:bar", sep = ':', maxsplit = 2) == @["", "foo", "bar"]
doAssert rsplit(":foo:bar", sep = ':', maxsplit = 3) == @["", "foo", "bar"]
doAssert rsplit("foothebar", sep = "the") == @["foo", "bar"]
block: # splitWhitespace
let s = " this is an example "
doAssert s.splitWhitespace() == @["this", "is", "an", "example"]
doAssert s.splitWhitespace(maxsplit = 1) == @["this", "is an example "]
doAssert s.splitWhitespace(maxsplit = 2) == @["this", "is", "an example "]
doAssert s.splitWhitespace(maxsplit = 3) == @["this", "is", "an", "example "]
doAssert s.splitWhitespace(maxsplit = 4) == @["this", "is", "an", "example"]
block: # removeSuffix
var s = "hello\n\r"
s.removeSuffix
doAssert s == "hello"
s.removeSuffix
doAssert s == "hello"
s = "hello\n\n"
s.removeSuffix
doAssert s == "hello"
s = "hello\r"
s.removeSuffix
doAssert s == "hello"
s = "hello \n there"
s.removeSuffix
doAssert s == "hello \n there"
s = "hello"
s.removeSuffix("llo")
doAssert s == "he"
s.removeSuffix('e')
doAssert s == "h"
s = "hellos"
s.removeSuffix({'s','z'})
doAssert s == "hello"
s.removeSuffix({'l','o'})
doAssert s == "he"
s = "aeiou"
s.removeSuffix("")
doAssert s == "aeiou"
s = ""
s.removeSuffix("")
doAssert s == ""
s = " "
s.removeSuffix
doAssert s == " "
s = " "
s.removeSuffix("")
doAssert s == " "
s = " "
s.removeSuffix(" ")
doAssert s == " "
s = " "
s.removeSuffix(' ')
doAssert s == ""
# Contrary to Chomp in other languages
# empty string does not change behaviour
s = "hello\r\n\r\n"
s.removeSuffix("")
doAssert s == "hello\r\n\r\n"
block: # removePrefix
var s = "\n\rhello"
s.removePrefix
doAssert s == "hello"
s.removePrefix
doAssert s == "hello"
s = "\n\nhello"
s.removePrefix
doAssert s == "hello"
s = "\rhello"
s.removePrefix
doAssert s == "hello"
s = "hello \n there"
s.removePrefix
doAssert s == "hello \n there"
s = "hello"
s.removePrefix("hel")
doAssert s == "lo"
s.removePrefix('l')
doAssert s == "o"
s = "hellos"
s.removePrefix({'h','e'})
doAssert s == "llos"
s.removePrefix({'l','o'})
doAssert s == "s"
s = "aeiou"
s.removePrefix("")
doAssert s == "aeiou"
s = ""
s.removePrefix("")
doAssert s == ""
s = " "
s.removePrefix
doAssert s == " "
s = " "
s.removePrefix("")
doAssert s == " "
s = " "
s.removePrefix(" ")
doAssert s == " "
s = " "
s.removePrefix(' ')
doAssert s == ""
# Contrary to Chomp in other languages
# empty string does not change behaviour
s = "\r\n\r\nhello"
s.removePrefix("")
doAssert s == "\r\n\r\nhello"
block: # delete
var s = "0123456789ABCDEFGH"
delete(s, 4, 5)
doAssert s == "01236789ABCDEFGH"
delete(s, s.len-1, s.len-1)
doAssert s == "01236789ABCDEFG"
delete(s, 0, 0)
doAssert s == "1236789ABCDEFG"
block: # find
doAssert "0123456789ABCDEFGH".find('A') == 10
doAssert "0123456789ABCDEFGH".find('A', 5) == 10
doAssert "0123456789ABCDEFGH".find('A', 5, 10) == 10
doAssert "0123456789ABCDEFGH".find('A', 5, 9) == -1
doAssert "0123456789ABCDEFGH".find("A") == 10
doAssert "0123456789ABCDEFGH".find("A", 5) == 10
doAssert "0123456789ABCDEFGH".find("A", 5, 10) == 10
doAssert "0123456789ABCDEFGH".find("A", 5, 9) == -1
doAssert "0123456789ABCDEFGH".find({'A'..'C'}) == 10
doAssert "0123456789ABCDEFGH".find({'A'..'C'}, 5) == 10
doAssert "0123456789ABCDEFGH".find({'A'..'C'}, 5, 10) == 10
doAssert "0123456789ABCDEFGH".find({'A'..'C'}, 5, 9) == -1
block: # rfind
doAssert "0123456789ABCDEFGAH".rfind('A') == 17
doAssert "0123456789ABCDEFGAH".rfind('A', last=13) == 10
doAssert "0123456789ABCDEFGAH".rfind('H', last=13) == -1
doAssert "0123456789ABCDEFGAH".rfind("A") == 17
doAssert "0123456789ABCDEFGAH".rfind("A", last=13) == 10
doAssert "0123456789ABCDEFGAH".rfind("H", last=13) == -1
doAssert "0123456789ABCDEFGAH".rfind({'A'..'C'}) == 17
doAssert "0123456789ABCDEFGAH".rfind({'A'..'C'}, last=13) == 12
doAssert "0123456789ABCDEFGAH".rfind({'G'..'H'}, last=13) == -1
doAssert "0123456789ABCDEFGAH".rfind('A', start=18) == -1
doAssert "0123456789ABCDEFGAH".rfind('A', start=11, last=17) == 17
doAssert "0123456789ABCDEFGAH".rfind("0", start=0) == 0
doAssert "0123456789ABCDEFGAH".rfind("0", start=1) == -1
doAssert "0123456789ABCDEFGAH".rfind("H", start=11) == 18
doAssert "0123456789ABCDEFGAH".rfind({'0'..'9'}, start=5) == 9
doAssert "0123456789ABCDEFGAH".rfind({'0'..'9'}, start=10) == -1
doAssert "/1/2/3".rfind('/') == 4
doAssert "/1/2/3".rfind('/', last=1) == 0
doAssert "/1/2/3".rfind('0') == -1
block: # trimZeros
var x = "1200"
x.trimZeros()
doAssert x == "1200"
x = "120.0"
x.trimZeros()
doAssert x == "120"
x = "0."
x.trimZeros()
doAssert x == "0"
x = "1.0e2"
x.trimZeros()
doAssert x == "1e2"
x = "78.90"
x.trimZeros()
doAssert x == "78.9"
x = "1.23e4"
x.trimZeros()
doAssert x == "1.23e4"
x = "1.01"
x.trimZeros()
doAssert x == "1.01"
x = "1.1001"
x.trimZeros()
doAssert x == "1.1001"
x = "0.0"
x.trimZeros()
doAssert x == "0"
x = "0.01"
x.trimZeros()
doAssert x == "0.01"
x = "1e0"
x.trimZeros()
doAssert x == "1e0"
block: # countLines
proc assertCountLines(s: string) = doAssert s.countLines == s.splitLines.len
assertCountLines("")
assertCountLines("\n")
assertCountLines("\n\n")
assertCountLines("abc")
assertCountLines("abc\n123")
assertCountLines("abc\n123\n")
assertCountLines("\nabc\n123")
assertCountLines("\nabc\n123\n")
block: # parseBinInt, parseHexInt, parseOctInt
# binary
doAssert "0b1111".parseBinInt == 15
doAssert "0B1111".parseBinInt == 15
doAssert "1111".parseBinInt == 15
doAssert "1110".parseBinInt == 14
doAssert "1_1_1_1".parseBinInt == 15
doAssert "0b1_1_1_1".parseBinInt == 15
rejectParse "".parseBinInt
rejectParse "_".parseBinInt
rejectParse "0b".parseBinInt
rejectParse "0b1234".parseBinInt
# hex
doAssert "0x72".parseHexInt == 114
doAssert "0X72".parseHexInt == 114
doAssert "#72".parseHexInt == 114
doAssert "72".parseHexInt == 114
doAssert "FF".parseHexInt == 255
doAssert "ff".parseHexInt == 255
doAssert "fF".parseHexInt == 255
doAssert "0x7_2".parseHexInt == 114
rejectParse "".parseHexInt
rejectParse "_".parseHexInt
rejectParse "0x".parseHexInt
rejectParse "0xFFG".parseHexInt
rejectParse "reject".parseHexInt
# octal
doAssert "0o17".parseOctInt == 15
doAssert "0O17".parseOctInt == 15
doAssert "17".parseOctInt == 15
doAssert "10".parseOctInt == 8
doAssert "0o1_0_0".parseOctInt == 64
rejectParse "".parseOctInt
rejectParse "_".parseOctInt
rejectParse "0o".parseOctInt
rejectParse "9".parseOctInt
rejectParse "0o9".parseOctInt
rejectParse "reject".parseOctInt
block: # parseHexStr
doAssert "".parseHexStr == ""
doAssert "00Ff80".parseHexStr == "\0\xFF\x80"
try:
discard "00Ff8".parseHexStr
doAssert false, "Should raise ValueError"
except ValueError:
discard
try:
discard "0k".parseHexStr
doAssert false, "Should raise ValueError"
except ValueError:
discard
doAssert "".toHex == ""
doAssert "\x00\xFF\x80".toHex == "00FF80"
doAssert "0123456789abcdef".parseHexStr.toHex == "0123456789ABCDEF"
block: # toHex
doAssert(toHex(100i16, 32) == "00000000000000000000000000000064")
doAssert(toHex(-100i16, 32) == "FFFFFFFFFFFFFFFFFFFFFFFFFFFFFF9C")
when not defined js:
doAssert(toHex(high(uint64)) == "FFFFFFFFFFFFFFFF")
doAssert(toHex(high(uint64), 16) == "FFFFFFFFFFFFFFFF")
doAssert(toHex(high(uint64), 32) == "0000000000000000FFFFFFFFFFFFFFFF")
block: # insertSep
doAssert(insertSep($1000_000) == "1_000_000")
doAssert(insertSep($232) == "232")
doAssert(insertSep($12345, ',') == "12,345")
doAssert(insertSep($0) == "0")
block: # repeat, spaces
doAssert(' '.repeat(8) == " ")
doAssert(" ".repeat(8) == " ")
doAssert(spaces(8) == " ")
doAssert(' '.repeat(0) == "")
doAssert(" ".repeat(0) == "")
doAssert(spaces(0) == "")
block: # toBin, toOct
block:# bug #11369
var num: int64 = -1
when not defined js:
doAssert num.toBin(64) == "1111111111111111111111111111111111111111111111111111111111111111"
doAssert num.toOct(24) == "001777777777777777777777"
block: # replace
doAssert "oo".replace("", "abc") == "oo"
# bug #8911
static:
let a = ""
let a2 = a.replace("\n", "\\n")
static:
let b = "b"
let b2 = b.replace("\n", "\\n")
block:
let c = ""
let c2 = c.replace("\n", "\\n")
block: # replaceWord
doAssert "-ld a-ldz -ld".replaceWord("-ld") == " a-ldz "
doAssert "-lda-ldz -ld abc".replaceWord("-ld") == "-lda-ldz abc"
doAssert "-lda-ldz -ld abc".replaceWord("") == "-lda-ldz -ld abc"
block: # multiReplace
doAssert "abba".multiReplace(("a", "b"), ("b", "a")) == "baab"
doAssert "Hello World.".multiReplace(("ello", "ELLO"), ("World.",
"PEOPLE!")) == "HELLO PEOPLE!"
doAssert "aaaa".multiReplace(("a", "aa"), ("aa", "bb")) == "aaaaaaaa"
# `parseEnum`, ref issue #14030
# check enum defined at top level # xxx this is probably irrelevant, and pollutes scope
# for remaining tests
type
Foo = enum
A = -10
B = "bb"
C = (-5, "ccc")
D = 15
E = "ee" # check that we count enum fields correctly
block: # parseEnum
block:
let a = parseEnum[Foo]("A")
let b = parseEnum[Foo]("bb")
let c = parseEnum[Foo]("ccc")
let d = parseEnum[Foo]("D")
let e = parseEnum[Foo]("ee")
doAssert a == A
doAssert b == B
doAssert c == C
doAssert d == D
doAssert e == E
try:
let f = parseEnum[Foo]("Bar")
doAssert false
except ValueError:
discard
# finally using default
let g = parseEnum[Foo]("Bar", A)
doAssert g == A
block: # check enum defined in block
type
Bar = enum
V
W = "ww"
X = (3, "xx")
Y = 10
Z = "zz" # check that we count enum fields correctly
let a = parseEnum[Bar]("V")
let b = parseEnum[Bar]("ww")
let c = parseEnum[Bar]("xx")
let d = parseEnum[Bar]("Y")
let e = parseEnum[Bar]("zz")
doAssert a == V
doAssert b == W
doAssert c == X
doAssert d == Y
doAssert e == Z
try:
let f = parseEnum[Bar]("Baz")
doAssert false
except ValueError:
discard
# finally using default
let g = parseEnum[Bar]("Baz", V)
doAssert g == V
block: # check ambiguous enum fails to parse
type
Ambig = enum
f1 = "A"
f2 = "B"
f3 = "A"
doAssert not compiles((let a = parseEnum[Ambig]("A")))
block: # check almost ambiguous enum
type
AlmostAmbig = enum
f1 = "someA"
f2 = "someB"
f3 = "SomeA"
let a = parseEnum[AlmostAmbig]("someA")
let b = parseEnum[AlmostAmbig]("someB")
let c = parseEnum[AlmostAmbig]("SomeA")
doAssert a == f1
doAssert b == f2
doAssert c == f3
block: # parseEnum TODO: merge above
type MyEnum = enum enA, enB, enC, enuD, enE
doAssert parseEnum[MyEnum]("enu_D") == enuD
doAssert parseEnum("invalid enum value", enC) == enC
block: # indentation
doAssert 0 == indentation """
hey
low
there
"""
doAssert 2 == indentation """
hey
low
there
"""
doAssert 2 == indentation """ hey
low
there
"""
doAssert 2 == indentation """ hey
low
there"""
doAssert 0 == indentation ""
doAssert 0 == indentation " \n \n"
doAssert 0 == indentation " "
block: # indent
doAssert " foo\n bar".indent(4, "Q") == "QQQQ foo\nQQQQ bar"
block: # unindent
doAssert """~~!!foo
~~!!bar
~~!!baz""".unindent(2, "~~!!") == "foo\nbar\nbaz"
doAssert """~~!!foo
~~!!bar
~~!!baz""".unindent(2, "~~!!aa") == "~~!!foo\n~~!!bar\n~~!!baz"
doAssert """~~foo
~~ bar
~~ baz""".unindent(4, "~") == "foo\n bar\n baz"
doAssert """foo
bar
baz
""".unindent(4) == "foo\nbar\nbaz\n"
doAssert """foo
bar
baz
""".unindent(2) == "foo\n bar\n baz\n"
doAssert """foo
bar
baz
""".unindent(100) == "foo\nbar\nbaz\n"
doAssert """foo
foo
bar
""".unindent() == "foo\nfoo\nbar\n"
block: # formatBiggestFloat
disableVm:
doAssert formatBiggestFloat(1234.567, ffDecimal, -1) == "1234.567000"
when not defined(js):
doAssert formatBiggestFloat(1234.567, ffDecimal, 0) == "1235." # bugs 8242, 12586
doAssert formatBiggestFloat(1234.567, ffDecimal, 1) == "1234.6"
doAssert formatBiggestFloat(0.00000000001, ffDecimal, 11) == "0.00000000001"
doAssert formatBiggestFloat(0.00000000001, ffScientific, 1, ',') in
["1,0e-11", "1,0e-011"]
block: # formatFloat
disableVm:
# bug #6589
when not defined(js):
doAssert formatFloat(123.456, ffScientific, precision = -1) == "1.234560e+02"
block: # `%`
doAssert "$# $3 $# $#" % ["a", "b", "c"] == "a c b c"
doAssert "${1}12 ${-1}$2" % ["a", "b"] == "a12 bb"
doAssert "$animal eats $food." % ["animal", "The cat", "food", "fish"] ==
"The cat eats fish."
block: # formatSize
disableVm:
when not defined(js):
doAssert formatSize((1'i64 shl 31) + (300'i64 shl 20)) == "2.293GiB" # <=== bug #8231
doAssert formatSize((2.234*1024*1024).int) == "2.234MiB"
doAssert formatSize(4096) == "4KiB"
doAssert formatSize(4096, prefix = bpColloquial, includeSpace = true) == "4 kB"
doAssert formatSize(4096, includeSpace = true) == "4 KiB"
doAssert formatSize(5_378_934, prefix = bpColloquial, decimalSep = ',') == "5,13MB"
block: # formatEng
disableVm:
doAssert formatEng(0, 2, trim = false) == "0.00"
doAssert formatEng(0, 2) == "0"
doAssert formatEng(53, 2, trim = false) == "53.00"
doAssert formatEng(0.053, 2, trim = false) == "53.00e-3"
doAssert formatEng(0.053, 4, trim = false) == "53.0000e-3"
doAssert formatEng(0.053, 4, trim = true) == "53e-3"
doAssert formatEng(0.053, 0) == "53e-3"
doAssert formatEng(52731234) == "52.731234e6"
doAssert formatEng(-52731234) == "-52.731234e6"
doAssert formatEng(52731234, 1) == "52.7e6"
doAssert formatEng(-52731234, 1) == "-52.7e6"
doAssert formatEng(52731234, 1, decimalSep = ',') == "52,7e6"
doAssert formatEng(-52731234, 1, decimalSep = ',') == "-52,7e6"
doAssert formatEng(4100, siPrefix = true, unit = "V") == "4.1 kV"
doAssert formatEng(4.1, siPrefix = true, unit = "V",
useUnitSpace = true) == "4.1 V"
doAssert formatEng(4.1, siPrefix = true) == "4.1" # Note lack of space
doAssert formatEng(4100, siPrefix = true) == "4.1 k"
doAssert formatEng(4.1, siPrefix = true, unit = "",
useUnitSpace = true) == "4.1 " # Includes space
doAssert formatEng(4100, siPrefix = true, unit = "") == "4.1 k"
doAssert formatEng(4100) == "4.1e3"
doAssert formatEng(4100, unit = "V", useUnitSpace = true) == "4.1e3 V"
doAssert formatEng(4100, unit = "", useUnitSpace = true) == "4.1e3 "
# Don't use SI prefix as number is too big
doAssert formatEng(3.1e22, siPrefix = true, unit = "a",
useUnitSpace = true) == "31e21 a"
# Don't use SI prefix as number is too small
doAssert formatEng(3.1e-25, siPrefix = true, unit = "A",
useUnitSpace = true) == "310e-27 A"
block: # align
doAssert align("abc", 4) == " abc"
doAssert align("a", 0) == "a"
doAssert align("1232", 6) == " 1232"
doAssert align("1232", 6, '#') == "##1232"
block: # alignLeft
doAssert alignLeft("abc", 4) == "abc "
doAssert alignLeft("a", 0) == "a"
doAssert alignLeft("1232", 6) == "1232 "
doAssert alignLeft("1232", 6, '#') == "1232##"
block: # center
doAssert center("foo", 13) == " foo "
doAssert center("foo", 0) == "foo"
doAssert center("foo", 3, fillChar = 'a') == "foo"
doAssert center("foo", 10, fillChar = '\t') == "\t\t\tfoo\t\t\t\t"
block: # count
doAssert count("foofoofoo", "foofoo") == 1
doAssert count("foofoofoo", "foofoo", overlapping = true) == 2
doAssert count("foofoofoo", 'f') == 3
doAssert count("foofoofoobar", {'f', 'b'}) == 4
block: # isAlphaAscii
doAssert isAlphaAscii('r')
doAssert isAlphaAscii('A')
doAssert(not isAlphaAscii('$'))
block: # isAlphaNumeric
doAssert isAlphaNumeric('3')
doAssert isAlphaNumeric('R')
doAssert(not isAlphaNumeric('!'))
block: # isDigit
doAssert isDigit('3')
doAssert(not isDigit('a'))
doAssert(not isDigit('%'))
block: # isSpaceAscii
doAssert isSpaceAscii('\t')
doAssert isSpaceAscii('\l')
doAssert(not isSpaceAscii('A'))
block: # isEmptyOrWhitespace
doAssert(isEmptyOrWhitespace(""))
doAssert(isEmptyOrWhitespace(" "))
doAssert(isEmptyOrWhitespace("\t\l \v\r\f"))
doAssert(not isEmptyOrWhitespace("ABc \td"))
block: # isLowerAscii
doAssert isLowerAscii('a')
doAssert isLowerAscii('z')
doAssert(not isLowerAscii('A'))
doAssert(not isLowerAscii('5'))
doAssert(not isLowerAscii('&'))
doAssert(not isLowerAscii(' '))
block: # isUpperAscii
doAssert isUpperAscii('A')
doAssert(not isUpperAscii('b'))
doAssert(not isUpperAscii('5'))
doAssert(not isUpperAscii('%'))
block: # unescape
doAssert(unescape(r"\x013", "", "") == "\x013")
block: # join
doAssert join(["foo", "bar", "baz"]) == "foobarbaz"
doAssert join(@["foo", "bar", "baz"], ", ") == "foo, bar, baz"
doAssert join([1, 2, 3]) == "123"
doAssert join(@[1, 2, 3], ", ") == "1, 2, 3"
block: # startsWith / endsWith
var s = "abcdef"
doAssert s.startsWith('a')
doAssert s.startsWith('b') == false
doAssert s.endsWith('f')
doAssert s.endsWith('a') == false
doAssert s.endsWith('\0') == false
static: main()
main()