[cfe-dev] [LLVMdev] [RFC] TileGX, a new backend for Tilera's many core processor

[Reed Kotler]

On 03/13/2013 05:24 AM, Joerg Sonnenberger wrote:
> On Wed, Mar 13, 2013 at 05:07:39AM -0700, Reed Kotler wrote:
>> It's possible to split up the patch. I think it took me 40 patches
>> to push the mips16 port into LLVM.
> Extending an existing port is a bit different. Frankly, I don't see how
> pushing the target info and the codegen in separate parts makes anything
> simpler. It's not like the amount of code to review shrinks that way.
>
> Joerg
The bottom line is that I don't think anybody is going to take giant 
patches from you, even if it's a new port.
You already made you counter argument and nobody is accepting it. I 
tried to make similar counter arguments when I did the mips 16 port and 
got nowhere.

I tried to explain the reasons behind this policy as I understand them.

The quickest way to get your port pushed is to start to submit small 
patches and test cases for them.

Sometimes a bigger patch will be accepted if it's repetitious.

You can always try and push a bigger patch and see if somone will 
approve and commit it.
You tried this time and everyone is saying that they won't approve and 
commit it.

It's very tedious pushing changes at times. I spend 5 times as much time 
writing the test cases and checking everything as I did doing the patch 
sometimes. You have to schedule for that or you will become frustrated 
with this process and hopefully your management will realize that it's 
one thing to push to an internal repository and another thing to push to 
open source.


Reed