But I don't think it's true that it's all plain-sailing today.

You can compile one TU with fast-math and one without. The header code they
consume is token-identical.

There's no ODR violation but you can get differences depending on which
version the linker chooses. And these differences need not be small. If the
result of a floating-point calculation is used to determine if a grid cell
is open or closed, tiny numerical differences can blow up into boolean
yes/no differences.

Isn't this something the community already lives with? Contracts may
increase the surface area for such things but I don't think it's novel.

O.

On Mon, 20 Oct 2025 at 16:19, John Spicer <jhspicer_at_[hidden]> wrote:

> By “do everything right” I meant “you don’t have an ODR violation”.
>
> Not that your program is perfect.
>
> I agree that in perfect programs you don’t need contracts.
>
> But for every actual program, people want contracts do what they can
> reasonably expect that they will do.
>
> John.
>
> On Oct 20, 2025, at 11:16 AM, Oliver Rosten via SG21 <
> sg21_at_[hidden]> wrote:
>
> Hi John,
>
> Contracts add the new issue that you can do everything right and still get
>> surprising behavior
>
>
> If you've done everything right then contracts are completely redundant in
> your program.
>
> O.
>
>
> On Mon, 20 Oct 2025 at 16:14, John Spicer <jhspicer_at_[hidden]> wrote:
>
>> In general, if you do things correctly, you don’t need to understand ODR
>> violations.
>>
>> If you do things incorrectly, it might be hard to figure out. But having
>> header files mean different things in difference places is a long-standing
>> issue that goes back to C and has always been hard to figure out.
>>
>> Contracts add the new issue that you can do everything right and still
>> get surprising behavior and it is even harder to figure out because your
>> tools won’t help you.
>>
>> John.
>>
>> On Oct 20, 2025, at 9:56 AM, Oliver Rosten <oliver.rosten_at_[hidden]>
>> wrote:
>>
>> PS Most people I teach have no prior idea what an ODR violation is.
>>
>> On Mon, 20 Oct 2025 at 14:55, Oliver Rosten <oliver.rosten_at_[hidden]>
>> wrote:
>>
>>> That's a proof of existence but not of wide-spread usage.
>>>
>>> I am honestly ignorant here. As far as I know the C++ ecosystem as a
>>> whole is not making rigorous use of things like this. But I may be wrong...
>>>
>>> On Mon, 20 Oct 2025 at 14:53, Ville Voutilainen <
>>> ville.voutilainen_at_[hidden]> wrote:
>>>
>>>> On Mon, 20 Oct 2025 at 16:46, Oliver Rosten
>>>> <oliver.rosten_at_[hidden]> wrote:
>>>> >
>>>> > Hi John,
>>>> >
>>>> > I'm not convinced by this:
>>>> >
>>>> >> No, it is not a pre-existing problem.
>>>> >> Other than contracts, if you end up with different function
>>>> definitions it is an ODR violation and your program is IFNDR and can be
>>>> rejected by your tools.
>>>> >
>>>> >
>>>> > There's a difference between "can be in principle" and "is in general
>>>> practice". Is it not the case that, in most instances, for all practical
>>>> purposes there is no difference between an ODR violation that's IFNDR and
>>>> the contracts mixed-mode: you get what the linker gives you?
>>>>
>>>> See, for example,
>>>> https://devblogs.microsoft.com/oldnewthing/20160803-00/?p=94015
>>>> See also https://maskray.me/blog/2022-11-13-odr-violation-detection
>>>>
>>>
>> _______________________________________________
> SG21 mailing list
> SG21_at_[hidden]
> Subscription: https://lists.isocpp.org/mailman/listinfo.cgi/sg21
> Link to this post: http://lists.isocpp.org/sg21/2025/10/11470.php
>
>
>