Boost C++ Libraries: Ticket #6160: support for (istream >> array < char >)

Sebastian Redl — Tue, 22 Nov 2011 10:05:14 GMT

the standard library provides the syntax (istream >> char []) and (ostream << char const [])

No, it provides istream >> char* and ostream << char const*. As such, you can quite easily use the c_array() and data() members to get pointers to the data: istream >> ar.c_array() and ostream << ar.data().

ne01026@… — Tue, 22 Nov 2011 10:19:51 GMT

Replying to cornedbee:

the standard library provides the syntax (istream >> char []) and (ostream << char const [])

No, it provides istream >> char* and ostream << char const*. As such, you can quite easily use the c_array() and data() members to get pointers to the data: istream >> ar.c_array() and ostream << ar.data().

The call ~~istream >> ar.c_array ()~~ invites buffer overflow and undefined behaviour. If it were that simple, I would not bother writing so many words here.

status changed

Marshall Clow — Mon, 28 Nov 2011 15:52:36 GMT

status new → assigned

I don't see the difference between: istream >> char * , istream >> char [] and istream >> boost::array<char> in terms of buffer overflow prevention.

It seems to me that in each case, it is the responsibility of the person providing the input to make sure that the input fits into the provided buffer.

Am I missing something here?

ne01026@… — Mon, 28 Nov 2011 16:07:04 GMT

char [] has unknown length, array < char > has known length, therefore >> array is safer than >> char [], and my code, attached, q.v., uses this information to prevent buffer overflow.

ne01026@… — Mon, 28 Nov 2011 16:10:45 GMT

Replying to marshall:

It seems to me that in each case, it is the responsibility of the person providing the input to make sure that the input fits into the provided buffer.

This is not possible for general streams, unless your machine happens to be clairvoyant.

Marshall Clow — Mon, 28 Nov 2011 16:15:56 GMT

Replying to ne01026@…:

char [] has unknown length, array < char > has known length, therefore >> array is safer than >> char [], and my code, attached, q.v., uses this information to prevent buffer overflow.

char [] has known length - sizeof is your friend here.

I see where your code is using the length to prevent a buffer overflow - but doing that at every stream extraction is error-prone, at best. It seems to me that (famous last words) that if this were to be implemented, it should never overflow the bounds of the array.

Do you know if this was considered for std::array?

ne01026@… — Mon, 28 Nov 2011 16:43:56 GMT

Replying to marshall:

Replying to ne01026@…:

char [] has unknown length, array < char > has known length, therefore >> array is safer than >> char [], and my code, attached, q.v., uses this information to prevent buffer overflow.

char [] has known length - sizeof is your friend here.

sizeof (char []) is undefined.

I see where your code is using the length to prevent a buffer overflow - but doing that at every stream extraction is error-prone, at best. It seems to me that (famous last words) that if this were to be implemented, it should never overflow the bounds of the array.

The extraction operator will not extract more characters than the array length, minus one, so where is the error?

Do you know if this was considered for std::array?

I do not know; however, given that std takes from boost and not the other way round, I decided to put it here first. At least, the thing seems obvious and straighforward to me. The standard library has special semantics for (native) character arrays, and so should we.

I tried to discuss the matter at #boost and I was told to file an enhancement, so here it is.

Marshall Clow — Mon, 28 Nov 2011 17:01:16 GMT

Replying to ne01026@…:

Replying to marshall:

Replying to ne01026@…:

char [] has unknown length, array < char > has known length, therefore >> array is safer than >> char [], and my code, attached, q.v., uses this information to prevent buffer overflow.

char [] has known length - sizeof is your friend here.

sizeof (char []) is undefined.

Sorry - I was imprecise: char [N] has size N for all N > 0

Do you know if this was considered for std::array?

I do not know; however, given that std takes from boost and not the other way round, I decided to put it here first. At least, the thing seems obvious and straighforward to me. The standard library has special semantics for (native) character arrays, and so should we.

I tried to discuss the matter at #boost and I was told to file an enhancement, so here it is.

And I'm glad that you did - and now we are discussing it.

BTW - if this were to become part of boost, it would be more complicated. You should be able to extract from std::wcin to a boost::array<wchar_t>, for example.

ne01026@… — Mon, 28 Nov 2011 17:08:26 GMT

Replying to marshall:

Replying to ne01026@…:

Replying to marshall:

Replying to ne01026@…:

char [] has unknown length, array < char > has known length, therefore >> array is safer than >> char [], and my code, attached, q.v., uses this information to prevent buffer overflow.

char [] has known length - sizeof is your friend here.

sizeof (char []) is undefined.

Sorry - I was imprecise: char [N] has size N for all N > 0

Unfortunately, the standard library does not provide istream >> char (&) [N] as some older compilers are unable to match against such a template, so the best we can do is to define this thing for array.

BTW - if this were to become part of boost, it would be more complicated. You should be able to extract from std::wcin to a boost::array<wchar_t>, for example.

I already do, that is why the code is so ugly :-(

Marshall Clow — Mon, 28 Nov 2011 17:17:25 GMT

BTW - if this were to become part of boost, it would be more complicated. You should be able to extract from std::wcin to a boost::array<wchar_t>, for example.

I already do, that is why the code is so ugly :-(

I apologize. I clearly did not read your code closely enough.