Victor Bazarov wrote, On 02/05/08 13:21:
> Paul Jackson wrote:
>> Here's a brace-enclosed initialiser for a 3D array:
>>
>> int a4[2][3][4] = {
>> { 1, 2, 3, 4, {5, 6, 7 }, },
>> { {13, 14, 15 }, 17, 18, 19, 20, {21, 22, 23, 24} }};
>>
>> gcc produces some warnings when compiling this, but sets a4 to this,
>> which is what I'd expect:
It is, of course, allowed to produce warnings.
>> { { 1 2 3 4 } { 5 6 7 0 } {0 0 0 0 } }
>> { {13 14 15 0 } {17 18 19 20 } {21 22 23 24 } }
>>
>> It gets more interesting if I leave out initialiser '20':
>>
>> int a4[2][3][4] = {
>> { 1, 2, 3, 4, {5, 6, 7 }, },
I think this part is correct. I believe that footnote 129 makes it
somewhat clearer.
>> { {13, 14, 15 }, 17, 18, 19, {21, 22, 23, 24} }};
I think that this part should produce a diagnostic.
>> gcc now produces lots more warnings, and sets a4 to this:
>>
>> { { 1 2 3 4 } { 5 6 7 0 } { 0 0 0 0 } }
The above would have been correct.
>> { {13 14 15 0 } {17 18 19 21 } { 0 0 0 0 } }
>> Any ideas on what the 'correct' behaviour here should be? I've only
>> checked 6.7.8 in the C99 spec, but this seems to be wrong. I suspect
>> that either reporting an error or setting a4 to
>>
>> { { 1 2 3 4 } { 5 6 7 0 } { 0 0 0 0 } }
>> { {13 14 15 0 } {17 18 19 0 } {21 22 23 24 } }
>>
>> would have better.
>
> The point of the definition (not sure about C99, too lazy to check,
> using C++03, as you ought to) is that in your example with '20' left
> out, the "17, 18, 19" fill the row a4[1][1], and since they are not
> enclosed in their own braces, the following curly brace (before the
> '21') is supposed to enclose the initialiser list for a4[1][1][3].
> AFAICT, since there are more initialisers in the braced list starting
> with '21', the entire initialiser is ill-formed. GCC lets it slide,
> apparently, and simply ignores '22, 23, 24' initialising a4[1][1][3]
> with only '21'. The rest of the array it leaves uninitialised.
In C the rest of the array is definitely *not* left uninitialised. If
you initialise part of an aggregate type (e.g. an array) then the C
standard specifies that all members are initialised. The only exception
to this is unnamed members of a structure objects.
I suspect the rules are similar for C++.
> And
> did you try to use 'strict' mode?
Yes, always worth using "-ansi -pedantic" or "-std=c99 -pedantic" if you
want as close as it comes to C99 conformance. See
http://gcc.gnu.org/c99status.html for information on C99 conformance.
--
Flash Gordon
Similar Threads
