test: remove semiocolons.

The ken directory is untouched so we have some examples with explicit semis.

R=gri
CC=golang-dev
https://golang.org/cl/2157041
This commit is contained in:
Rob Pike 2010-09-04 10:36:13 +10:00
parent cd8f4cd206
commit 4f61fc96b2
88 changed files with 2512 additions and 2516 deletions

View file

@ -9,71 +9,71 @@ package main
type I interface {}
const (
// assume all types behave similarly to int8/uint8
Int8 int8 = 101;
Minus1 int8 = -1;
Uint8 uint8 = 102;
Const = 103;
Int8 int8 = 101
Minus1 int8 = -1
Uint8 uint8 = 102
Const = 103
Float32 float32 = 104.5;
Float float = 105.5;
ConstFloat = 106.5;
Big float64 = 1e300;
Float32 float32 = 104.5
Float float = 105.5
ConstFloat = 106.5
Big float64 = 1e300
String = "abc";
Bool = true;
String = "abc"
Bool = true
)
var (
a1 = Int8 * 100; // ERROR "overflow"
a2 = Int8 * -1; // OK
a3 = Int8 * 1000; // ERROR "overflow"
a4 = Int8 * int8(1000); // ERROR "overflow"
a5 = int8(Int8 * 1000); // ERROR "overflow"
a6 = int8(Int8 * int8(1000)); // ERROR "overflow"
a7 = Int8 - 2*Int8 - 2*Int8; // ERROR "overflow"
a8 = Int8 * Const / 100; // ERROR "overflow"
a9 = Int8 * (Const / 100); // OK
a1 = Int8 * 100 // ERROR "overflow"
a2 = Int8 * -1 // OK
a3 = Int8 * 1000 // ERROR "overflow"
a4 = Int8 * int8(1000) // ERROR "overflow"
a5 = int8(Int8 * 1000) // ERROR "overflow"
a6 = int8(Int8 * int8(1000)) // ERROR "overflow"
a7 = Int8 - 2*Int8 - 2*Int8 // ERROR "overflow"
a8 = Int8 * Const / 100 // ERROR "overflow"
a9 = Int8 * (Const / 100) // OK
b1 = Uint8 * Uint8; // ERROR "overflow"
b2 = Uint8 * -1; // ERROR "overflow"
b3 = Uint8 - Uint8; // OK
b4 = Uint8 - Uint8 - Uint8; // ERROR "overflow"
b5 = uint8(^0); // ERROR "overflow"
b6 = ^uint8(0); // OK
b7 = uint8(Minus1); // ERROR "overflow"
b8 = uint8(int8(-1)); // ERROR "overflow"
b8a = uint8(-1); // ERROR "overflow"
b9 byte = (1<<10) >> 8; // OK
b10 byte = (1<<10); // ERROR "overflow"
b11 byte = (byte(1)<<10) >> 8; // ERROR "overflow"
b12 byte = 1000; // ERROR "overflow"
b13 byte = byte(1000); // ERROR "overflow"
b14 byte = byte(100) * byte(100); // ERROR "overflow"
b15 byte = byte(100) * 100; // ERROR "overflow"
b16 byte = byte(0) * 1000; // ERROR "overflow"
b16a byte = 0 * 1000; // OK
b17 byte = byte(0) * byte(1000); // ERROR "overflow"
b18 byte = Uint8/0; // ERROR "division by zero"
b1 = Uint8 * Uint8 // ERROR "overflow"
b2 = Uint8 * -1 // ERROR "overflow"
b3 = Uint8 - Uint8 // OK
b4 = Uint8 - Uint8 - Uint8 // ERROR "overflow"
b5 = uint8(^0) // ERROR "overflow"
b6 = ^uint8(0) // OK
b7 = uint8(Minus1) // ERROR "overflow"
b8 = uint8(int8(-1)) // ERROR "overflow"
b8a = uint8(-1) // ERROR "overflow"
b9 byte = (1<<10) >> 8 // OK
b10 byte = (1<<10) // ERROR "overflow"
b11 byte = (byte(1)<<10) >> 8 // ERROR "overflow"
b12 byte = 1000 // ERROR "overflow"
b13 byte = byte(1000) // ERROR "overflow"
b14 byte = byte(100) * byte(100) // ERROR "overflow"
b15 byte = byte(100) * 100 // ERROR "overflow"
b16 byte = byte(0) * 1000 // ERROR "overflow"
b16a byte = 0 * 1000 // OK
b17 byte = byte(0) * byte(1000) // ERROR "overflow"
b18 byte = Uint8/0 // ERROR "division by zero"
c1 float64 = Big;
c2 float64 = Big*Big; // ERROR "overflow"
c3 float64 = float64(Big)*Big; // ERROR "overflow"
c4 = Big*Big; // ERROR "overflow"
c5 = Big/0; // ERROR "division by zero"
c1 float64 = Big
c2 float64 = Big*Big // ERROR "overflow"
c3 float64 = float64(Big)*Big // ERROR "overflow"
c4 = Big*Big // ERROR "overflow"
c5 = Big/0 // ERROR "division by zero"
)
func f(int);
func f(int)
func main() {
f(Int8); // ERROR "convert|wrong type|cannot"
f(Minus1); // ERROR "convert|wrong type|cannot"
f(Uint8); // ERROR "convert|wrong type|cannot"
f(Const); // OK
f(Float32); // ERROR "convert|wrong type|cannot"
f(Float); // ERROR "convert|wrong type|cannot"
f(ConstFloat); // ERROR "truncate"
f(ConstFloat - 0.5); // OK
f(Big); // ERROR "convert|wrong type|cannot"
f(String); // ERROR "convert|wrong type|cannot|incompatible"
f(Bool); // ERROR "convert|wrong type|cannot|incompatible"
f(Int8) // ERROR "convert|wrong type|cannot"
f(Minus1) // ERROR "convert|wrong type|cannot"
f(Uint8) // ERROR "convert|wrong type|cannot"
f(Const) // OK
f(Float32) // ERROR "convert|wrong type|cannot"
f(Float) // ERROR "convert|wrong type|cannot"
f(ConstFloat) // ERROR "truncate"
f(ConstFloat - 0.5) // OK
f(Big) // ERROR "convert|wrong type|cannot"
f(String) // ERROR "convert|wrong type|cannot|incompatible"
f(Bool) // ERROR "convert|wrong type|cannot|incompatible"
}