bf51655a7b
Signed-off-by: Jacek J. Łakis <jacek.lakis@intel.com> Signed-off-by: Samuel Ortiz <sameo@linux.intel.com>
282 lines
6.5 KiB
Go
282 lines
6.5 KiB
Go
// Copyright 2015 The Go Authors. All rights reserved.
|
||
// Use of this source code is governed by a BSD-style
|
||
// license that can be found in the LICENSE file.
|
||
|
||
package runes
|
||
|
||
import (
|
||
"strings"
|
||
"testing"
|
||
"unicode"
|
||
|
||
"golang.org/x/text/cases"
|
||
"golang.org/x/text/language"
|
||
"golang.org/x/text/transform"
|
||
)
|
||
|
||
var (
|
||
toUpper = cases.Upper(language.Und)
|
||
toLower = cases.Lower(language.Und)
|
||
)
|
||
|
||
type spanformer interface {
|
||
transform.SpanningTransformer
|
||
}
|
||
|
||
func TestPredicate(t *testing.T) {
|
||
testConditional(t, func(rt *unicode.RangeTable, t, f spanformer) spanformer {
|
||
return If(Predicate(func(r rune) bool {
|
||
return unicode.Is(rt, r)
|
||
}), t, f)
|
||
})
|
||
}
|
||
|
||
func TestIn(t *testing.T) {
|
||
testConditional(t, func(rt *unicode.RangeTable, t, f spanformer) spanformer {
|
||
return If(In(rt), t, f)
|
||
})
|
||
}
|
||
|
||
func TestNotIn(t *testing.T) {
|
||
testConditional(t, func(rt *unicode.RangeTable, t, f spanformer) spanformer {
|
||
return If(NotIn(rt), f, t)
|
||
})
|
||
}
|
||
|
||
func testConditional(t *testing.T, f func(rt *unicode.RangeTable, t, f spanformer) spanformer) {
|
||
lower := f(unicode.Latin, toLower, toLower)
|
||
|
||
for i, tt := range []transformTest{{
|
||
desc: "empty",
|
||
szDst: large,
|
||
atEOF: true,
|
||
in: "",
|
||
out: "",
|
||
outFull: "",
|
||
t: lower,
|
||
}, {
|
||
desc: "small",
|
||
szDst: 1,
|
||
atEOF: true,
|
||
in: "B",
|
||
out: "b",
|
||
outFull: "b",
|
||
errSpan: transform.ErrEndOfSpan,
|
||
t: lower,
|
||
}, {
|
||
desc: "short dst",
|
||
szDst: 2,
|
||
atEOF: true,
|
||
in: "AAA",
|
||
out: "aa",
|
||
outFull: "aaa",
|
||
err: transform.ErrShortDst,
|
||
errSpan: transform.ErrEndOfSpan,
|
||
t: lower,
|
||
}, {
|
||
desc: "short dst writing error",
|
||
szDst: 1,
|
||
atEOF: false,
|
||
in: "A\x80",
|
||
out: "a",
|
||
outFull: "a\x80",
|
||
err: transform.ErrShortDst,
|
||
errSpan: transform.ErrEndOfSpan,
|
||
t: lower,
|
||
}, {
|
||
desc: "short dst writing incomplete rune",
|
||
szDst: 2,
|
||
atEOF: true,
|
||
in: "Σ\xc2",
|
||
out: "Σ",
|
||
outFull: "Σ\xc2",
|
||
err: transform.ErrShortDst,
|
||
t: f(unicode.Latin, toLower, nil),
|
||
}, {
|
||
desc: "short dst, longer",
|
||
szDst: 5,
|
||
atEOF: true,
|
||
in: "Hellø",
|
||
out: "Hell",
|
||
outFull: "Hellø",
|
||
err: transform.ErrShortDst,
|
||
// idem is used to test short buffers by forcing processing of full-rune increments.
|
||
t: f(unicode.Latin, Map(idem), nil),
|
||
}, {
|
||
desc: "short dst, longer, writing error",
|
||
szDst: 6,
|
||
atEOF: false,
|
||
in: "\x80Hello\x80",
|
||
out: "\x80Hello",
|
||
outFull: "\x80Hello\x80",
|
||
err: transform.ErrShortDst,
|
||
t: f(unicode.Latin, Map(idem), nil),
|
||
}, {
|
||
desc: "short src",
|
||
szDst: 2,
|
||
atEOF: false,
|
||
in: "A\xc2",
|
||
out: "a",
|
||
outFull: "a\xc2",
|
||
err: transform.ErrShortSrc,
|
||
errSpan: transform.ErrEndOfSpan,
|
||
t: lower,
|
||
}, {
|
||
desc: "short src no change",
|
||
szDst: 2,
|
||
atEOF: false,
|
||
in: "a\xc2",
|
||
out: "a",
|
||
outFull: "a\xc2",
|
||
err: transform.ErrShortSrc,
|
||
errSpan: transform.ErrShortSrc,
|
||
nSpan: 1,
|
||
t: lower,
|
||
}, {
|
||
desc: "invalid input, atEOF",
|
||
szDst: large,
|
||
atEOF: true,
|
||
in: "\x80",
|
||
out: "\x80",
|
||
outFull: "\x80",
|
||
t: lower,
|
||
}, {
|
||
desc: "invalid input, !atEOF",
|
||
szDst: large,
|
||
atEOF: false,
|
||
in: "\x80",
|
||
out: "\x80",
|
||
outFull: "\x80",
|
||
t: lower,
|
||
}, {
|
||
desc: "invalid input, incomplete rune atEOF",
|
||
szDst: large,
|
||
atEOF: true,
|
||
in: "\xc2",
|
||
out: "\xc2",
|
||
outFull: "\xc2",
|
||
t: lower,
|
||
}, {
|
||
desc: "nop",
|
||
szDst: large,
|
||
atEOF: true,
|
||
in: "Hello World!",
|
||
out: "Hello World!",
|
||
outFull: "Hello World!",
|
||
t: f(unicode.Latin, nil, nil),
|
||
}, {
|
||
desc: "nop in",
|
||
szDst: large,
|
||
atEOF: true,
|
||
in: "THIS IS α ΤΕΣΤ",
|
||
out: "this is α ΤΕΣΤ",
|
||
outFull: "this is α ΤΕΣΤ",
|
||
errSpan: transform.ErrEndOfSpan,
|
||
t: f(unicode.Greek, nil, toLower),
|
||
}, {
|
||
desc: "nop in latin",
|
||
szDst: large,
|
||
atEOF: true,
|
||
in: "THIS IS α ΤΕΣΤ",
|
||
out: "THIS IS α τεστ",
|
||
outFull: "THIS IS α τεστ",
|
||
errSpan: transform.ErrEndOfSpan,
|
||
t: f(unicode.Latin, nil, toLower),
|
||
}, {
|
||
desc: "nop not in",
|
||
szDst: large,
|
||
atEOF: true,
|
||
in: "THIS IS α ΤΕΣΤ",
|
||
out: "this is α ΤΕΣΤ",
|
||
outFull: "this is α ΤΕΣΤ",
|
||
errSpan: transform.ErrEndOfSpan,
|
||
t: f(unicode.Latin, toLower, nil),
|
||
}, {
|
||
desc: "pass atEOF is true when at end",
|
||
szDst: large,
|
||
atEOF: true,
|
||
in: "hello",
|
||
out: "HELLO",
|
||
outFull: "HELLO",
|
||
errSpan: transform.ErrEndOfSpan,
|
||
t: f(unicode.Latin, upperAtEOF{}, nil),
|
||
}, {
|
||
desc: "pass atEOF is true when at end of segment",
|
||
szDst: large,
|
||
atEOF: true,
|
||
in: "hello ",
|
||
out: "HELLO ",
|
||
outFull: "HELLO ",
|
||
errSpan: transform.ErrEndOfSpan,
|
||
t: f(unicode.Latin, upperAtEOF{}, nil),
|
||
}, {
|
||
desc: "don't pass atEOF is true when atEOF is false",
|
||
szDst: large,
|
||
atEOF: false,
|
||
in: "hello",
|
||
out: "",
|
||
outFull: "HELLO",
|
||
err: transform.ErrShortSrc,
|
||
errSpan: transform.ErrShortSrc,
|
||
t: f(unicode.Latin, upperAtEOF{}, nil),
|
||
}, {
|
||
desc: "pass atEOF is true when at end, no change",
|
||
szDst: large,
|
||
atEOF: true,
|
||
in: "HELLO",
|
||
out: "HELLO",
|
||
outFull: "HELLO",
|
||
t: f(unicode.Latin, upperAtEOF{}, nil),
|
||
}, {
|
||
desc: "pass atEOF is true when at end of segment, no change",
|
||
szDst: large,
|
||
atEOF: true,
|
||
in: "HELLO ",
|
||
out: "HELLO ",
|
||
outFull: "HELLO ",
|
||
t: f(unicode.Latin, upperAtEOF{}, nil),
|
||
}, {
|
||
desc: "large input ASCII",
|
||
szDst: 12000,
|
||
atEOF: false,
|
||
in: strings.Repeat("HELLO", 2000),
|
||
out: strings.Repeat("hello", 2000),
|
||
outFull: strings.Repeat("hello", 2000),
|
||
errSpan: transform.ErrEndOfSpan,
|
||
err: nil,
|
||
t: lower,
|
||
}, {
|
||
desc: "large input non-ASCII",
|
||
szDst: 12000,
|
||
atEOF: false,
|
||
in: strings.Repeat("\u3333", 2000),
|
||
out: strings.Repeat("\u3333", 2000),
|
||
outFull: strings.Repeat("\u3333", 2000),
|
||
err: nil,
|
||
t: lower,
|
||
}} {
|
||
tt.check(t, i)
|
||
}
|
||
}
|
||
|
||
// upperAtEOF is a strange Transformer that converts text to uppercase, but only
|
||
// if atEOF is true.
|
||
type upperAtEOF struct{ transform.NopResetter }
|
||
|
||
func (upperAtEOF) Transform(dst, src []byte, atEOF bool) (nDst, nSrc int, err error) {
|
||
if !atEOF {
|
||
return 0, 0, transform.ErrShortSrc
|
||
}
|
||
return toUpper.Transform(dst, src, atEOF)
|
||
}
|
||
|
||
func (upperAtEOF) Span(src []byte, atEOF bool) (n int, err error) {
|
||
if !atEOF {
|
||
return 0, transform.ErrShortSrc
|
||
}
|
||
return toUpper.Span(src, atEOF)
|
||
}
|
||
|
||
func BenchmarkConditional(b *testing.B) {
|
||
doBench(b, If(In(unicode.Hangul), transform.Nop, transform.Nop))
|
||
}
|