Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > comp.lang.c++ > #84265 > unrolled thread

Vector - contiguous memory question.

Back to article view | Back to comp.lang.c++

Contents

  Vector - contiguous memory question. Muttley@dastardlyhq.com - 2022-05-26 08:51 +0000
    Re: Vector - contiguous memory question. Bo Persson <bo@bo-persson.se> - 2022-05-26 11:28 +0200
    Re: Vector - contiguous memory question. "Alf P. Steinbach" <alf.p.steinbach@gmail.com> - 2022-05-26 12:36 +0200
      Re: Vector - contiguous memory question. Muttley@dastardlyhq.com - 2022-05-26 16:02 +0000
    Re: Vector - contiguous memory question. "Alf P. Steinbach" <alf.p.steinbach@gmail.com> - 2022-05-26 18:25 +0200
    Re: Vector - contiguous memory question. James Kuyper <jameskuyper@alumni.caltech.edu> - 2022-05-26 13:53 -0400
    Re: Vector - contiguous memory question. Vir Campestris <vir.campestris@invalid.invalid> - 2022-05-27 17:12 +0100
      Re: Vector - contiguous memory question. James Kuyper <jameskuyper@alumni.caltech.edu> - 2022-05-27 12:32 -0400
        Re: Vector - contiguous memory question. Muttley@dastardlyhq.com - 2022-05-28 09:07 +0000
          Re: Vector - contiguous memory question. Ben <ben.usenet@bsb.me.uk> - 2022-05-28 16:36 +0100
            Re: Vector - contiguous memory question. "Chris M. Thomasson" <chris.m.thomasson.1@gmail.com> - 2022-05-28 17:14 -0700
              Re: Vector - contiguous memory question. Ben <ben.usenet@bsb.me.uk> - 2022-05-29 01:47 +0100
              Re: Vector - contiguous memory question. "james...@alumni.caltech.edu" <jameskuyper@alumni.caltech.edu> - 2022-05-28 22:06 -0700
            Re: Vector - contiguous memory question. Muttley@dastardlyhq.com - 2022-05-29 07:32 +0000
              Re: Vector - contiguous memory question. Ben <ben.usenet@bsb.me.uk> - 2022-05-29 19:43 +0100

#84265 — Vector - contiguous memory question.

Is contiguous memory storage in a vector a recent thing or is it only
in more recent versions of C++? Ie is the following always guaranteed to
work no matter how old the compiler:

vector<u_char> mydata;
:
fill vector
:
write(fd,&mydata[0],mydata.size());

or 

write(fd,mydata.data(),mydata.size());

[toc] | [next] | [standalone]

#84266

On 2022-05-26 at 10:51, Muttley@dastardlyhq.com wrote:
> Is contiguous memory storage in a vector a recent thing or is it only
> in more recent versions of C++? Ie is the following always guaranteed to
> work no matter how old the compiler:
> 
> vector<u_char> mydata;
> :
> fill vector
> :
> write(fd,&mydata[0],mydata.size());
> 
> or
> 
> write(fd,mydata.data(),mydata.size());
> 

If by "recent compiler" you mean anyone following the C++ standard, then 
vector storage is contiguous. And has been since C++98.

If you use Turbo-C++ from 1990, I'm not sure.  :-)

[toc] | [prev] | [next] | [standalone]

#84267

On 26 May 2022 10:51, Muttley@dastardlyhq.com wrote:
> Is contiguous memory storage in a vector a recent thing or is it only
> in more recent versions of C++? Ie is the following always guaranteed to
> work no matter how old the compiler:
> 
> vector<u_char> mydata;
> :
> fill vector
> :
> write(fd,&mydata[0],mydata.size());
> 
> or
> 
> write(fd,mydata.data(),mydata.size());

`std::vector` has always had a guaranteed contiguous buffer.

That includes in the pre-standard Annotated Reference Manual days.

`std::string` got a contiguous buffer guarantee at the meeting at 
Lillehammer in 2005, incorporated into the official standard in C++11.


Cheers,

- Alf

[toc] | [prev] | [next] | [standalone]

#84273

On Thu, 26 May 2022 12:36:53 +0200
"Alf P. Steinbach" <alf.p.steinbach@gmail.com> wrote:
>On 26 May 2022 10:51, Muttley@dastardlyhq.com wrote:
>> Is contiguous memory storage in a vector a recent thing or is it only
>> in more recent versions of C++? Ie is the following always guaranteed to
>> work no matter how old the compiler:
>> 
>> vector<u_char> mydata;
>> :
>> fill vector
>> :
>> write(fd,&mydata[0],mydata.size());
>> 
>> or
>> 
>> write(fd,mydata.data(),mydata.size());
>
>`std::vector` has always had a guaranteed contiguous buffer.
>
>That includes in the pre-standard Annotated Reference Manual days.
>
>`std::string` got a contiguous buffer guarantee at the meeting at 
>Lillehammer in 2005, incorporated into the official standard in C++11.

Yes, I think I may have been confusing it with string.

[toc] | [prev] | [next] | [standalone]

#84274

On 26 May 2022 17:48, Stefan Ram wrote:
> ram@zedat.fu-berlin.de (Stefan Ram) writes:
>> In practice, probably yes. But the first paragraph of
>> ISO 14882 23.2.4 contains the clause "The elements of a
>> vector are stored contiguously" only in 2003, not in 1998.
> 
>    It was inserted by section 69 of n1516 "C++ Standard Library
>    Defect Report List (Revision 27)" submitted by Andrew Koenig
>    on 29 Jul 1998.

Yeah, means it was intended in C++98. C++03 was Technical Corrigendum 1 
(the only one) of C++98. The only really new functionality was value 
initialization, also Andrew Koenig's work.


Cheers,

- Alf

[toc] | [prev] | [next] | [standalone]

#84275

On 5/26/22 04:51, Muttley@dastardlyhq.com wrote:
> Is contiguous memory storage in a vector a recent thing or is it only
> in more recent versions of C++? Ie is the following always guaranteed to
> work no matter how old the compiler:
> 
> vector<u_char> mydata;
> :
> fill vector
> :
> write(fd,&mydata[0],mydata.size());
> 
> or 
> 
> write(fd,mydata.data(),mydata.size());

In the original, 1998 version of the C++ standard, there was no
requirement that std::vector<> be contiguous. I participated in several
discussions on comp.std.c++ about that issue. Some people argued that
contiguity could be inferred from other things that the standard said
about std::vector<>, but those inferrences were not correct. Notably,
all the reasons given for believing that std::vector<> had to be
contiguous applied equally well to std::valarray<>. This is a problem,
because std::valarray<> was explicitly required to be contiguous. If
those arguments were valid, then they should apply to std::valarray<>
too, with the result that there was no need for the standard to specify
that the it was contiguous.
You can find those discussions using groups.google.com and searching for
 	contig vector

My copy of C++98 is long gone, so I can't check this, but no one cited
any text from the standard mandating that it be contiguous. Since I'm an
inveterate standard-quoter, and was participating in that discussion, if
there had been any such text, I would have quoted it.

The consensus was that std::vector<> should have been required to be
contiguous, and that the standard was defective by reason of failing to
do so. Andrew Koenig filed DR69
<https://cplusplus.github.io/LWG/issue69>, and the proposed wording was
inserted in TC1.

[toc] | [prev] | [next] | [standalone]

#84284

On 26/05/2022 09:51, Muttley@dastardlyhq.com wrote:
> write(fd,&mydata[0],mydata.size());
> 
> or
> 
> write(fd,mydata.data(),mydata.size());

Nitpick for you - that won't write everything unless your array is chars.

Andy

[toc] | [prev] | [next] | [standalone]

#84285

On 5/27/22 12:12, Vir Campestris wrote:
> On 26/05/2022 09:51, Muttley@dastardlyhq.com wrote:
>> write(fd,&mydata[0],mydata.size());
>>
>> or
>>
>> write(fd,mydata.data(),mydata.size());
> 
> Nitpick for you - that won't write everything unless your array is chars.

Agreed - but his array was of u_char, presumably a typedef for unsigned
char or something similar.

[toc] | [prev] | [next] | [standalone]

#84309

On Fri, 27 May 2022 12:32:15 -0400
James Kuyper <jameskuyper@alumni.caltech.edu> wrote:
>On 5/27/22 12:12, Vir Campestris wrote:
>> On 26/05/2022 09:51, Muttley@dastardlyhq.com wrote:
>>> write(fd,&mydata[0],mydata.size());
>>>
>>> or
>>>
>>> write(fd,mydata.data(),mydata.size());
>> 
>> Nitpick for you - that won't write everything unless your array is chars.
>
>Agreed - but his array was of u_char, presumably a typedef for unsigned
>char or something similar.

Defined in sys/types.h

[toc] | [prev] | [next] | [standalone]

#84316

Muttley@dastardlyhq.com writes:

> On Fri, 27 May 2022 12:32:15 -0400
> James Kuyper <jameskuyper@alumni.caltech.edu> wrote:
>>On 5/27/22 12:12, Vir Campestris wrote:
>>> On 26/05/2022 09:51, Muttley@dastardlyhq.com wrote:
>>>> write(fd,&mydata[0],mydata.size());
>>>>
>>>> or
>>>>
>>>> write(fd,mydata.data(),mydata.size());
>>> 
>>> Nitpick for you - that won't write everything unless your array is chars.
>>
>>Agreed - but his array was of u_char, presumably a typedef for unsigned
>>char or something similar.
>
> Defined in sys/types.h

Since it's not a standard type (as far as C++ is concerned) you can't be
sure what it is, and I don't think the OP gave a definition nor an
include file it might be defined in.

-- 
Ben.

[toc] | [prev] | [next] | [standalone]

#84320

On 5/28/2022 8:36 AM, Ben wrote:
> Muttley@dastardlyhq.com writes:
> 
>> On Fri, 27 May 2022 12:32:15 -0400
>> James Kuyper <jameskuyper@alumni.caltech.edu> wrote:
>>> On 5/27/22 12:12, Vir Campestris wrote:
>>>> On 26/05/2022 09:51, Muttley@dastardlyhq.com wrote:
>>>>> write(fd,&mydata[0],mydata.size());
>>>>>
>>>>> or
>>>>>
>>>>> write(fd,mydata.data(),mydata.size());
>>>>
>>>> Nitpick for you - that won't write everything unless your array is chars.
>>>
>>> Agreed - but his array was of u_char, presumably a typedef for unsigned
>>> char or something similar.
>>
>> Defined in sys/types.h
> 
> Since it's not a standard type (as far as C++ is concerned) you can't be
> sure what it is, and I don't think the OP gave a definition nor an
> include file it might be defined in.
> 

A std::vector better be contiguous!

[toc] | [prev] | [next] | [standalone]

#84322

"Chris M. Thomasson" <chris.m.thomasson.1@gmail.com> writes:

> On 5/28/2022 8:36 AM, Ben wrote:
>> Muttley@dastardlyhq.com writes:
>> 
>>> On Fri, 27 May 2022 12:32:15 -0400
>>> James Kuyper <jameskuyper@alumni.caltech.edu> wrote:
>>>> On 5/27/22 12:12, Vir Campestris wrote:
>>>>> On 26/05/2022 09:51, Muttley@dastardlyhq.com wrote:
>>>>>> write(fd,&mydata[0],mydata.size());
>>>>>>
>>>>>> or
>>>>>>
>>>>>> write(fd,mydata.data(),mydata.size());
>>>>>
>>>>> Nitpick for you - that won't write everything unless your array is chars.
>>>>
>>>> Agreed - but his array was of u_char, presumably a typedef for unsigned
>>>> char or something similar.
>>>
>>> Defined in sys/types.h
>>
>> Since it's not a standard type (as far as C++ is concerned) you can't be
>> sure what it is, and I don't think the OP gave a definition nor an
>> include file it might be defined in.
>
> A std::vector better be contiguous!

The thread has moved on from that.  The point here is whether the size
of the copy is correct.

-- 
Ben.

[toc] | [prev] | [next] | [standalone]

#84324

On Saturday, May 28, 2022 at 8:14:24 PM UTC-4, Chris M. Thomasson wrote:
...
> A std::vector better be contiguous!

It seems a little odd to post that response to a message that contains no mention of the original question about std::vector<> contiguity.

That question asked about whether std::vector<> had ever lacked a guarantee of contiguity, and if fact it did NOT have such a guarantee in the first version of the C++ standard; that guarantee was only added in TC1.

[toc] | [prev] | [next] | [standalone]

#84328

On Sat, 28 May 2022 16:36:52 +0100
Ben <ben.usenet@bsb.me.uk> wrote:
>Muttley@dastardlyhq.com writes:
>
>> On Fri, 27 May 2022 12:32:15 -0400
>> James Kuyper <jameskuyper@alumni.caltech.edu> wrote:
>>>On 5/27/22 12:12, Vir Campestris wrote:
>>>> On 26/05/2022 09:51, Muttley@dastardlyhq.com wrote:
>>>>> write(fd,&mydata[0],mydata.size());
>>>>>
>>>>> or
>>>>>
>>>>> write(fd,mydata.data(),mydata.size());
>>>> 
>>>> Nitpick for you - that won't write everything unless your array is chars.
>>>
>>>Agreed - but his array was of u_char, presumably a typedef for unsigned
>>>char or something similar.
>>
>> Defined in sys/types.h
>
>Since it's not a standard type (as far as C++ is concerned) you can't be
>sure what it is, and I don't think the OP gave a definition nor an
>include file it might be defined in.

The OP was me and its been a standard posix typdef for as long as I've been
programming so if you're a Windows def you probably won't have come across it.

[toc] | [prev] | [next] | [standalone]

#84364

Muttley@dastardlyhq.com writes:

> On Sat, 28 May 2022 16:36:52 +0100
> Ben <ben.usenet@bsb.me.uk> wrote:
>>Muttley@dastardlyhq.com writes:
>>
>>> On Fri, 27 May 2022 12:32:15 -0400
>>> James Kuyper <jameskuyper@alumni.caltech.edu> wrote:
>>>>On 5/27/22 12:12, Vir Campestris wrote:
>>>>> On 26/05/2022 09:51, Muttley@dastardlyhq.com wrote:
>>>>>> write(fd,&mydata[0],mydata.size());
>>>>>>
>>>>>> or
>>>>>>
>>>>>> write(fd,mydata.data(),mydata.size());
>>>>> 
>>>>> Nitpick for you - that won't write everything unless your array is chars.
>>>>
>>>>Agreed - but his array was of u_char, presumably a typedef for unsigned
>>>>char or something similar.
>>>
>>> Defined in sys/types.h
>>
>>Since it's not a standard type (as far as C++ is concerned) you can't be
>>sure what it is, and I don't think the OP gave a definition nor an
>>include file it might be defined in.
>
> The OP was me and its been a standard posix typdef for as long as I've been
> programming so if you're a Windows def you probably won't have come
> across it.

I'm aware of what is often means.  I was pointing out that what it
actually it will always be something a guess without the context.

-- 
Ben.

[toc] | [prev] | [standalone]

Back to top | Article view | comp.lang.c++

csiph-web