www.digitalmars.com         C & C++   DMDScript  

digitalmars.D - Is the following code legal?

reply Shachar Shemesh <shachar weka.io> writes:
int[int] hash;

..

foreach( key, ref value; hash ) {
	if( value>12 )
		hash.remove(key);
}

Some hash implementations support this, some don't. The D documentation 
(https://dlang.org/spec/hash-map.html) leaves this not defined.

As reference, C++ does define this (in C++ it is allowed, at least since 
C++14: http://en.cppreference.com/w/cpp/container/unordered_map/erase)

Shachar
Nov 02 2016
next sibling parent reply Andrei Alexandrescu <SeeWebsiteForEmail erdani.org> writes:
On 11/02/2016 10:21 AM, Shachar Shemesh wrote:

 int[int] hash;

 ..

 foreach( key, ref value; hash ) {
     if( value>12 )
         hash.remove(key);
 }

 Some hash implementations support this, some don't. The D documentation
 (https://dlang.org/spec/hash-map.html) leaves this not defined.

 As reference, C++ does define this (in C++ it is allowed, at least since
 C++14: http://en.cppreference.com/w/cpp/container/unordered_map/erase)

 Shachar


We should render it defined, and document it as such. Could you please 
create an issue and I'll have someone look at it. Thanks! -- Andrei
Nov 02 2016
parent Shachar Shemesh <shachar weka.io> writes:
On 02/11/16 16:52, Andrei Alexandrescu wrote:

 We should render it defined, and document it as such. Could you please
 create an issue and I'll have someone look at it. Thanks! -- Andrei


https://issues.dlang.org/show_bug.cgi?id=16659
Nov 03 2016
prev sibling parent reply Adam D. Ruppe <destructionator gmail.com> writes:
On Wednesday, 2 November 2016 at 14:21:32 UTC, Shachar Shemesh 
wrote:

 The D documentation (https://dlang.org/spec/hash-map.html) 
 leaves this not defined.


The foreach statement is defined to not allow it:

http://dlang.org/spec/statement.html#ForeachStatement

"The aggregate must be loop invariant, meaning that elements to 
the aggregate cannot be added or removed from it in the 
NoScopeNonEmptyStatement."
Nov 02 2016
next sibling parent Andrei Alexandrescu <SeeWebsiteForEmail erdani.org> writes:
On 11/02/2016 11:17 AM, Adam D. Ruppe wrote:

 On Wednesday, 2 November 2016 at 14:21:32 UTC, Shachar Shemesh wrote:

 The D documentation (https://dlang.org/spec/hash-map.html) leaves this
 not defined.


 The foreach statement is defined to not allow it:

 http://dlang.org/spec/statement.html#ForeachStatement

 "The aggregate must be loop invariant, meaning that elements to the
 aggregate cannot be added or removed from it in the
 NoScopeNonEmptyStatement."


Yah, we'd do good to relax that to allow removal of the currently 
iterated element. -- Andrei
Nov 02 2016
prev sibling parent Steven Schveighoffer <schveiguy yahoo.com> writes:
On 11/2/16 11:17 AM, Adam D. Ruppe wrote:

 On Wednesday, 2 November 2016 at 14:21:32 UTC, Shachar Shemesh wrote:

 The D documentation (https://dlang.org/spec/hash-map.html) leaves this
 not defined.


 The foreach statement is defined to not allow it:

 http://dlang.org/spec/statement.html#ForeachStatement

 "The aggregate must be loop invariant, meaning that elements to the
 aggregate cannot be added or removed from it in the
 NoScopeNonEmptyStatement."


That's only for builtins. Obviously, there are cases where it can work, 
and it needs to be defined by the aggregate/range. I think the 
documentation should be updated to reflect that.

To answer the original question, it's not valid with the current 
implementation AFAIK. I don't think we should define ever that it is 
valid, even if we have an implementation that supports it, as this 
restricts our implementation to always supporting it.

I'll point at my dcollections library as an example where the currently 
iterated value can be removed:

https://github.com/schveiguy/dcollections/blob/master/concepts.txt#L81

I think Java also allows this, and C++ allows this.

-Steve
Nov 03 2016