Among the CppCon 2014 slide packs there is this nice one:

"Types Don't Know #", by Howard Hinnant:

https://github.com/CppCon/CppCon2014/tree/master/Presentations/Types%20Don%27t%20Know%20%23%20-%20Howard%20Hinnant%20-%20CppCon%202014

It shows a nice idea to perform transitive hashing in a flexible and efficient way. Perhaps the idea can be used in D too. It suggests the introduction of a hashAppend standard method.

Bye,
bearophile

On Tuesday, 16 September 2014 at 14:53:52 UTC, bearophile wrote:
> Among the CppCon 2014 slide packs there is this nice one:
>
> "Types Don't Know #", by Howard Hinnant:
>
> https://github.com/CppCon/CppCon2014/tree/master/Presentations/Types%20Don%27t%20Know%20%23%20-%20Howard%20Hinnant%20-%20CppCon%202014
>
> It shows a nice idea to perform transitive hashing in a flexible and efficient way. Perhaps the idea can be used in D too. It suggests the introduction of a hashAppend standard method.
>
> Bye,
> bearophile

Funny thing is that as soon as you have to combine hashes in C++ atm you realize that current way is broken and the way HH proposes is kindof obvious correct design.

On Tue, Sep 16, 2014 at 02:53:51PM +0000, bearophile via Digitalmars-d wrote:
> Among the CppCon 2014 slide packs there is this nice one:
> 
> "Types Don't Know #", by Howard Hinnant:
> 
> https://github.com/CppCon/CppCon2014/tree/master/Presentations/Types%20Don%27t%20Know%20%23%20-%20Howard%20Hinnant%20-%20CppCon%202014
> 
> It shows a nice idea to perform transitive hashing in a flexible and efficient way. Perhaps the idea can be used in D too. It suggests the introduction of a hashAppend standard method.
[...]

This is all the more reason to make rt.util.hash.hashOf available to user code. It already supports chaining, which would be the equivalent of hashAppend.


T

-- 
The computer is only a tool. Unfortunately, so is the user. -- Armaphine, K5

On Tuesday, 16 September 2014 at 14:53:52 UTC, bearophile wrote:
> Among the CppCon 2014 slide packs there is this nice one:
>
> "Types Don't Know #", by Howard Hinnant:
>
> https://github.com/CppCon/CppCon2014/tree/master/Presentations/Types%20Don%27t%20Know%20%23%20-%20Howard%20Hinnant%20-%20CppCon%202014
>
> It shows a nice idea to perform transitive hashing in a flexible and efficient way. Perhaps the idea can be used in D too. It suggests the introduction of a hashAppend standard method.

Yes, I've never understood why std::hash was designed the way it
is.  Or why allowing the user to supply a seed value seems to be
unusual for publicly available hash routines (which is basically
what Howard has proposed).  rt.util.hash.hashOf accepts a seed
value, and any hash routine we made publicly callable should as
well.  Not to do so is really pretty useless.

I guess this also has implications for composite types in D.  If
I have a class that references other classes, I might want to do
the same basic thing.  In which case, calling Object.toHash isn't
sufficient as it fails in the same manner.  I suppose we really
need to extend toHash to accept an optional seed value as well.