www.digitalmars.com         C & C++   DMDScript  

digitalmars.D - How to debug broken inference?

reply Steven Schveighoffer <schveiguy gmail.com> writes:
I have a dilemma here. I have a non-trivial template function which is 
inside a template struct, and this template is inferred to be  system. 
I'm testing using a  safe unittest.

I put  safe on the template to see which calls are actually system, and 
now it compiles.

This is not a solution, because some template parameters might cause the 
function to truly be  system, and then it just won't compile at all.

I'm struggling to try and figure out how to properly make this function. 
Not only am I not sure what to test, but I will have to duplicate the 
function for both  safe and  system versions, or use a mixin.

But this is going to be really painful. I very much depend on the 
compiler inferring safety for templates, and it's what we recommend to 
people who want to build their code for all purposes.

Does anyone have any recommendations of how to approach this? Even if I 
wanted to report a bug, I have no idea why this is happening in this one 
spot. Why can't the compiler *try*  safe and if it doesn't work, just 
decay to  system?

-Steve
May 01
parent reply Steven Schveighoffer <schveiguy gmail.com> writes:
On 5/1/20 4:40 PM, Steven Schveighoffer wrote:
 
 Does anyone have any recommendations of how to approach this? Even if I 
 wanted to report a bug, I have no idea why this is happening in this one 
 spot. Why can't the compiler *try*  safe and if it doesn't work, just 
 decay to  system?
I figured out the issue. It was something very stupid, but probably a bug in the compiler. In my efforts to avoid bounds checking, I did this: char[4] buf; auto slice = buf.ptr[idx .. buf.length]; This compiles if it's tagged safe, but will not infer safety. But I don't need to do this trick anyway, it's a static array! So I just fixed that part of the code. But I'd say it's a bug in the compiler, it shouldn't accept that as safe. Indeed if I just put that in a function, it fails to compile as safe. I'll need to dustmite it. For reference, I used a manual dustmite-like process to narrow down this function to the part that fails. -Steve
May 01
parent reply welkam <wwwelkam gmail.com> writes:
On Friday, 1 May 2020 at 21:04:34 UTC, Steven Schveighoffer wrote:
 char[4] buf;

 auto slice = buf.ptr[idx .. buf.length];
onlineapp.d(3): Error: buf.ptr cannot be used in safe code, use &buf[0] instead
May 01
parent Steven Schveighoffer <schveiguy gmail.com> writes:
On 5/1/20 6:46 PM, welkam wrote:
 On Friday, 1 May 2020 at 21:04:34 UTC, Steven Schveighoffer wrote:
 char[4] buf;

 auto slice = buf.ptr[idx .. buf.length];
onlineapp.d(3): Error: buf.ptr cannot be used in safe code, use &buf[0] instead
Yes, I know. That doesn't happen in my full use case. Which is why I said I need to dustmite it. But what I did find is that once I changed that line, then it infers safe. Essentially, there's a bug in the compiler that has to do with my whole project. -Steve
May 01