use warnings;
use Inline C => Config =>
    BUILD_NOISY => 1;

use Inline C => <<'EOC';

struct s {
    int field1;
    int field2;
    int field3;
};
void do_simulation() {
    int i;
    struct s * m = (struct s*)malloc(10000000 * sizeof(struct s));

    for (i=0; i<10000000; i++) {
        m[i].field1 = 0;
        m[i].field2 = 1;
        m[i].field3 = 2;
    } 


    printf("Sleeping for 5\n");   
    sleep(5);
    
    free(m);
    printf("Memory freed ... sleep for another 5\n");

    sleep(5);
    printf("Done\n");
}

void do_simulation2() {
    int i;
    struct s* m;

    New(1, (struct s*)m, 10000000, struct s);

    for ( i=0; i<10000000; i++ ) {
        m[i].field1 = 0;
        m[i].field2 = 1;
        m[i].field3 = 2;
    } 

    printf("Sleeping for 5\n");   
    sleep(5);   

    Safefree(m);
    printf("Memory freed ... sleep for another 5\n");

    sleep(5);
    printf("Done\n");
}



EOC

do_simulation();
do_simulation2();
[download]

Thank you, Rob, for your participation and reproducing the code on your Win32/perl-5-12 environment. I described my problem bearing in mind perl 5.6 (I know, it's an old version ... maybe even an archaic one), but your perl version triggered my intention to try a more recent perl.

Firstly I wanted to try on a fresh environment. After reinstalling Win/MSVS6/Perl everything worked the same way ... Then I tried Perl 5.8 with the SAME (only #include "ppport.h" was added by h2xs) code and it worked correctly.

Maybe it's premature to state that it's a Perl 5.6 (more precisely ActiveState Perl 5.6.1 Build 638) bug, but Perl 5.8 doesn't have this memory leak problem.

Now I need to upgrade Perl/mod_perl. A last question would be what do you recommend for a 'not-enterprise production environment': 5.8, 5.10, or 5.12? Is there a stable mod_perl module built from 5.12 sources for Apache 2.2 (all Win32)? Or is it better to use mod_perl with Apache 2.0?

I described my problem bearing in mind perl 5.6

Aaah ... when I run my demo on perl-5.6 I can see what you mean - the memory usage does not drop down until the script exits.

However, I think it's as ikegami says below: ...the memory is being put back into Perl's free memory pool and you just don't realize it.
With perl-5.6 (running the script I posted), note that when do_simulation2() is called, the memory usage does *not* jump up. This is a sure sign that it is re-using the memory that was freed by do_simulation().

So there's no memory leaking going on, and no bug - it's just that with 5.6, freed memory stays locked up in the pool, whereas with 5.12 freed memory is being released back to the system. Task Manager therefore sees the memory release under 5.12, but doesn't see the memory release under 5.6. (It seems to me that 5.6's behaviour in this regard is probably more efficient than 5.12's ... but I expect that there are pros and cons.)

Now I need to upgrade Perl/mod_perl. A last question would be what do you recommend ...

Sorry - I'm way out of touch with Apache and mod_perl. I don't think you should be worrying about updating just on the strength of this memory handling behaviour of 5.6 ... unless, of course, you really do need the freed memory to be released back to the system.

Cheers,
Rob

I've tried without "stdlib.h". The same result for both cases (malloc()/free() and New()/Safefree())

I aslo created a new MSVC6 project with the same code (perl paths included):

#include "EXTERN.h"
#include "perl.h"
#include "XSUB.h"

struct s {
    int field1;
    int field2;
    int field3;
};

void do_simulation() {
    int i;
    struct s * m = (struct s*)malloc(10000000 * sizeof(struct s));

    for (i=0; i<10000000; i++) {
        m[i].field1 = 0;
        m[i].field2 = 1;
        m[i].field3 = 2;
    }    
    
    /* sleep for 5 seconds to be able to analyze memory usage
    sleep(5);
    
    free(m);

    /* sleep for 5 seconds to be able to see if memory is freed 
    sleep(5);
}


int main(void) {

    do_simulation();

    return 1;
}
[download]

... and everithing works as it should: first 5 seconds the process uses 200+ MB (VM included), the last 5 seconds memory usage drops down to ~600 KB