On 5 May 2014 17:48, David T. Lewis lewis@mail.msen.com wrote:
On Sat, May 03, 2014 at 02:01:57PM -0400, David T. Lewis wrote:
On Sat, May 03, 2014 at 11:18:11AM -0500, Chris Muller wrote:
On Fri, May 2, 2014 at 4:38 PM, David T. Lewis lewis@mail.msen.com wrote:
On Thu, Apr 10, 2014 at 10:06 AM, David T. Lewis lewis@mail.msen.com wrote:
> On Wed, 09 Apr 2014 16:52:54 -0700 David T. Lewis lewis@mail.msen.com wrote: >> >> Create a new class, attempt to edit the class comment ==> Error: >> RemoteString past end of file >> >> Surely I cannot be the first person to have edited a class comment in >> the >> last month or so ... I hope. >> > > WriteStream>>nextChunkPut > > add a self flush at the end will fix it. > > Discussion several months ago fingered the gcc 4 compiler so something. > > cheers. > > tty >
Ah, right, thank you! I do recall the discussion now that you mention it. I use VMs compiled on my own machine, which happens to be a new Ubuntu system at the moment. It probably has the buggy gcc library.
Thanks a lot, Dave
Okay, I was just getting this again in our release image. Strange -- it happens on one laptop but not the other..! The suggested patch seemed to work for me, don't we need to push this fix to trunk and squeak45?
I should double check to be sure, but I think this was a bug in the run time libraries of the operating system, as opposed to something in the VM or the image. I can't say I like the idea of putting a kludge into the image to work around a bug in Ubuntu, although it might be worthwhile in this case just to prevent new users from having a bad experience.
Come to think of it, we should be able to write a unit test that would detect this problem when running Squeak on a buggy OS.
Dave
Strange I encountered the bug on Ubuntu 14.04 instead of 12.04 instead of the other way around.. Actually not, Ubuntu seems to be getting worse in some ways..
I am having a hard time reducing this problem to something I can put into a unit test. I can't see any obvious problem in the way the writes are being performed on a file stream, and the actual failure seems to be happening when creating the initial RemoteStream for a class comment. When a new class is created without a class comment, the classComment in the ClassOrganizer is nil. When accepting a new comment in a browser for the first time, a RemoteStream is created to reference the comment that will be saved in the changes file. Something gets out of whack at that point, but I have not yet spotted the cause.
I pushed the fix^h^h^hworkaround for this to the inbox. I bothers me a bit that I can't come up with a reproduceable test case for this, but the change seems harmless.
This is the same issue that cropped up just after 4.4's release, right? IIRC I patched trunk & 4.4, but then soon reverted the workarounds because the community considered the workaround a Bad Thing (tm), possibly making the proper fix more difficult to find.
I'm happy if folks have since changed their mind.
If no objections, this will go into trunk.
Not from me.
frank
Dave