[racket-dev] Should BSL signal an error on (require ...) in an unsaved buffer?

On Jun 27, 2011, at 5:47 PM, Robby Findler wrote:

> On Tue, Jun 28, 2011 at 8:41 AM, John Clements
> <clements at brinckerhoff.org> wrote:
>> 
>> On Jun 27, 2011, at 5:28 PM, Robby Findler wrote:
>> 
>>> On Tue, Jun 28, 2011 at 8:19 AM, John Clements
>>> <clements at brinckerhoff.org> wrote:
>>>> 
>>>> On Jun 27, 2011, at 5:10 PM, Robby Findler wrote:
>>>> 
>>>>> On Tue, Jun 28, 2011 at 7:52 AM, John Clements
>>>>> <clements at brinckerhoff.org> wrote:
>>>>>> 
>>>>>> On Jun 27, 2011, at 4:45 PM, Robby Findler wrote:
>>>>>> 
>>>>>>> We're talking about relative requires only, right?
>>>>>>> 
>>>>>>> How are you proposing to signal the error?
>>>>>> 
>>>>>> My guess about how this works--correct me if I'm wrong--is that for unsaved buffers, DrR sets a parameter such that the expanded code has the current directory as (uh, part of?) the syntax-source of the expanded source.  I'm guessing that I'm not right, or it would be as simple as disabling this for programs written in BSL et al.
>>>>> 
>>>>> DrRacket does indeed set current-directory and could instead signal an
>>>>> error instead of setting it. That would mean no unsaved file would
>>>>> run, however. Is that really desired?
>>>> 
>>>> Would it not be possible to set the syntax-source in such a way that relative requires would fail? We may be out of DrR and into just plain R here.
>>> 
>>> Yes, we have to be, I'm pretty sure.
>>> 
>>>> Indeed, I would be happy if buffers without a recorded save location couldn't make relative requires in *any* language level, though I realize that such a change would be more likely to have repercussions elsewhere.
>>> 
>>> I think that would probably require a new parameter that controls
>>> where relative requires are relative to being exported and allowing it
>>> to have a value that says "no relative requires allowed".
>>> (current-load-relative-directory is maybe already this, without the
>>> extra functionality, maybe?)
>>> 
>>>> I suppose there could be situations where users dynamically create syntax objects and set the current directory rather than setting the syntax-source of the objects, though that doesn't sound like the most robust way to write the code.
>>> 
>>> I don't understand this comment.
>>> 
>>> (I believe that the source field of a syntax objects is not involved
>>> in resolving relative requires.)
>> 
>> Ah! I understand better now.
>> 
>> Okay, out of the realm of how-things-should-be and into the realm of how-things-are; you write that DrR sets 'current-directory'.  Is there some reason to set 'current-directory' rather than 'current-load-relative-directory'? It looks to me like setting 'current-load-relative-directory' fixes the bug I'm tackling, but I want the stepper to accurately simulate DrR's behavior here.
> 
> Oh, I see that current-load-relative-directory can already be #f. I
> think that situation corresponds to "I'm not requiring anything", tho
> and, according to the docs, that parameter is set by the require
> machinery (not by drracket altho I didn't grep the sources to be
> sure). So again, I think we're stuck if all we change is drracket.
> 
> (I'm not sure, but I think that this is probably going to be more
> trouble than it is worth.)

Okey dokey. I've verified that setting "current-directory" has the desired effect, so I'm just going to set it to the result of the tab's "get-directory" before calling the expander to pull the exprs out of the defns window.  Let me know if this is obviously wrong.

Thanks!

John

(lost dev on intermediate response, sorry)


-------------- next part --------------
A non-text attachment was scrubbed...
Name: smime.p7s
Type: application/pkcs7-signature
Size: 4624 bytes
Desc: not available
URL: <http://lists.racket-lang.org/dev/archive/attachments/20110627/4190cb5f/attachment.p7s>