Re: Bug in GNU Parted

[Wouter Coene]

According to Andrew Clausen (address@hidden):
[snip]
> > I've submitted this patch before, but it was refused claiming this
> > condition couldn't happen if parted was programmed according to the API
> > specification.
> 
> Exactly.  I want to find out why Parted "isn't programmed
> according to the API".

Well, the code itself strikes me as rather odd, too. In version 1.4.9 (not
the latest version, but I don't believe that part of the code has changed
much lately, plus I don't have access to any newer versions right now),
ped_device_close() (libparted/device.c:812) calls _do_refresh_close()
(libparted/device.c:738) while decrementing dev->open_count. However, the
contents of _do_refresh_close() is a mere:

        static int
        _do_refresh_close (PedDevice* dev)
        {
                return 1;
        }

I encountered this problem in a program I worked on some time ago too, but
my patch (which is still in the version of parted I use to build said
program) solved that nicely.

I think it has something to do with dev->open_count being only incremented
on the opening of the device by the disk layer, and that somehow right after
the parted command-line editor is started, unlike the device layer, the disk
layer remains uninitialized until a command is executed.

This would also mean that, since ped_disk_close() closes the device (and
thus decrements dev->open_count), there must be a small memory leak in the
parted command line editor at program termination, caused by
ped_disk_close() not being called for every opened disk "object", otherwise
this problem would have been noticed earlier.

Wouter

ps: sorry about my previous post (which was rather rude I believe). I think
    I'll blame the weather (hot and humid) for that one.
-- 
* Wouter Coene * C/PHP/TCL * address@hidden *
 *** I do not speak for my employer until I say I do ***
  ** "First they ignore you. Then they laugh at you. **
   ** Then they fight you. Then you win." -- Gandhi **