I can make your example, using the Zn variants, work with the following change:

StdioStream>>#atEnd
  ^ peekBuffer isNil or: [ (peekBuffer := self next) isNil ]

Which is a literal implementation of your statement that you can only know that 
you are atEnd by reading (and thus waiting/blocking) and checking for nil, 
which seems logical to me, given the fact that you *are* waiting for user input.

BTW, at least on macOS you have to enter ctrl-D (^D) on a separate line, I am 
not sure how relevant that is, but that is probably another argument that stdin 
is special (being line-buffered by the OS, EOF needing to be on a separate 
line).

And FWIW, I have been writing networking code in Pharo for years, and I have 
never had issues with unclear semantics of these primitives (#atEnd, #next, 
#peek) on network streams, either the classic SocketStream or the Zdc* streams 
(TLS or not). That is why I think we have to be careful.

That being said, it is important to continue this discussion, I find it very 
interesting. I am trying to write some test code using stdin myself, to better 
understand the topic.

> On 11 Apr 2018, at 16:06, Alistair Grant <akgrant0...@gmail.com> wrote:
> 
> On 11 April 2018 at 15:11, Sven Van Caekenberghe <s...@stfx.eu> wrote:
>> 
>> 
>>> On 11 Apr 2018, at 11:12, Sven Van Caekenberghe <s...@stfx.eu> wrote:
>>> 
>>> How does one modify #atEnd to block ? I suppose you are talking about 
>>> StdioStream>>#atEnd ?
>>> 
>>> ^ self peek isNil
>>> 
>>> ?
>> 
>> Still the same question, how do you implement a blocking #atEnd for stdin ?
>> 
>> I have seen your stdio.cs which is indeed needed as the current 
>> StdioStream>>#atEnd is bogus for sure.
>> 
>> But that is still a non-blocking one, right ?
>> 
>> Since there is a peekBuffer in StdioStream, why can't that be used ?
> 
> I think you've created a chicken-and-egg problem with this question,
> but ignoring that for now:
> 
> 
> StdioStream>>peek
> "Answer what would be returned if the message next were sent to the
> receiver. If the receiver is at the end, answer nil.  "
> 
>    self atEnd ifTrue: [^ nil ].
> 
>    peekBuffer ifNotNil: [ ^ peekBuffer ].
> 
>    ^ peekBuffer := self next.
> 
> 
> 
> So when we first start the program, i.e. the user hasn't entered any
> input yet, and #peek is called:
> 
> 1. #atEnd returns false because Ctrl-D (or similar) hasn't been
> entered (assuming it is non-blocking).
> 2. peekBuffer is nil because we haven't previously called #peek.
> 3. The system now blocks on "self next".
> 
> 
> Just a reminder: for terminal input the end-of-file isn't reached
> until the user explicitly enters the end of file key (Ctrl-D).
> 
> So, if there is no buffered input (either none has been entered yet,
> or all input has been consumed)
> 
> #atEnd (after the patch) calls #primAtEnd:.
> 
> At the moment, #primAtEnd: ends up calling the libc function feof(),
> which is non-blocking and answers the end-of-file flag for the FILE*.
> Since the user hasn't entered Ctrl-D, that's false.
> 
> If we want to control iteration over the stream and ensure that we
> don't need to do a "stream next == nil" check, then #primAtEnd: is
> going to have to peek for the next character, and that means waiting
> for the user to enter that character.
> 
> In c that is typically done using:
> 
> atEnd = ungetc(fgetc(fp), fp);
> 
> and fgetc() will block until the user enters something.
> 
>> I have run your example testAtEnd.st now, and it works/fails as advertised.
> 
> :-)
> 
> 
> Cheers,
> Alistair


Reply via email to