Hi: I am triying to allocate a n*n matrix of a type defined structure of data. The pd external object example I'm sending attached is very simple: each time it receives any message except the "clear" message it increments a counter and reallocate a matrix of n*n members where n is the number stored in the counter (so, 1, 4, 9, 16 and so on). If it receives the "clear" message, it reset the counter to 1 and reallocate the matrix to 1x1. The fact is that the object crashed when reallocating and I can't be able to know what am I doing wrong. I use the traditional "realloc" function. Any help will be welcome. Best Isi
Isidro Gonzalez wrote:
Hi: I am triying to allocate a n*n matrix of a type defined structure of data. The pd external object example I'm sending attached is very simple: each time it receives any message except the "clear" message it increments a counter and reallocate a matrix of n*n members where n is the number stored in the counter (so, 1, 4, 9, 16 and so on). If it receives the "clear" message, it reset the counter to 1 and reallocate the matrix to 1x1. The fact is that the object crashed when reallocating and I can't be able to know what am I doing wrong. I use the traditional "realloc" function. Any help will be welcome.
realloc does no initialization of data.
In the for loop in the dummy_alloc function you are calling realloc on uninitialized pointers. The fact that this ever fails to crash is because you are by chance getting NULL pointers from the realloc when you increase the size of a.
You need to manually set the newly acquired bytes in a to be NULL before you can safely call realloc with them as arguments.
Many thanks for your kind answer. I´m surprised because the way I use to allocate memory is the most commonly used. Wouldn´t initializing the pointers to NULL destroy all the contents of previous memory they hold, if any...? So, why to use "realloc" this way, then? I would use "free" and "malloc", instead. Your advice works fine but when I try to use the memory, the object makes PD to crash. Any further ideas? Many thanks Isi
--- On Tue, 10/13/09, Justin Glenn Smith noisesmith@gmail.com wrote:
From: Justin Glenn Smith noisesmith@gmail.com Subject: Re: [PD-dev] Memory reallocation problem To: "Isidro Gonzalez" isidro_gonsoy@yahoo.com Cc: pd-dev@iem.at Date: Tuesday, October 13, 2009, 4:10 PM Isidro Gonzalez wrote:
Hi: I am triying to allocate a n*n matrix of a type
defined structure of data.
The pd external object example I'm sending attached is
very simple: each time it receives any message except the "clear" message it increments a counter and reallocate a matrix of n*n members where n is the number stored in the counter (so, 1, 4, 9, 16 and so on). If it receives the "clear" message, it reset the counter to 1 and reallocate the matrix to 1x1.
The fact is that the object crashed when reallocating
and I can't be able to know what am I doing wrong. I use the traditional "realloc" function.
Any help will be welcome.
realloc does no initialization of data.
In the for loop in the dummy_alloc function you are calling realloc on uninitialized pointers. The fact that this ever fails to crash is because you are by chance getting NULL pointers from the realloc when you increase the size of a.
You need to manually set the newly acquired bytes in a to be NULL before you can safely call realloc with them as arguments.
Isidro Gonzalez wrote:
Many thanks for your kind answer. IŽm surprised because the way I use to allocate memory is the most commonly used. WouldnŽt initializing the pointers to NULL destroy all the contents of previous memory they hold, if any...? So, why to use "realloc" this way, then? I would use "free" and "malloc", instead. Your advice works fine but when I try to use the memory, the object makes PD to crash. Any further ideas?
The issue is that you need to initialize the data that realloc gives you, but not touch the data you already had.
Using realloc is common enough, but you cannot count on the value of the memory returned by realloc, it will be NULL much of the time, but can also have random data in it. If you pass a NULL pointer to a call to realloc you get a fresh patch of memory, if you send a random uninitialized pointer that did not happen to be NULL, that is where you are getting your segmentation violation.
This should be simple enough to fix if you use an extra variable to store the previous dimensions of a, and then initialize your loop variables accordingly.
I also recommend using the pd library functions getbytes and resizebytes, which work like malloc and realloc except you need to track the size of objects they return yourself, and I believe they are designed to prevent hiccups in audio when used realtime.
-
Isidro Gonzalez -
Justin Glenn Smith