Re: Questions and remarks on draft-ietf-sieve-include-01.txt

[Aaron Stone <aaron@xxxxxxxxxxxxxx>]

On Tue, 05 May 2009 16:47:11 -0400, Jeffrey Hutzelman <jhutz@xxxxxxx>
wrote:
> --On Tuesday, May 05, 2009 01:14:09 PM -0700 Aaron Stone 
> <aaron@xxxxxxxxxxxxxx> wrote:
> 
>>
>> On Tue, 05 May 2009 15:25:17 -0400, Jeffrey Hutzelman <jhutz@xxxxxxx>
>> wrote:
>>> --On Tuesday, May 05, 2009 11:07:59 AM +0200 Stephan Bosch
>>> <stephan@xxxxxxxxxxxx> wrote:
>>>
>>>> Also, you completely lifted the global command's placement
>>>> requirements.
>>>> I am not fundamentally opposed to this, but this can make the
inclusion
>>>> of a global variable conditional, i.e. inside an if statement. Until I
>>>> actually implement this I won't really know for sure whether there
will
>>>> be implementation issues.
>>>>
>>>> And if the possibility for a conditional global command is desired, I
>>>> would expect an explicit mention of this possibility somewhere in the
>>>> command explanation. Now it could come to a surprise to the more naive
>>>> implementer. This also implies that a global variable is not included
>>>> into the local scope until after the first global command that
>>>> specifies
>>>> it.
>>>
>>>
>>> I would expect that a global command has effect only from the command
to
>>> the end of the enclosing scope; as a result, a global inside a
>> conditional
>>> has effect only on the conditional code, and never on something
>>> occurring
>>
>>> after the conditional, regardless of whether the conditional is taken.
>>
>> Would require variable scoping rules that follow the curly braces, but
>> we're doing it PHP / Python style in Sieve.
>>
>>> Alternately, a global command could be defined to apply to the entire
>> file
>>> regardless of where it occurs.
>>
>> That presumes pre-parsing the script, which I don't want to sneakily
make
>> a requirement of interpreters.
>>
>> I went with this approach: the variable is local to the immediate script
>> until it is called global at some point later in the script. I flipped a
>> coin, and decided the local value would then overwrite the global value.
>> I
>> could as easily see it go the other way, though. I can't think of any
>> particularly good examples from other languages that might apply here.
>> Feedback appreciated!
> 
> Hrm.  OK, let's make it simple...
> 
> The global command is legal at any point in the script.  However, it must

> appear _and be executed_ before any use of the named variable(s). 
> Otherwise, the results are undefined.

Or an error is thrown at runtime. I'm not a fan of this idea, but at least
it's definitive.

> This means the question of what happens if you use a local variable and 
> then call it global is up to the implementation.  So is the question of 
> what happens if the global command appears in a conditional and that
branch
> is not executed, and then the variable is used.
> 
> I think it gets us all the same implementation simplicity as requiring it

> to be "first", without painting ourselves into the corner of someday
having
> multiple things that want to be "first".

Yuck, I don't like it when a language doesn't specify something like this,
especially knowing it is probably going to be implemented differently right
from the get-go.

Aaron