Re: [PATCH 6/9] hashmap: Add re-entrancy support to foreach function

[Denis Kenzior <denkenz@gmail.com>]

[-- Attachment #1: Type: text/plain, Size: 1440 bytes --]

Hi Jukka,

On 02/10/2015 08:42 AM, Jukka Rissanen wrote:
> Allow user to remove and add values while inside a callback
> from foreach function. This is very much needed as it is
> easy to get things wrong here and it helps to avoid complicated
> call flows in calling applications.
>

Can you tell me why this is needed?  This sounds like abuse of 
hashmap_foreach and an alternate data structure might be in order.

> So foreach logic is changed so that if user tries to remove
> an entry from the hash while inside foreach callback, the
> entry is not yet removed from hash but marked as removable.
> After foreach has finished calling the callback function,
> it checks what elements it needs to remove from the hash.
>

So let me politely say: "No way are we doing this" ;)

> This valgrind report was seen and it relates to this issue.
> Here a DBus message was received and its callback function
> eventually called l_dbus_unregister() because of application
> logic. This then caused the signal_list hash to get corrupted
> because the unregister removed an entry from hash while
> traversing it.
>

So I think I understand why you're doing this...

l_dbus_register and l_dbus_unregister need to actually go away.  They 
are not workable long term.  We need a way smarter data structure to 
handle signal filtering.

Look at libsystemd/sd-bus/bus-match.c for some inspiration.

Regards,
-Denis