Re: Still confused on new thread starting idle functions to update UI.

[Bernhard Schuster <schuster bernhard gmail com>]

> The observed bad behavior is that sometimes messages don't appear at 
> all, or appear twice in my textview. It's quite unpredictable. Sounds 
> like a race condition.
>
> I spawn a thread from the main program - and from now on I don't talk 
> about main - and this new thread prepares messages in a character 
> string. The thread then passes a pointer to the character string to 
> an idle function so that the idle function can update a textview in 
> the UI. When done, the idle function stops itself by returning a 
> G_SOURCE_REMOVE boolean.

That sounds sane so far.

> The thread allocates memory for the character string, and the idle 
> function does not free it. Instead, the thread free's the memory just 
> before it stops. It waits a bit to make sure the idle function has 
> finished using the memory containing the message.

That does not sound sane. This actually could create a race condition, 
you never know which thread get scheduled when.

I'd recommend you to use a GAsyncQueue, push a ref counted (not 
necessary in your case here) object (i.e. GByteArray or a custom ref 
counted version of GString) into it.

Keep the idle function until the queue is empty (+deref/destroy the 
char array/GByteArray/GString).

> But I lied...
>
> Because there are several messages (roughly 30), the thread actually 
> allocates memory for an *array* of character strings. An index is 
> then used to specify which one we're using.
>
> This may seem awkward and unnecessary, but if I just use a single 
> character string, it is possible for the thread to replace the 
> contents of the string with the next message while an idle function 
> is still working with the previous message.
‘3’

This is not deterministic, you are trying to rely on probabilistic 
behaviour and that the mainloop is able to process a certain amount x 
before your thread shoots again.

> What I had before which worked was a single character string and a 
> sleep (1) after each g_idle_add(), but that made things very sluggish.

Indeed, do not use sleep in order to fix an already broken behaviour 
(this is not a rule of thumb, this is a do never ever thing).

<snip>...</snip>

> I would've liked to pass the message string to the idle function by 
> value, because then it would only work with its own ephemeral copy of 
> the string, but g_idle_add() only allows me to pass a pointer to 
> data, not the data itself.

Use a refcount char array "container" (like GByteArray and GString is).


Hope I could help you, keep asking if something is not clear or if you 
are uncertain how to do implement things properly.


Best

Bernhard