@nogc

[Chris Cain via Digitalmars-d]

On Friday, 11 July 2014 at 07:45:24 UTC, bearophile wrote:
> Joseph Rushton Wakeling:
>
>> I've habitually used 'format' to prepare assert messages with 
>> variable content, which has a similar problem -- IIRC it 
>> conflicts with nothrow and I guess it too would conflict with 
>> @nogc.
>
> A solution is to add to Phobos a way to format to an output 
> range, where the range you use a stack-allocated array of chars 
> large enough to contain your formatted message. So on the whole 
> it can be @nogc.
>
> Bye,
> bearophile

Although I think that is certainly something that would be 
helpful/necessary, it wouldn't solve the problem of printing out 
assert messages/exception messages.

Basically, the only way I could think to do it is to make a 
special `NoGCException` that has a `detailMsg` function (taking a 
`sink`, of course) and your subclass of `NoGCException` would 
store all the info necessary to print out the error at the catch 
site and you'd just overload `detailMsg` to print it out properly.

https://gist.github.com/Zshazz/47ed52c3246e5348062a

That's an example. I store info at line 84 as a tuple in my 
`SpecialNoGCException`, and set the storage when I'm about to 
throw at line 202. My apologies for the poor code (I made that 
stuff as an experiment). It works, but pre-allocating the initial 
exception is still problematic, IMO. I'd really like a way of 
specifying that an exception was malloc'd and it's the catcher's 
job to free it. Obviously, I could "just do it" but it seems 
pretty unsafe and error-prone. *shrug*