Proposal for SentinelInputRange

[Dmitry Olshansky]

28-Feb-2013 20:33, Walter Bright пишет:
> On 2/28/2013 7:38 AM, Dmitry Olshansky wrote:
>> 28-Feb-2013 14:44, Walter Bright пишет:
>>> On 2/28/2013 12:25 AM, Jacob Carlborg wrote:
>>>> On 2013-02-28 08:23, Walter Bright wrote:
>>>>
>>>>> There is always a terminating 0, even if the file ends in 0x1a.
>>>>>
>>>>> (The 0x1A comes from some old editors that end a file with a
>>>>> control Z.)
>>>>
>>>> http://dlang.org/lex.html, "End of File" shows this:
>>>>
>>>> EndOfFile:
>>>>      physical end of the file
>>>>      \u0000
>>>>      \u001A
>>>>
>>>> Doesn't that mean that the input is empty if the file is empty, \u0000
>>>> or \u001A
>>>> is encountered?
>>>>
>>>
>>
>>
>>> Please follow the source code from file to input to the lexer.
>>
>> Source as spec is no good. Either change the spec or  admit that
>> having a tuple
>> of sentinels as manifest constant is fine. In any case constant tuples
>> are
>> easily unrolled into case statements.
>
> The spec is correct, so is the code, and tuple sentinels are entirely
> unnecessary.
>

line 300:
     case 0x1A:          // ^Z means end of file
     case 0:
                         break;

On the lines you noted it claimed that that 0x1a is outdate. Along with 
the fact that you allocate filesize+2 and fill the last 2 bytes with zeros.

In any case I see 0 and 0x1a as 2 values that act like sentinels i.e. a 
tuple. And this is what spec says - any one of them is a sentinel. 
Correct me if I'm wrong.

-- 
Dmitry Olshansky