johnm@wdi.disney.com wrote:
michal starke Michal.Starke@lettres.unige.ch wrote:
Playing around for a bit with the morph interface, is also almost guaranteed to get the VM in a closed loop.
Is morphic running correctly on anyone's (intel) linux box [modulo sound] or is this a generalised problem? (I am considering a switch to linux but need to work on morhpic a lot...)
I don't know how I didn't notice this thread!
Here are a couple of landmines for Linux users (maybe this should be on the Swiki somewhere?):
1) Make sure you have a VM that is up to date; there was a time over last summer where new images actually wouldn't work with old VM's. Probably any VM made since September will be fine.
2) If you compiled your own sources, check that sq.h does NOT #define ioLowResMSecs. The default definition is based on clock(), which won't work on Linux. clock() only advances when the process is consuming CPU; thus, using a clock()-based definition on Unix can cause Squeak to hang in times of low CPU usage. If you notice Squeak hanging with 0 CPU usage, this might be the reason. Delete the #define and recompile.
3) Set up sound. If you haven't specifically compiled in sound support, then setting the disableSounds preference is the safest thing. There are combinations of the Unix VM and Squeak image which will hang the image at 100% CPU once sound starts playing. (Though I'd think alt-period would still work in this case). Note that certain morphic operations, in particular bookmorph page turns, cause sounds to be played.
These aside, I have seen a *few* lockups at 100% CPU that I couldn't explain. They all happened over the last couple of weeks, when I was at home over a PPP link instead of over ethernet. "strace" reported that Squeak was doing lots of gettimeofday()'s, for whatever that helps. It always happened after I downloaded my email using Squeak's mail reader. Squeak would lock hard, at 100% CPU, and even alt-period had no visible effect.
But as John mentioned, I worked over a PPP link all of last summer and didn't see any of these unexplainable lockups.... And I'm not seeing them now that I'm back at Tech. It may well have been some change I had added to that image which wasn't so bright after all--I seem to do that a lot! It's hard to say, which is why I hadn't mentioned it before. But I mention it now to be fair.
I guess the main point here is that Squeak is certainly being used on Linux. It's not some old dusty proof of concept that has been forgotten. If you're seeing problems, there are people who would like a stab at fixing them.
Lex
On Mon, 4 Jan 1999, Lex Spoon wrote:
johnm@wdi.disney.com wrote:
michal starke Michal.Starke@lettres.unige.ch wrote:
Playing around for a bit with the morph interface, is also almost guaranteed to get the VM in a closed loop.
Is morphic running correctly on anyone's (intel) linux box [modulo sound] or is this a generalised problem? (I am considering a switch to linux but need to work on morhpic a lot...)
Here are a couple of landmines for Linux users (maybe this should be on the Swiki somewhere?): [...]
These aside, I have seen a *few* lockups at 100% CPU that I couldn't explain. They all happened over the last couple of weeks, when I was at home over a PPP link instead of over ethernet. "strace" reported that Squeak was doing lots of gettimeofday()'s, for whatever that helps. It always happened after I downloaded my email using Squeak's mail reader. Squeak would lock hard, at 100% CPU, and even alt-period had no visible effect.
The 100%-CPU-GUI-lockup-problem (with alt-. not responding) occurs every other week or so in my Swiki. No Morphic, no PPP, no Sound, just plain Squeak 2.2 / 2.3b. The interesting thing is, nobody outside notices it because the *server is still serving*.
IMHO it's a network/socket/*ConnectionQueue* problem that occurs at some rare conditions. I think it could be solved with accepting sockets - *please* put it into 2.3 (even if the Mac would implement it using the ConnectionQueue stuff :-)
Cheers, Bert
At 5:04 AM -0500 1/5/99, Bert Freudenberg wrote:
[snip]
The 100%-CPU-GUI-lockup-problem (with alt-. not responding) occurs every other week or so in my Swiki. No Morphic, no PPP, no Sound, just plain Squeak 2.2 / 2.3b. The interesting thing is, nobody outside notices it because the *server is still serving*.
IMHO it's a network/socket/*ConnectionQueue* problem that occurs at some rare conditions. I think it could be solved with accepting sockets - *please* put it into 2.3 (even if the Mac would implement it using the ConnectionQueue stuff :-)
I suspect that this is the ListenLoop bug that was thrashed out in Nov on the PWS list. A fix is available, though I don't know if it got into 2.3 yet (John Maloney would know...John?)
The problem (I fuzzily recall) was that ListenLoop was trying to reuse a Socket that plain didn't exist so you ended up in a tight high priority loop.
The final solution was to kill the Socket object.
I'm not clear if this is in fact your problem, but it may be connected, so to speak. Check your version of ConnectionQueue>>ListenLoop. It should be of 11/10/98.
Cheers, Bijan Parsia
squeak-dev@lists.squeakfoundation.org