On Mar 2, 2013, at 5:40 AM, Rafael Espíndola <[email protected]> wrote:

>> I'm concerned that this kind of development approach doesn't actually
>> make code more robust if we're designing the program to continue down
>> other untested code paths (OK - so it doesn't infloop here, it does
>> something else unexpected later). Are clients of this API intending to
>> account for invalid FileIDs being returned, for example.
> 
> Having worked on a codebase that went down this path a long time ago I
> must second this concern. Changes like this make bug easier to ignore,
> and the result is just that, they are ignored.
> 
> In other codebases what I have seen is a assert-like macro that causes
> crashes on non debug builds too. That makes bugs easier to reproduce
> and much harder to ignore. Could we do something like that in here?

libclang from the beginning has chosen to avoid crashing the process, see 
llvm::CrashRecoveryContext.

> 
> Cheers,
> Rafael


_______________________________________________
cfe-commits mailing list
[email protected]
http://lists.cs.uiuc.edu/mailman/listinfo/cfe-commits

Reply via email to