Re: [boost] GSOC 2015 : Project on Concurrent Hash Tables

Subject: Re: [boost] GSOC 2015 : Project on Concurrent Hash Tables
From: Niall Douglas (s_sourceforge_at_[hidden])
Date: 2015-02-14 09:15:09

On 14 Feb 2015 at 17:04, Amarnath V A wrote:

> I have a few questions regarding the code base before I jump in try to
> complete the test.
> 
> 1. As I have experience only on C++98, I am not sure what is the purpose
> of the move constructors. Reading up on internet gave me fair idea about
> move constructors in general. In our context, is there anything specific
> to be taken care? Could you provide few ideas of what is the expected
> result of the move? The important question I have is should the move
> rehash all the hashes? (Is that why you were mentioning about rehash()?) 

Move construction is (usually) destructive to the original, so no 
rehashing should occur. In fact, no element copying should occur 
either. When you examine the assembler generated by your 
implementation, you should expect to see mostly pointer copies, and 
the total move constructor implementation ought to not exceed maybe 
40-50 assembler instructions.

Note that I will be examining the assembler output by candidate 
submissions as part of my assessment.

However, there is more still. concurrent_unordered_map must also be 
thread safe during both copying and moving. This is where it gets 
tricky, and this is why it is a good test of whether a candidate has 
what it takes to do a successful GSoC.

Finally, both copy and move construction need to be exception safe 
i.e. if during a copy or move an exception is thrown, the map needs 
to be left in a known good state. Achieving this perfectly correctly 
I don't expect of a GSoC candidate, though if one were to achieve it 
they would be noticed.

> 2. Similarly, does the copy constructor need to just perform a deep copy
> of the hash table? Or should it also perform a rehashing? 

A copy constructor copies the source exactly. Yes, as per STL 
convention it is a deep copy. I would not expect to see a rehash 
during a copy.

> 3. Could you also point me to any white paper or publication that gave
> you how to go about with the implementation of concurrent_unordered_map? 

It is my own creation. The concepts used are very conventional, 
indeed Chandler Carruth gave a talk at CppCon 2014 about the exact 
same techniques. I personally believe he was inspired to that topic 
from conversations with me at C++ Now the preceding May :), but I 
guess we'll never know. His talk can be watched at 
https://www.youtube.com/watch?v=fHNmRkzxHWs.

> 4. Being a Linux user, I am using Ubuntu as my dev box. I have currently
> build the source
> code using gcc v4.8.2 and clang v3.4. Is this fine?

It is fine. Though for the GSoC you would be expected to keep 
compatibility with VS2013.

> Sorry if my mail ends up as a cross post. I am new to this list. And
> thanks for bearing with my (maybe silly) questions. 

Asking good questions is the mark of a good engineer. I myself am 
just about to go ask stackoverflow how you programmatically delete a 
file using just its inode on Linux and FreeBSD, because I can't find 
an answer.

Niall

--- 
Boost C++ Libraries Google Summer of Code 2015 admin
https://svn.boost.org/trac/boost/wiki/SoC2015