www.digitalmars.com         C & C++   DMDScript  

digitalmars.D - More std.allocator goodies coming your way: better statistics

reply Andrei Alexandrescu <SeeWebsiteForEmail erdani.org> writes:
Beautiful statistics accumulated for the allocator instance and/or per 
call location:

http://erdani.com/d/phobos-prerelease/std_experimental_allocator_stats_collector.html

https://github.com/andralex/phobos/blob/allocator/std/experimental/allocator/stats_collector.d

Destructors welcome.


Andrei
May 18 2015
next sibling parent reply Timon Gehr <timon.gehr gmx.ch> writes:
On 05/18/2015 08:53 PM, Andrei Alexandrescu wrote:

 Beautiful statistics accumulated for the allocator instance and/or per
 call location:

 http://erdani.com/d/phobos-prerelease/std_experimental_allocator_stats_collector.html


 https://github.com/andralex/phobos/blob/allocator/std/experimental/allocator/stats_collector.d


 Destructors welcome.


 Andrei


- The documentation for numAllocateOK is wrong. (Allocating zero bytes 
is always successful.)

- bytesHighTide is not properly updated. (I suggest to just move the 
check into the implementation of 'add' and be done with it.) Also, is 
this really the only value where it is useful to track a maximum, or 
other summaries?

- The documentation is inconsistent on whether bytesContracted should be 
changed in deallocate; the implementation does it. (I don't see why it 
should.)

- 'static struct Voldermort'

- I think line 602 is not what you wanted.
   'if (!s.next && root != &s)'
   !s.next <-> 's' has not been added, or it is the last element of the 
list
   root != &s <-> 's' is not the first element of the list

   I.e. if 's' is the last element of a list that contains more than one 
element, the list will become cyclic.
May 18 2015
parent Andrei Alexandrescu <SeeWebsiteForEmail erdani.org> writes:
On 5/18/15 1:30 PM, Timon Gehr wrote:

 On 05/18/2015 08:53 PM, Andrei Alexandrescu wrote:

 Beautiful statistics accumulated for the allocator instance and/or per
 call location:

 http://erdani.com/d/phobos-prerelease/std_experimental_allocator_stats_collector.html



 https://github.com/andralex/phobos/blob/allocator/std/experimental/allocator/stats_collector.d



 Destructors welcome.


 Andrei


 - The documentation for numAllocateOK is wrong. (Allocating zero bytes
 is always successful.)


Fixed.


 - bytesHighTide is not properly updated. (I suggest to just move the
 check into the implementation of 'add' and be done with it.) Also, is
 this really the only value where it is useful to track a maximum, or
 other summaries?


Couldn't think of any, but I'm sure use will bring new ideas. In fact 
for better capacity I just made Option a 64-bit value.


 - The documentation is inconsistent on whether bytesContracted should be
 changed in deallocate; the implementation does it. (I don't see why it
 should.)


Well, in per-call stats I "reuse" it for counting calls to deallocate. I 
clarified that in the documentation. I could add another one, but 
bytesUsed is redundant with bytesAllocated - bytesDeallocated (the 
latter doesn't exist) and redundancy sux. Then bytesUsed is very useful 
because it's easy to think about ("how many bytes I'm currently using").


 - 'static struct Voldermort'


Ouch.


 - I think line 602 is not what you wanted.
    'if (!s.next && root != &s)'
    !s.next <-> 's' has not been added, or it is the last element of the
 list
    root != &s <-> 's' is not the first element of the list

    I.e. if 's' is the last element of a list that contains more than one
 element, the list will become cyclic.


That explains a number of things :o). Fixed.


Thanks!

Andrei
May 18 2015
prev sibling parent Jacob Carlborg <doob me.com> writes:
On 2015-05-18 20:53, Andrei Alexandrescu wrote:

 Beautiful statistics accumulated for the allocator instance and/or per
 call location:

 http://erdani.com/d/phobos-prerelease/std_experimental_allocator_stats_collector.html


 https://github.com/andralex/phobos/blob/allocator/std/experimental/allocator/stats_collector.d


 Destructors welcome.


Since Options.all is the default value in StatsCollector. I think "all" 
should be the first member of "Options". Then this will be the default 
value when declaring variables of the type "Options" without 
initializing them.

You have abbreviated "options" to "opts" in PerCallStatistics, I don't 
think that gains much and you haven't abbreviated anything else, 
exception the Options members.

-- 
/Jacob Carlborg
May 18 2015