GNOME Bugzilla – Bug 161525
Screenshot application complains that it cannot save the screenshot file
Last modified: 2006-12-22 12:42:25 UTC
When one tries to capture a screenshot under GNOME 2.9, the following error is thrown: Unable to save the screenshot to disk: (null) I looked at the code in gnome-utils, and I think there is a logic bug. The attached patch corrects the problem, and seems pretty correct to me.
Created attachment 34928 [details] [review] Allow screenshot application to save screenshots
That patch doesn't make sense to me. If we have an error message on the pipe, then we have an error. We assume that a save occurred iff the pipe just closes. Can you see what the error is by putting a print statement near the write() on line 184?
That's just it, nothing is written to that pipe. However, G_IO_IN is being set when pipe_from_child[1] is closed. If read_pipe_from_child() is changed so as not to display the dialog and exit if error_message is NULL, then gnome-screenshot will work again. Alternatively, since G_IO_HUP is also set when the pipe is closed, the code could also be modified to check to see if G_IO_HUP and G_IO_IN are set, and if so, then don't print the error dialog and exit.
Why is G_IO_IN set there? I'm not getting that here, just the HUP. I'll double check on Monday why it's set, but in the interim here's a patch that checks the result of g_io_channel_read_line(). It should hopefully fix the problem you had. This is on freebsd, right?
Created attachment 34949 [details] [review] Quick patch to check errors.
Yeah, this is on FreeBSD 5-STABLE with glib-2.5.7. I haven't updated to 2.6.0 yet on this machine.
So is the patch committed?
not yet. I didn't get to it over Xmas as it merged with a larger patch of mine. I'll try to it this week.
Note: I did some more testing with this since gnome-nettool had a similar problem. All BSDs set POLLIN along with POLLHUP on a pipe EOF. MacOS X doesn't set POLLHUP on EOF, but rather just POLLIN, and read() will return 0 to indicate EOF. Therefore, all G_IO_IN tests should probably check for EOF, and cleanup appropriately.
Prolly should update the glib docs to reflect this.
This is working fine for me with the latest version of screenshot 2.9.4 and the version of GNOME i am using is 2.9(cvs head) built on 3rd Jan 2005. The Operating System i am using is Novell Linux Desktop.
Linux and Solaris are not affected by this. Only *BSD and MacOS X will be affected.
Now that gnome-utils has branched for 2.10, I hope the larger patch can be committed ;-)
maybe gnome bugzilla bug #325708 is related to this? http://bugzilla.gnome.org/show_bug.cgi?id=325708
tino (comment #14): nope, this bug affects *BSD. jrb: was the patch committed? can we close this bug?
jonathan, joe, what's the state of this? patch has not been committed to CVS it seems - can this get in?
I am still using this patch in our local ports repository. I would like to see it committed, but I have no say in that.
Something new on this ? I still see this issue.
Ditto I am seeing this issue also Running Gnome 2.16 on Gentoo Linux x86 (stable) seems this bug's pervading other archs... where do I find the patch to see if it fixes the problem here?
the problem is also on on my ubuntu (linux 2.6.19) box since atleast gnome-utils 2.17 The read_pipe_from_child() calls: g_io_channel_read_line() and it returns G_IO_STATUS_EOF on linux too as far as I can see it does not send a IO_HUP at all. the read_pipe_from_child() is called after close opertion on the child filedescriptor (screenshot-save.c:206) I dont know why this has worked before because it is a bug. check the return status of g_io_channel_read_line() should fix the problem.
danny, the patch is attached here at comment #5: http://bugzilla.gnome.org/attachment.cgi?id=34949
For what is worth, patch attached on Comment #5 fixed the issue for me too. Using gnome-utils 2.17.
quick patch from comment #5 also worked for me on gnome-utils 2.16.2 cheers danny
thanks to all BSDers who tested the patch. should be fixed in HEAD; if my calendar is right, there should be a 2.16 point release in january: will backport the patch to gnome-2-16.
committed to gnome-2-16, hence closing as FIXED.