Re: strscpy

Le 29/05/2020 à 20:36, matthew green a écrit :
>>>> I would want to keep the difference minimal between the two. I think
>>>> that the
>>>> version that returns -1 is nice and addresses the concern. What do you
>>>> think?
>>>
>>> Personally I have got no strong opinion. I would change "size > INT_MAX"
>>> to "size > SSIZE_MAX".
>>>
>>> Returning -1 instead of -EINVAL and -E2BIG is good, as there few users
>>> who check return value at all, and if so, they check whether it is less
>>> than 0.
>>
>> I have updated my patch:
>>
>> 	https://m00nbsd.net/garbage/libkern/strscpy.diff
>>
>> We simply return -1. Will commit soon unless there is valid objection.
> 
> i strongly object.
> 
> please don't commit more dual-return value APIs.
> 
> kre's point about it is 100% valid.  they're extremely
> dangerous.

I have still not seen any valid technical reason why they are "dangerous".
Now they're suddenly "extremely dangerous", but you fail to explain how.

> i spent a couple of months tracking down a
> mismatch between linux and netbsd errno values that
> happens because linux APIs regularly return "-errno",
> and we (bsd generally, not just netbsd) have avoided
> that idiom in most places by design.

You did notice that I am _not_ returning a negative errno, right?

This is an anecdote about Linux-to-BSD mishaps, caused by Linux using
negative errnos. I could easily quote you a hundred BSD-to-BSD mishaps
I've seen in imported subsystems and drivers.

Maxime