Re: [gmime-devel] GObject unref hell

[Jeffrey Stedfast <fejj gnome org>]

On 11/6/2013 4:21 AM, Mario Theodoridis wrote:
> Hello,
>
> i'm currently in the process of using GMime for a mail process 
> application. First of all i'd like to thank Jeffrey and potential 
> other contributors to GMime for writing this wonderful library, as 
> MIME parsing and generating is everything but fun.
>
> I find myself with a problem that i can't seem to wrap my head around.
> The following code snippet deterministically works in some instances, 
> but not in others.
>
> #define rfunref(obj) if (obj) { if G_IS_OBJECT(obj) { \
>     g_debug("[%s:%d] unref "#obj, __FILE__, __LINE__); \
>     g_object_unref(obj); obj = NULL; \
>     g_debug("[%s:%d] unrefed "#obj, __FILE__, __LINE__); }}
>
>   GMimeDataWrapper *content = NULL;
>   GMimeStream *mem = NULL;
>   GMimePart *part = NULL;
>   GMimeMultipart *multipart = NULL;
>   char *fileName = NULL;
>   char *type = NULL, *subtype = NULL, *param = NULL, *value = NULL;
>
>   do{
>     /* other stuff here */
>
>     /* set type, subtype, filename, etc. */
>     /* find type */
>     getType(msg, res, resLen, &type, &subtype, &param, &value);
>     part = g_mime_part_new_with_type(type, subtype);
>     g_mime_part_set_filename(part, fileName);
>     g_mime_object_set_content_type_parameter((GMimeObject*)part,
>                                               param, value);
>     if (g_ascii_strcasecmp(type, "text") == 0) {
>         g_mime_part_set_content_encoding(
>                 part, GMIME_CONTENT_ENCODING_QUOTEDPRINTABLE);
>     } else {
>         g_mime_part_set_content_encoding (
>                 part, GMIME_CONTENT_ENCODING_BASE64);
>     }
>     g_mime_part_set_content_object (part, content);
>     rfunref (content);
>
>     g_mime_multipart_add (multipart, (GMimeObject *) part);

where does multipart get set?

> rfunref (part);
>   } while(attachments);
>
>   g_mime_message_set_mime_part (message, (GMimeObject*) multipart);
>   headerList = g_mime_object_get_header_list((GMimeObject*)multipart);
>   rfunref (multipart);
> [snip]
>
> The unref macro is a mere act of desperation.
>
> I have 2 general questions here.
> 1. What is going on here. No code was skipped between unref part and 
> multipart.

Hard to say without knowing where multipart is getting set.

> 2. Is there a way to safely unref a GObject, i.e. unref it if it needs 
> to be? This is mainly because it is not always clear to me when and 
> when not due to issues like the current.

No, but as a general set of rules:

1. if you create a new object and set it as a child on another object, 
you need to unref it
2. when you get an object from another object, you don't - think if it 
as returning a const pointer.

example of rule #1:

child = g_mime_child_new ();
g_mime_parent_add_child (parent, child);
g_object_unref (child);

The reason you unref here is because the parent adds a ref to the child 
when you add it.

Essentially you own a ref to child, and, once you've added the child to 
the parent, the parent *also* owns a ref to the child. Thus there are 2 
references to it, but if you are done with the child, you need to unref 
it (or you leak a reference to it).

example of rule #2:

child = g_mime_parent_get_child (parent);
// no need to unref child because the parent does not ref the child when 
it returns it

// to extend this example further...
g_mime_parent_add_child (parent2, child);
// still don't unref the child because you never ref'd it

In this example, when you get the child from the parent, only the parent 
owns a reference to the child. When you add it to parent2, parent2 adds 
another reference, which means that now both parent objects own a ref to 
the child, but since you still don't own a reference to the child, you 
don't unref it.


Hope that helps,

Jeff

Attachment: smime.p7s
Description: S/MIME Cryptographic Signature