llvm dev - Oct 2014 - [LLVMdev] lld coding style

On Oct 6, 2014, at 6:03 PM, Nick Kledzik <kledzik at apple.com>
wrote:
>> I'd like to hear the reason. :)
> 
> Up until now the thread has been about “formatting”.  You suggested
renaming every variable in the project!
If that's what it takes. 
> On Oct 6, 2014, at 5:37 PM, Chris Lattner <clattner at apple.com>
wrote:
>> Right.  Specifically, why is it for the “greater good” to perpetuate a
schism between lld and the rest of llvm?
>> 
> 
> Well to turn this around, why is llvm maintaining a schism with the rest of
the C++ world?  That is, lld (and probably lldb) is more like all the other C++
conventions in the world than llvm is.  If llvm had conventions like every other
convention, it would be easier for outside developers to contribute to llvm and
improve the "greater good".

I can see where you're coming from, and as I hope you know, I deeply respect
you and your work.  You're right that what we're talking about has a
profound impact on 100% of the lld project.  That said, we must keep in
perspective lld vs the rest of the llvm eco system.

By my count, lld is about 48K LOC (including include/lib/tools, but not tests). 
When you compare this to LLVM (about ~1.1M LOC, or clang's ~900KLOC) what
we're really discussing is whether it is better to favor the core
preferences of the lld authors vs the opportunity cost of developers jumping
from llvm/clang and being effective in the lld code base.  Given that we're
talking about a 40x size difference, along with greater unity in the llvm
community, this seems like a simple decision to me.

I see your point and can concede that not everyone in the C++ community follows
the same approach as the LLVM coding standards, but a) the greater C++ community
is far from unified on the lld style, and b) a random C++ developer is far less
likely to make the leap to work on LLD than a random LLVM developer is.

I really think that both LLD and LLDB should move to the LLVM coding standards. 
If it makes you feel better, doing so would be a bigger change for the LLDB
folks than the LLD ones.

-Chris

-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20141007/0f2ad6dd/attachment.html>

On 8 October 2014 05:25, Chris Lattner <clattner at apple.com>
wrote:
> > Up until now the thread has been about “formatting”.  You suggested
renaming
> > every variable in the project!
>
> If that's what it takes.
If we're still talking about applying it to the HEAD, not the whole
history, I think it's feasible.

A blunt refactoring job would have to be one (or very few) commit(s)
and a lot of testing put on to make sure the change didn't introduce
new bugs.

A series of blunt changes would make things worse (by not knowing
where the change started/stopped). So the only alternative would be to
make small changes on all future patches, which would contradict our
other rule that formatting should go on their own patches.

> a random C++
> developer is far less likely to make the leap to work on LLD than a random
> LLVM developer is.
That's a good point. And we want to increase the number of LLVM
developers helping LLD or it'll never become an integral part of the
LLVM umbrella (like Clang did).

cheers,
--renato

> On Oct 8, 2014, at 1:55 AM, Renato Golin <renato.golin at linaro.org>
wrote:
> 
> On 8 October 2014 05:25, Chris Lattner <clattner at apple.com> wrote:
>>> Up until now the thread has been about “formatting”.  You suggested
renaming
>>> every variable in the project!
>> 
>> If that's what it takes.
> 
> If we're still talking about applying it to the HEAD, not the whole
> history, I think it's feasible.
Yep, to be clear, I mean one big change to head, not trying to rewrite history.

-Chris