www.digitalmars.com         C & C++   DMDScript  

digitalmars.D.learn - Reducing array.length triggers reallocation

reply milentin <valentin.milea gmail.com> writes:
I've just started learning D and noticed a bug, but wanted to 
confirm it here before reporting it.

According to spec: "If the new array length is shorter, the array 
is not reallocated, and no data is copied. It is equivalent to 
slicing the array". Contradicted by a trivial program:

void main() {
     int[] arr;
     arr.length = 7;
     arr.length = 6; // not ok -- allocation
     int[] slice = arr[0..5]; // ok -- no allocation
}

-------------------------------------------------------
dmd -profile=gc test.d
(DMD32 D Compiler v2.069.2)
-------------------------------------------------------
bytes allocated, allocations, type, function, file:line
              28               1 int[] D main test.d:3
              24               1 int[] D main test.d:4
Dec 27 2015
parent reply =?UTF-8?Q?Ali_=c3=87ehreli?= <acehreli yahoo.com> writes:
On 12/27/2015 02:09 AM, milentin wrote:

 I've just started learning D and noticed a bug, but wanted to confirm it
 here before reporting it.

 According to spec: "If the new array length is shorter, the array is not
 reallocated, and no data is copied. It is equivalent to slicing the
 array". Contradicted by a trivial program:

 void main() {
      int[] arr;
      arr.length = 7;
      arr.length = 6; // not ok -- allocation
      int[] slice = arr[0..5]; // ok -- no allocation
 }

 -------------------------------------------------------
 dmd -profile=gc test.d
 (DMD32 D Compiler v2.069.2)
 -------------------------------------------------------
 bytes allocated, allocations, type, function, file:line
               28               1 int[] D main test.d:3
               24               1 int[] D main test.d:4


I don't understand why that happens. I found one related bug:

   https://issues.dlang.org/show_bug.cgi?id=13750

I can understand that assignment to arr.length cannot be  nogc but I 
would expect a check against length so that there would be no allocation.

At least there are no copies and .ptr property of the array does not change.

[Several hours later...]

You know what... I bet there is no actual allocation at all. I think 
what happens is, the code calls GC.realloc(24) and realloc() does not do 
anything. However, it still reports to the profiler that there was an 
allocation (attempt).

Can someone verify that please. At least, can someone show where 
GC.realloc() source is.

Thank you,
Ali
Dec 27 2015
next sibling parent Ivan Kazmenko <gassa mail.ru> writes:
On Sunday, 27 December 2015 at 22:36:32 UTC, Ali Çehreli wrote:

 [Several hours later...]

 You know what... I bet there is no actual allocation at all. I 
 think what happens is, the code calls GC.realloc(24) and 
 realloc() does not do anything. However, it still reports to 
 the profiler that there was an allocation (attempt).

 Can someone verify that please. At least, can someone show 
 where GC.realloc() source is.

 Thank you,
 Ali


I believe it boils down to calling gc.gc.reallocNoSync in 
druntime: 
https://github.com/D-Programming-Language/druntime/blob/master/src/gc/gc.d#L603
.
Dec 27 2015
prev sibling parent milentin <valentin.milea gmail.com> writes:
On Sunday, 27 December 2015 at 22:36:32 UTC, Ali Çehreli wrote:

 I don't understand why that happens. I found one related bug:

   https://issues.dlang.org/show_bug.cgi?id=13750

 I can understand that assignment to arr.length cannot be  nogc 
 but I would expect a check against length so that there would 
 be no allocation.

 At least there are no copies and .ptr property of the array 
 does not change.

 [Several hours later...]

 You know what... I bet there is no actual allocation at all. I 
 think what happens is, the code calls GC.realloc(24) and 
 realloc() does not do anything. However, it still reports to 
 the profiler that there was an allocation (attempt).

 Can someone verify that please. At least, can someone show 
 where GC.realloc() source is.

 Thank you,
 Ali


Thanks for the feedback, I have opened 
https://issues.dlang.org/show_bug.cgi?id=15481.
Dec 27 2015