[Ur] Confusing error message when using effectful javascript ffi function in onload handler

[Adam Chlipala]

I've pushed a small changeset with a new error message, inspired by your 
suggestion.

By the way, I expect the error message you reported originally was 
especially confusing because you must have configured an empty URI for 
your application's main page handler.  The "()" part of your message 
would have said, e.g., "(main)" if some nonempty name were configured 
for that handler.  My error message tweak is also aimed at explaining 
this part of the message better.

On 06/14/2013 01:27 PM, Daniel Patterson wrote:
> I still think the message is misleading. Because seeing that message
> (and not having source information to track down what was causing it),
> I removed _all_ links from the application - ie, there were no calls
> to "url", and no "<a>" tags. So it isn't that I'm using a link when I
> should be using a form - it is that a page that is accessible via a
> url is causing effects.
>
> If the error was more along the lines of "A handler accessible via GET
> requests could cause side effects - try accessing it only via forms,
> or removing it from the list of exported handlers", I would have had a
> lot easier time figuring out what was wrong. Of course, being able to
> specify what handler (ie, the handler at 101:0, or even better,
> sourceloc and name) would be even better.
>
> On Fri, Jun 14, 2013 at 12:53 PM, Adam Chlipala<adamc at csail.mit.edu>  wrote:
>    
>> I agree that it would be nice to get the proper source location information
>> propagated into that error message.  However, I don't think the message
>> itself is so misleading.  Presumably your 'onload' handler is on a page that
>> can be reached via a GET request to a link.  The HTTP standard or some
>> related advice specifies that GET requests should not have side effects, and
>> Ur/Web is detecting that your [Js.init] function might actually cause a side
>> effect upon loading of the link.  From a user's perspective, there is little
>> practical difference between "fetching URL causes a side effect" and
>> "fetching URL returns a page whose JavaScript immediately causes a side
>> effect."
>>
>> I hope the Ur/Web manual makes clear that 'benignEffectful' is for functions
>> with only page-local side effects, while 'effectful' is for functions that
>> may cause persistent side effects.  Does your [Js.init] really belong in the
>> second category?  If so, then the compiler seems to me to be making a
>> reasonable inference that you aren't following the rules about GET requests
>> not causing persistent side effects.
>>
>>
>> On 06/13/2013 08:14 PM, Daniel Patterson wrote:
>>      
>>> I have an ffi function declared like:
>>>
>>> ffi js
>>> jsFunc Js.init=init
>>> effectful Js.init
>>>
>>> I have it in my body onload handler as:
>>>
>>> Js.init ...
>>>
>>> And the error that it produces is:
>>>
>>> :0:0: (to 0:0) A link () could cause side effects; try implementing it
>>> with a form instead
>>> make: *** [app] Error 1
>>>
>>>
>>> Which is confusing because:
>>>
>>> a. it has no source location information, so I have no idea where the
>>> error is occuring and
>>>
>>> b. it is not caused by a link.
>>>
>>> The problem goes away if I either don't say that it is effectful, or
>>> say it is benignEffectful, but I'm not sure why this is - because it
>>> seems like onload handlers should be able to cause effects...
>>>