llvm dev - Jan 2017 - [cfe-dev] Modernizing LLVM Coding Style Guide and enforcing Clang-tidy

On Mon, Jan 9, 2017 at 7:25 AM, Piotr Padlewski via llvm-dev <
llvm-dev at lists.llvm.org> wrote:
> 2017-01-09 16:15 GMT+01:00 Renato Golin <renato.golin at linaro.org>:
>
>> On 9 January 2017 at 14:17, Piotr Padlewski via cfe-dev
>> <cfe-dev at lists.llvm.org> wrote:
>> > - prefer "using' instead of "typedef"
>> > - use default member initialization
>> > - use default, override, delete
>> > - skip "virtual" with override
>>
>> I thought we had all of those already...
>>
>> Nope, some people use it, but I still see a lot of new code with
typedefs.
> I would like to have it written in style guide so it will be easier to
> convince to change in review.
>
The last two are enforced by compiler warnings now. The second is hard
because of bitfields.

I object to the first. If you need a new type name, use a typedef. It's
time honored and everyone, including C programmers, will know what you're
doing. I don't understand why people push the new thing just for the sake
of new-ness.
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20170109/eb3d1ebc/attachment.html>

> On Jan 9, 2017, at 10:20 AM, Reid Kleckner via llvm-dev <llvm-dev at
lists.llvm.org> wrote:
> 
> On Mon, Jan 9, 2017 at 7:25 AM, Piotr Padlewski via llvm-dev <llvm-dev
at lists.llvm.org <mailto:llvm-dev at lists.llvm.org>> wrote:
> 2017-01-09 16:15 GMT+01:00 Renato Golin <renato.golin at linaro.org
<mailto:renato.golin at linaro.org>>:
> On 9 January 2017 at 14:17, Piotr Padlewski via cfe-dev
> <cfe-dev at lists.llvm.org <mailto:cfe-dev at lists.llvm.org>>
wrote:
> > - prefer "using' instead of "typedef"
> > - use default member initialization
> > - use default, override, delete
> > - skip "virtual" with override
> 
> I thought we had all of those already...
> 
> Nope, some people use it, but I still see a lot of new code with typedefs.
> I would like to have it written in style guide so it will be easier to
convince to change in review.
> 
> The last two are enforced by compiler warnings now. The second is hard
because of bitfields.
> 
> I object to the first. If you need a new type name, use a typedef. It's
time honored and everyone, including C programmers, will know what you're
doing. I don't understand why people push the new thing just for the sake of
new-ness.
`using` handles strictly more cases than  `typedef`, in particular partial
specialization of templates. So because we’ll end up with `using` anyway,
uniformity can be valuable. So that could be a motivation: since `using` is
needed anyway, might be better to just use it always (I’m not saying it is a
“strong” motivation though, just some piece of rational).
(I also find that typedef of function pointers in particular are terrible, and
`using` is much better for these, but that can be a matter of taste).

—
Mehdi
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20170109/ea8f21a8/attachment.html>

On 9 January 2017 at 18:20, Reid Kleckner <rnk at google.com>
wrote:
> I object to the first. If you need a new type name, use a typedef. It's
time
> honored and everyone, including C programmers, will know what you're
doing.
> I don't understand why people push the new thing just for the sake of
> new-ness.
The text read "prefer", which I think it's fair. I'm opposed
to any
hard rule for the sake of new-ness, consistency, or personal
preference.

If we encode all nit-rules as "prefer", then it should be more like a
guideline than a rule book, which has been our motto for everything
else.

cheers,
--renato

2017-01-09 19:25 GMT+01:00 Mehdi Amini <mehdi.amini at apple.com>:
>
> On Jan 9, 2017, at 10:20 AM, Reid Kleckner via llvm-dev <
> llvm-dev at lists.llvm.org> wrote:
>
> On Mon, Jan 9, 2017 at 7:25 AM, Piotr Padlewski via llvm-dev <
> llvm-dev at lists.llvm.org> wrote:
>
>> 2017-01-09 16:15 GMT+01:00 Renato Golin <renato.golin at
linaro.org>:
>>
>>> On 9 January 2017 at 14:17, Piotr Padlewski via cfe-dev
>>> <cfe-dev at lists.llvm.org> wrote:
>>> > - prefer "using' instead of "typedef"
>>> > - use default member initialization
>>> > - use default, override, delete
>>> > - skip "virtual" with override
>>>
>>> I thought we had all of those already...
>>>
>>> Nope, some people use it, but I still see a lot of new code with
>> typedefs.
>> I would like to have it written in style guide so it will be easier to
>> convince to change in review.
>>
>
> The last two are enforced by compiler warnings now. The second is hard
> because of bitfields.
>
> I object to the first. If you need a new type name, use a typedef. It's
> time honored and everyone, including C programmers, will know what
you're
> doing. I don't understand why people push the new thing just for the
sake
> of new-ness.
>
>
> `using` handles strictly more cases than  `typedef`, in particular partial
> specialization of templates. So because we’ll end up with `using` anyway,
> uniformity can be valuable. So that could be a motivation: since `using` is
> needed anyway, might be better to just use it always (I’m not saying it is
> a “strong” motivation though, just some piece of rational).
> (I also find that typedef of function pointers in particular are terrible,
> and `using` is much better for these, but that can be a matter of taste).
>
> —
> Mehdi
>
+1 Exactly this.
I don't think C programmer will not understand using. The "="
makes it much
simpler to read, even if it is the first time you see it, which is not the
case of typedef.

typedef MyType::NestedType (*fptr)(const MyOhterType&);
or
using fptr = MyType::NestedType (*)(const MyOhterType&);

Typedefs with function pointers are used in couple of places in LLVM and I
find it terible to read.

So it is not about new-ness. Trust me, I would never use typedef if using
was first :)
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20170109/23a4600a/attachment.html>

2017-01-09 19:20 GMT+01:00 Reid Kleckner <rnk at google.com>:
> On Mon, Jan 9, 2017 at 7:25 AM, Piotr Padlewski via llvm-dev <
> llvm-dev at lists.llvm.org> wrote:
>
>> 2017-01-09 16:15 GMT+01:00 Renato Golin <renato.golin at
linaro.org>:
>>
>>> On 9 January 2017 at 14:17, Piotr Padlewski via cfe-dev
>>> <cfe-dev at lists.llvm.org> wrote:
>>> > - prefer "using' instead of "typedef"
>>> > - use default member initialization
>>> > - use default, override, delete
>>> > - skip "virtual" with override
>>>
>>> I thought we had all of those already...
>>>
>>> Nope, some people use it, but I still see a lot of new code with
>> typedefs.
>> I would like to have it written in style guide so it will be easier to
>> convince to change in review.
>>
>
> The last two are enforced by compiler warnings now. The second is hard
> because of bitfields.
>
Clang is smart enough to propose using override, but only if it see that it
was used somewhere, so it is not totally enforced. Also I don't think it
warns about redundant "virtual" if one writes "virtual void foo()
override".

Yes, the bitfields is the only one reason we would need initializations in
constructors in some places, but it is rare (as far as what clang-tidy
check have found)
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20170109/c9e97d17/attachment.html>

2017-01-09 19:29 GMT+01:00 Renato Golin <renato.golin at linaro.org>:
> On 9 January 2017 at 18:20, Reid Kleckner <rnk at google.com> wrote:
> > I object to the first. If you need a new type name, use a typedef.
It's
> time
> > honored and everyone, including C programmers, will know what
you're
> doing.
> > I don't understand why people push the new thing just for the sake
of
> > new-ness.
>
> The text read "prefer", which I think it's fair. I'm
opposed to any
> hard rule for the sake of new-ness, consistency, or personal
> preference.
>
> If we encode all nit-rules as "prefer", then it should be more
like a
> guideline than a rule book, which has been our motto for everything
> else.
>
> cheers,
> --renato
>
That's right, everyone has some preferences, but as long as most (over 75%)
of the community prefer one thing over another then I guess it would make
sense to convice the rest and make it official by writing it down. Then we
won't have to spend time dicussing the same things over and over on
different reviews. It is just much simpler to point new (or not)
contributor to a rule in style guide.

Piotr
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20170109/399264f8/attachment.html>

> On Jan 9, 2017, at 10:29 AM, Renato Golin via llvm-dev <llvm-dev at
lists.llvm.org> wrote:
> 
> On 9 January 2017 at 18:20, Reid Kleckner <rnk at google.com> wrote:
>> I object to the first. If you need a new type name, use a typedef.
It's time
>> honored and everyone, including C programmers, will know what
you're doing.
>> I don't understand why people push the new thing just for the sake
of
>> new-ness.
> 
> The text read "prefer", which I think it's fair. I'm
opposed to any
> hard rule for the sake of new-ness, consistency, or personal
> preference.
> 
> If we encode all nit-rules as "prefer", then it should be more
like a
> guideline than a rule book, which has been our motto for everything
> else.
This is not correct according to the number of “should” and the imperative tone
for many aspects of
http://llvm.org/docs/CodingStandards.html#source-code-formatting

— 
Mehdi