Hacker News new | comments | show | ask | jobs | submit login
C++ compiler accepts explicit constructor call (microsoft.com)
24 points by 430gj9j 1631 days ago | hide | past | web | 29 comments | favorite

While I'm all for improved adherence to standards, the original linked bug report seems like a terrible case to pick a bone about attitudes towards standards compliance with.

There are real issues causing real problems which really should be fixed, very much including those related to standards compliance, but this doesn't appear to be one of them. The cure would be worse than the poison: Breaking existing codebases further, making it even harder to update legacy codebases to new toolsets, for an error made -- I'm guessing wildly here instead of verifying -- back in the VS6 era perhaps? That's a lot of code. What do we gain? We... make it slightly harder to accidentally write MSVC specific code. We already have a much, much better and more thorough tool for that: Compiling with a non MSVC compiler. I'd rather see the dev time put towards other issues.

As for not hitting C++11 standards compliance straight off the bat, they tried that for C++98 in VS6 and got burned by last minute changes. Yes, it'd be nice if they implemented everything correctly. The bug reports calling out problems with that compliance are very well and good. That said, they aren't billing themselves as feature-complete WRT C++11 yet, and I'd rather take this lackadaisical tempo than see yet another round of implementation mistakes which then need indefinite support.

Of course, this is probably colored by the fact that I don't get to use C++11 yet anyways -- out of the environments I'm currently stuck supporting, MSVC is at the bleeding edge of the curve.

Please don't editorialise submission titles.

Titling a submission "Visual C++ team's attitude to standards compliance" and pointing it at a bug report for some minor non-conforming behaviour tells us nothing useful about the msvc teams attitude to standards. Neither does the brief response from the team on that page. All it says is that there is an issue that they're not going to fix at the moment. Microsoft have a lot of customers, and those customers have a lot of code that they don't want to suddenly break without a good reason.

Do some work. Find or write a useful article investigating the c++ teams attitude to standards compliance and post a link to that. If it compared msvc to other compilers then I'd read it.

Is the implication that other teams are better with standards compliance?

Here is a much, much more serious issue. Variable length arrays of C++ types. Totally not in the C++ standard anywhere. Accepted by g++ and clang++, even if I turn on all warnings, and use '-std=c++11', which is supposed to turn off extensions. I have to add -pedantic to finally get a warning.

Now, I'm not saying they should break this code, but this is a much, much, much more serious unlabelled breakage of the standard, which is never going to get fixed, and goes back to the start of g++ and clang++.

    struct X {};

    int main(int argc, char** argv)
    { X a[argc]; }

--std=c++11 is not supposed to turn off all extensions, only the extensions that conflict with the standard. The VLA extension only applies to programs that would be malformed according to the standard, so it isn't turned off with -std=c++11. The -pedantic flag is for adhering to the standard exactly.

Dynamic arrays will be in C++14.


That being said, you are correct that it would be good to see warnings in something other than just -pedantic for that.

Standards nonconformance has always been less of a problem when gcc does it.

(Of course, part of that is presumably that gcc's nonconformance is generally useful, as in this case...)

FOSS developers like to forget that GCC is also full of language extensions and nonconformance issues.

This is pretty standard type of issues for any ANSI/ISO language regardless of the vendor.

Oh boy the joys of doing C and C++ development across multiple OS in the late 90's with commercial compilers.

Gcc was pretty wild-n-wooly back in the '90s, and seemed to add language extensions left and right, but that attitude changed quite a while ago, and it's very good about standards conformance these days (part of the reason for this, of course is it's no longer so necessary: many of these extensions have picked up by the standard in one form or another).

In my experience, gcc is also significantly better with standards conformance than VC++, either with default settings or with strict warning/conformance options turned on. [I was on a team doing shared g++/vc++ develeopment, and it fell to me to fix all the code checked in by devs using non-standard VC++ extensions... way, way, way too much time... >< ]

The example given in the "story" is useful.

Not terribly - you can get the same effect with `return Thing()'. This doesn't seem as useful as the similar bug/extension that allows pre-C11 scoped enums (see, e.g., http://stackoverflow.com/questions/441552/scope-resolution-o...).

Just for fun, I checked that code with gcc 4.6 and clang 3.2. gcc gives the error:

error: cannot call constructor ‘Thing::Thing’ directly [-fpermissive]

clang compiles the code without issuing an error.

Unless this breaks valid C++ code, I don't see how they need care too much about fixing it, especially since that would break existing code.

A quote from Stroustrup:

"Use of a supplier's language extensions and non-standard-conforming features limits the portability of your code and can prevent you from choosing a new implementation supplier."


A sibling thread claims the same syntax is supported by clang, and allowed by gcc using -fpermissive, so it seems like that's not a practical concern in this instance.

Quite a valid observation from Stroustrup. But in this case the programmer has chosen to use a non-standard, vendor specific feature. So it is not clear how it applies here? (Especially given the cost of breaking existing code)

The programmer inadvertently used a vendor-specific feature. This is the key difference. Nobody writes perfect, standards-compliant code all the time, so the compiler needs to tell you when you are doing this.

Right, but it's a trivial fix, in this case at least. I'm happy as long as standard code works in their compiler. Considering how much trouble MS are having with getting their compiler standard-conformant, I'd rather they work on the more important features.

Their approach seems reasonable. If they fix this, they will break code that currently works by depending on this. They've not said that they will never fix it, just that they won't for the time being and will reconsider for a future release.

The fix could be just to issue a warning of non-compliance to C++ standards.

Except that can break compiling projects as well due to /WX (treating warnings as errors).

Those projects want to be warned of non-compliant code and they have chosen to have their build break on questionable constructs. I think that this is a weak argument for not issuing a warning.

Having said that the issue looks fairly non-urgent and inclusion in a later release would be perfectly reasonable.

C++ noob here. This reminded me of "Obscure C++ Features" [1] (discussion [2]) - the 'Placement new' feature. Basically, we can allocate memory using malloc and then call the constructor on the allocated memory:

  // Must allocate our own memory
  Test *ptr = (Test *)malloc(sizeof(Test));

  // Use placement new
  new (ptr) Test;

  // Must call the destructor ourselves

  // Must release the memory ourselves

My point is: is possible that the explicit constructor call feature is intended to be used in this use case?

[1]: http://madebyevan.com/obscure-cpp-features/

[2]: https://news.ycombinator.com/item?id=5577631

When I last worked with Windows and would end up on that Microsoft Connect site fairly regularly by Googling, I don't think I ever saw a response that would be satisfactory to the reporter. "Thank for for holding; your call is important to us" or "No", possibly rephrased a few times, was about your lot.

How is this different from any other commercial vendor?

seems standard. visual studio releases are years behind each other so they're not going to fix anything that's not critical.

In the scope of compliance issues, that is not a huge issue because no C++ developer would write that code.

In a changing world, you can depend on Microsoft: "Where you want to go today? Meh, as if we care; we have already decided on that."

Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | DMCA | Apply to YC | Contact