async headaches

[muppet <scott asofyet org>]

had a devil of a time solving a race condition problem.  i'd like your 
opinions on what might be a better way to solve this one.


---


i'm creating a file open dialog specifically for images, which shows 
some image properties and a thumbnail for each image the user selects in 
the file list.  i'm using imagemagick's convert and some home-grown 
image loading display code for various reasons that aren't all that 
relevant.

to the "select-row" signal of the file CList i bound a callback which 
would attempt to load and preview the image.  so far so good.  each time 
it runs, it forks a call to convert (via piped open), so there is no 
caching of thumbnails (though given time i would like to cache them, but 
the regeneration logic is more than i want to deal with right now).

unfortunately, since the loading of the image takes long enough that the 
double-click to open feature is now inoperative...  the user has to 
select the file, wait for the preview to finish loading, and *then* 
double click.  this is uncool.

so, thinking myself rather clever, i changed the "select-row" callback 
to instead install a timeout for a quarter second later to load the 
image.  then after the user clears the dialog, i remove the timeout, 
clean up, and go on my merry way.

except that the program now segfaults because the timeout event got put 
on the event queue before i had a chance to remove the timeout, and it's 
trying to load an image into a buffer that's already been destroyed.

after a bit of mucking about, i added a field called "dying" to the 
parent-to-callback communication hash.  if the timeout handler sees the 
"dying" flag set it just doesn't do anything.  this hack relies on 
perl's reference counting to make sure the communication hash even 
sticks around long enough to to this job.


this seems to work, but also feels unclean and unsafe.


am i going about this all the wrong way?  would simply lengthening the 
delay solve the problem?  would relying on access and modification times 
be sufficient for knowing when to regenerate thumbnails?