steakhal wrote:

I am getting some push back internally against this change.

The problem is that some projects are close-to-the-metal (firmware, boot, 
hardware stuff) and there actually the only way to communicate with the 
hardware is via fixed addresses and a bunch of volatile pointers.

Of course, this checker wasn't really designed for such code bases - if 
anything, for the opposite.
However, projects in such a domain are also likely to have a security aware 
mindset, which fundamentally opposes them to just disable some checkers - 
especially if they are in the `core` package. I find this argument pretty solid.

So the question is, why should we do about this?
 - Clarify the docs that this checker is not intended to be used in firmware 
stuff?
 - I want to especially avoid the users picking up the only suppression to this 
checker, using the `address_space(NNN)` attribute. That attribute is not 
documented, also has consequences on the binary so I definitely don't want to 
imply using that for suppression. This is kinda what users think right now, 
sadly.
 - Come up with some new way of suppressing this checker?
 - We could potentially move this to the `optin` package.
 
 @NagyDonat @balazske WDYT?

https://github.com/llvm/llvm-project/pull/132404
_______________________________________________
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

Reply via email to