[cfe-dev] [RFC] automatic variable initialization

[Peter Collingbourne via cfe-dev]

On Tue, Nov 27, 2018 at 2:34 PM JF Bastien <jfbastien at apple.com> wrote:

>
>
> On Nov 27, 2018, at 2:14 PM, Peter Collingbourne <peter at pcc.me.uk> wrote:
>
>
>
> On Tue, Nov 27, 2018 at 1:50 PM JF Bastien via cfe-dev <
> cfe-dev at lists.llvm.org> wrote:
>
>>
>>
>> On Nov 27, 2018, at 10:16 AM, David Blaikie via cfe-dev <
>> cfe-dev at lists.llvm.org> wrote:
>>
>>
>>
>> On Mon, Nov 26, 2018 at 6:55 PM Reid Kleckner <rnk at google.com> wrote:
>>
>>> On Mon, Nov 26, 2018 at 6:43 PM Kostya Serebryany via cfe-dev <
>>> cfe-dev at lists.llvm.org> wrote:
>>>
>>>> On Sat, Nov 17, 2018 at 9:00 AM David Blaikie via cfe-dev <
>>>> cfe-dev at lists.llvm.org> wrote:
>>>>
>>>>> Would it be that drastic to have this require a code change/compiler
>>>>> rebuild to enable? It could be designed so the change is small/easy
>>>>> (changing a constant) but that the default compilers we all ship around (&
>>>>> especially not the official releases) don't allow access to this
>>>>> functionality.
>>>>>
>>>>> Anyone wanting to gather data would have to make this small change,
>>>>> rebuild their compiler, build their target with this feature & gatehr
>>>>> results from there.
>>>>>
>>>>
>>>> This will cripple our ability to do measurements because in many cases
>>>> we can only build things with whatever is the production compiler.
>>>> I'd rather just rename the flag to something like
>>>> -ftrivial-auto-var-init=zero-SCARY-WARNING-ABOUT-VOID-WARRANTY-GOES-HERE
>>>>
>>>
>>> Reminds me of -fheinous-gnu-extensions. ;-)
>>>
>>
>> Pretty much - and I don't think any choice of spelling is bad enough that
>> it'd make it substantially less likely that people would end up depending
>> on the behavior in their non-buggy codepaths. Once the flag is written into
>> someone's build system, there it is... even if it's absurdly
>> long/verbose/angry/whatever, generally it'll slip under the radar after
>> it's written into the build system.
>>
>>
>> I’d like to get us to a point where we all agree to a solution. I think
>> these statements are true:
>>
>> 0. We want to minimize the size / perf hit because that’ll maximize
>> adoption.
>> 1. We need to gather size / perf data of uninit versus zero-init versus
>> pattern-init.
>> 2. The easier it is to gather data, the faster it’ll be to find and
>> address regressions we don’t know about (likely in other people’s
>> codebases).
>> 3. We don’t want to have zero-init long term, and we don’t want people to
>> rely on it.
>> 4. Giving the flag / attribute an ugly name won’t prevent people from
>> using them.
>>
>> These seem at odds because we don’t usually remove flags. Can we
>> purposefully change the name of the zero-init flag / attribute on every
>> release, until we finally remove it? Or is the only option to give up on 2.
>> And add friction when trying to collect data?
>>
>
> We can perhaps take inspiration from how the web platform introduces new
> experimental features. Historically this was done using vendor prefixes,
> but this didn't prevent people from using them (i.e. your 4). The approach
> that Blink (and I believe other vendors) has taken is to put these features
> behind a user configurable "enable experimental web platform features”
> setting. The closest analog to the setting for us would probably be a cmake
> build setting. This makes it easy for folks to collect data with the
> feature (no code changes, just reconfigure and rebuild), while making it
> impossible for an arbitrary project to start relying on the flag (unless
> they ship their own copy of clang, but then again that copy of clang could
> be modified in arbitrary ways as well).
>
>
> That still requires a rebuild of clang, right? Experimental features for
> web stuff is often enabled through e.g. about:flags or something similar
> (such as a command flag), you don’t rebuild your browser. Something closer
> for us is a command-line flag, we could also require an environment
> variable with the command-line flag...
>

Yes it requires a rebuild but that's probably an order of magnitude more
straightforward than requiring code changes. Although it's not exactly the
same as web platform the basic idea is the same: enabling the feature is
split up into two steps, where typically in "production" scenarios
different entities are in control of the different steps. In the web
platform case it's the web developer and the browser user and in our case
it's the toolchain builder and the project maintainer. I guess if it's
gated on an environment variable it's still possible for a project to
enable the feature on its own by using "env" in its build commands.

Thanks,
-- 
-- 
Peter
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-dev/attachments/20181127/e2c809e3/attachment.html>