Re^3: Could we save the memory occupied by "undef" in an array?

Before you complain about the speed of a hash, benchmark it. The difference between a hash and an array shouldn't be much. I doubt a sparse array would be any faster.

Your program is already heavily invested in the performance of hashes. Most objects use hashes as their underlying data structure. Every reference to a package variable results in multiple hash lookups (I think). And every call to a subroutine results in multiple hash lookups (I think).

Is there a method or package that can squash the memory occupied by "undef" in an array and return the the positions back when I try to fill something into the position that was occupied by a undef, sort of like autovivification?

The benefit of arrays come from knowing exactly where to look into memory for a given element knowing only a base address and the element index. That becomes impossible if some of the elements are missing.

Now, there could be something more efficient data structure than a hash for your needs, but accessing it would require calling a subroutine/method. I believe that results in a hash lookup to find the code pointer associated with the subroutine name, so you're worse off than you started.

On the plus side, arrays use the most efficient of two kinds of undef by default if I read what follows correctly. Instead of creating a new undef SV, uninitialized array elements refer to a global undef constant. That means only sizeof(void*) bytes are wasted per unused slot instead of sizeof(void*) + sizeof(SV).

>perl -MDevel::Peek -e"$a[2]=undef; $a[3]=3; Dump \@a"
SV = RV(0x183c518) at 0x22609c
  REFCNT = 1
  FLAGS = (TEMP,ROK)
  RV = 0x226db0
  SV = PVAV(0x22bfcc) at 0x226db0
    REFCNT = 2
    FLAGS = ()
    IV = 0
    NV = 0
    ARRAY = 0x1833f9c
    FILL = 3
    MAX = 3
    ARYLEN = 0x0
    FLAGS = (REAL)
    Elt No. 0                        # Not allocated
    Elt No. 1                        # Not allocated
    Elt No. 2                        # Allocated but undef
    SV = NULL(0x0) at 0x225f7c
      REFCNT = 1
      FLAGS = ()
    Elt No. 3                        # Allocated and defined
    SV = IV(0x18287b4) at 0x22606c
      REFCNT = 1
      FLAGS = (IOK,pIOK)
      IV = 3
[download]

Comment on Re^3: Could we save the memory occupied by "undef" in an array? Select or Download Code

Replies are listed 'Best First'.
Re^4: Could we save the memory occupied by "undef" in an array? by TimToady (Parson) on Nov 24, 2008 at 17:30 UTC
Your program is already heavily invested in the performance of hashes. Most objects use hashes as their underlying data structure. Every reference to a package variable results in multiple hash lookups (I think). And every call to a subroutine results in multiple hash lookups (I think). Well, then, think again. `:)` To the first approximation, only packages use hashes for their internal structure, and most package symbol lookups happen only at compile time. (Any stored GV* is post-hash-lookup.) Apart from symbolic references (which are discouraged), the only major operation that historically used hash lookups implicitly at run time was method calls, and I think even that has been heavily optimized these days when the method name is known in advance. There are other implicit uses of hashes, but these typically come into play only when you start using higher Unicodes characters in patterns, and even there the internals typically try to avoid using the hash in the common cases. In any event, normal sub calls don't use any implicit hashes. They're already too slow as it is...	[reply] [d/l]
Re^5: Could we save the memory occupied by "undef" in an array? by ikegami (Patriarch) on Nov 24, 2008 at 18:26 UTC
Thanks for the correction. I can't figure out what lead me to think function lookups were made at run-time.	[reply]
Re^4: Could we save the memory occupied by "undef" in an array? by matrixmadhan (Beadle) on Nov 23, 2008 at 08:24 UTC
This looks interesting Would you mind if I request you to explain that? Could you please explain the Dumper output if you have some free time? Thanks	[reply]
Re^5: Could we save the memory occupied by "undef" in an array? by ikegami (Patriarch) on Nov 23, 2008 at 16:32 UTC
Could you please explain the Dumper output if you have some free time? By "I think" and "if I read what follows correctly", I meant to communicate that I'm not clear on the details relevant to this conversation. I've already described the relevant details beyond my understanding. The only other relevant details might be `FILL = 3` (4 elements are visible) and `MAX = 3` (4 elements have been allocated). Perl sometimes allocate a larger array than it needs to speed up future push statements. It also has the option to speed up `pop` statements by not resizing the allocated block. >perl -MDevel::Peek -le"@a=qw(a b c); Dump(\@a, 1); pop @a; Dump(\@a, +1)" SV = RV(0x183c550) at 0x2260b4 REFCNT = 1 FLAGS = (TEMP,ROK) RV = 0x226dbc SV = PVAV(0x22bfcc) at 0x226dbc REFCNT = 2 FLAGS = () IV = 0 NV = 0 ARRAY = 0x1833fc4 FILL = 2 # 0..2 in use MAX = 3 # 0..3 allocated ARYLEN = 0x0 FLAGS = (REAL) SV = RV(0x183c550) at 0x2260a8 REFCNT = 1 FLAGS = (TEMP,ROK) RV = 0x226dbc SV = PVAV(0x22bfcc) at 0x226dbc REFCNT = 2 FLAGS = () IV = 0 NV = 0 ARRAY = 0x1833fc4 FILL = 1 # 0..1 in use MAX = 3 # 0..3 allocated ARYLEN = 0x0 FLAGS = (REAL) [download] Would you mind if I request you to explain that? Did you ask the same question twice, or is this a different "that"?	[reply] [d/l] [select]


Just another Perl shrine
	PerlMonks