[Geany-devel] multiple instance save-settings - Re: Race condition when saving geany.conf

[Enrico Tröger]

On Fri, 26 Feb 2010 15:15:02 +0000, Nick wrote:

>On Wed, 24 Feb 2010 12:04:20 +0300
>Eugene Arshinov <earshinov at gmail.com> wrote:
>
>> Hi all.
>> 
>> When several instances of Geany quit in the same time, there is a
>> high possibility of a conflict. I can reproduce it easily on my
>> machine, using either trunk or SM version.
>> 
>> To reproduce: open three instances of geany, "geany", "geany -i" and
>> another "geany" (absence of file names implies -i automatically in
>> this case). It would be better to open three different files in
>> the instances, to distinguish them. Then logout or reboot without
>> quitting geany manually. On my machine, after I (in case of
>> trunk) or SM code (in case of SM) restart geany, the default session
>> is always cleared. Expected behaviour: the default session is
>> managed by the first of the three instances and contains the files,
>> which were opened in that instance, after restart.
>> 
>> I can see two solutions for this problem. The first is an
>> additional POSIX process-shared semaphore / mutex for Windows to
>> guard geany.conf. This should eliminate the problem completely.
>> AFAIK, there are no wrappers for process synchronization primitives
>> in GLib, so I'll need to write a thin wrapper myself.
>> 
>> The second option is to change the behaviour of "new instances". If
>> such an instance (#1) detects a "main instance" (#2) running, it
>> should not touch geany.conf. Actually, to deal with the described
>> issue, it is enough to implement this behaviour only when #1 tries
>> to save geany.conf while quitting.
>> 
>> The second option is easier to write as it does not require
>> additional synchronization primitives and it's possible to reuse the
>> code of socket.c. Actually, I already have this option implemented,
>> to check whether it indeed solves the problem. But, you see, this
>> solution can't prevent the race condition completely, in distinction
>> from the first solution. Moreover, some of you may consider the
>> second solution "hackish", which is enough to decline it.
>> 
>> So, the first solution is right, but the second is easy :-) What do
>> you think?
>
>Perhaps without clearly answering you, here's my non-technical
>thoughts:
>
>To be easy to implement, perhaps maybe only the first/main instance
>should save settings. We could have a Tools->Save Config menu item
>enabled for the first instance to allow the user to restart other
>instances with the new settings.

I completely agree here.
As said in another reply, file locks/semaphores are most pobably
overkill here.
Just make the first instance the master (by definition) and this
instance is the only one which writes the config. All other instances
may read but must not write.
This is the easy implementation but also the most logical to me, more
logical than 'last instance wins' especially in terms of closing
multiple instances on system logout where 'last' is quite randomly as
anyone said in another reply.
If the master setting works correctly, we don't need to fiddle with
semaphores, they would be only an additional tool to enforce the master
instance but IMO it's not worth.


Regards,
Enrico

-- 
Get my GPG key from http://www.uvena.de/pub.asc
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 198 bytes
Desc: not available
URL: <http://lists.geany.org/pipermail/devel/attachments/20100228/178862ee/attachment.pgp>