[Black Rose]

You can ignore this, but I'll leave it for the sake of "new x[]" vs. "malloc()" for Arkshine to consider. However I realize that it might not have crashed if I hand't sent it a negative number...
The problem was that when using chunked transfers it passed the next size marker and I tried to allocate -166B. I feel like a moron now.

Spoiler

I'm stressing my HTTP thingy by running and restarting it and the server keeps crashing randomly.
I have traced it down to this line inside socket_recv():

// native socket_recv(_socket, _data[], _length);
static cell AMX_NATIVE_CALL socket_recv(AMX *amx, cell *params)  /* 2 param */
{
    int socket = params[1];
    int length = params[3];
    int tmp = -1;
    // First we dynamicly allocate a block of heap memory for recieving our data
    char *tmpchar = new char[length];
    if(tmpchar == NULL) return -1; // If we didn't got a block, we have to quit here to avoid sigsegv
    // And set it all to 0, because the memory could contain old trash
    memset(tmpchar, 0, length);
    // Now we recieve
    tmp = recv(socket, tmpchar, length-1, 0);
    if (tmp == -1)
    {
        delete [] tmpchar;
        return -1;
    }
    // And put a copy of our recieved data into amx's string
    tmpchar[tmp]='\0';
    int nlen = 0;
    //int max = params[3];
    int max = length-1;
    const char* src = tmpchar;
    cell* dest = MF_GetAmxAddr(amx,params[2]);
    while(max--&&nlen<tmp){
     *dest++ = (cell)*src++;
     nlen++;
    }
    *dest = 0;
    // And we need to free up the space to avoid wasting memory
    delete [] tmpchar;
    // And finnally, return the what recv returnd
    return tmp;
}

If I can remember correctly I think there's a #pragma to increase the memory allocated to a plugin. Would that fix the problem? I'm assuming no since this is inside the module, but I don't understand scripting fully at this low level. Is there a better approach?

Is this considered a bug or am I doing something wrong?

I have included all the files required for testing.
This is what I changed in the .dll:

Spoiler

// native socket_recv(_socket, _data[], _length);
static cell AMX_NATIVE_CALL socket_recv(AMX *amx, cell *params)  /* 2 param */
{
    MF_Log("socket_recv():1");
    int socket = params[1];
    int length = params[3];
    int tmp = -1;
    MF_Log("socket_recv():2");
    // First we dynamicly allocate a block of heap memory for recieving our data
    char *tmpchar = new char[length];
    MF_Log("socket_recv():3");
    if(tmpchar == NULL) return -1; // If we didn't got a block, we have to quit here to avoid sigsegv
    // And set it all to 0, because the memory could contain old trash
    MF_Log("socket_recv():4");
    memset(tmpchar, 0, length);
    // Now we recieve
    MF_Log("socket_recv():5");
    tmp = recv(socket, tmpchar, length-1, 0);
    MF_Log("socket_recv():6");
    if (tmp == -1)
    {
        delete [] tmpchar;
        return -1;
    }
    MF_Log("socket_recv():7");
    // And put a copy of our recieved data into amx's string
    tmpchar[tmp]='\0';
    int nlen = 0;
    //int max = params[3];
    int max = length-1;
    const char* src = tmpchar;
    MF_Log("socket_recv():8");
    cell* dest = MF_GetAmxAddr(amx,params[2]);
    MF_Log("socket_recv():9");
    while(max--&&nlen<tmp){
     *dest++ = (cell)*src++;
     nlen++;
    }
    MF_Log("socket_recv():10");
    *dest = 0;
    MF_Log("socket_recv():11");
    // And we need to free up the space to avoid wasting memory
    delete [] tmpchar;
    MF_Log("socket_recv():12");
    // And finnally, return the what recv returnd
    return tmp;
}

Here's the typical output when the error occurs:

Spoiler

L 05/17/2016 - 19:00:38: [SOCKET] socket_recv():1
L 05/17/2016 - 19:00:38: [SOCKET] socket_recv():2
L 05/17/2016 - 19:00:38: [SOCKET] socket_recv():3
L 05/17/2016 - 19:00:38: [SOCKET] socket_recv():4
L 05/17/2016 - 19:00:38: [SOCKET] socket_recv():5
L 05/17/2016 - 19:00:38: [SOCKET] socket_recv():6
L 05/17/2016 - 19:00:38: [SOCKET] socket_recv():7
L 05/17/2016 - 19:00:38: [SOCKET] socket_recv():8
L 05/17/2016 - 19:00:38: [SOCKET] socket_recv():9
L 05/17/2016 - 19:00:38: [SOCKET] socket_recv():10
L 05/17/2016 - 19:00:38: [SOCKET] socket_recv():11
L 05/17/2016 - 19:00:38: [SOCKET] socket_recv():12
Position 3 reached
Position 4 reached
Position 5 reached
Position 6 reached
Native call GetFilename():Start
Native call GetFilename():End
Native call IsFilesizeLarge():Start
Native call IsFilesizeLarge():End
Native call GetFilesize():Start
Native call GetFilesize():End
Native call IsFilesizeLarge():Start
Native call IsFilesizeLarge():End
Native call GetBytesReceived():Start
Native call GetBytesReceived():End
Native call GetNewBytesReceived()
Native call GetFilesize():Start
Native call GetFilesize():End
Native call GetBytesReceived():Start
Native call GetBytesReceived():End
Downloaded data: 1685/-1 (+1685)
native call GetData(string[], 1023)(27, 2)
Datalen: 1023 = min(1023, 1685)
String was set
gBufferPos was increased to 1023
Native call GetNewBytesReceived()
Reading data 1023/1685
native call GetData(string[], 1023)(27, 2)
Datalen: 662 = min(1023, 662)
String was set
gBufferPos was increased to 1685
Native call GetNewBytesReceived()
Reading data 662/1685
native call GetData(string[], 1023)(27, 2)
Position 7 reached
Think Download:End
Think Download:Start
Working with 4
Socket changed
Position 1 reached
Position 2 reached
L 05/17/2016 - 19:00:38: [SOCKET] socket_recv():1
L 05/17/2016 - 19:00:38: [SOCKET] socket_recv():2
*At this point the server crashes and the output stops*

Edit:
I switched out...

char *tmpchar = new char[length];

... to...

char *tmpchar = (char*)malloc(sizeof(char) * length);

...which avoided the crash. But the allocation still fails randomly. It also seems to have increased the frequency of the problem.
I have about 12GB of memory free.