Am 12.10.2016 um 01:31 schrieb Waldek Hebisch: > Kurt Pagani wrote: >> >> Am 11.10.2016 um 04:39 schrieb Waldek Hebisch: >>> >>> I am affraid there are dragons here. IIUC you use FriCAS functions >>> in two different threads. Currently FriCAS is not thread safe >>> and doing this may lead to troubles. More precisely, for >>> performance FriCAS depends on various global caches. Updating >>> them simultanously from independent threads may lead to corruption. >>> This is worst kind of bugs... >>> >> >> At first I also was afraid of what you described above. And yet I couldn't >> avoid >> doing some extensive tests. Apparently the webserver does not interfere in >> any >> way with Fricas (thanks to the bordeaux-threads I suppose). I fired over >> 1000 >> queries (automated). Surprisingly enough it also works when starting the >> server >> in a interpreter window (-iw) and one can even interact with the current >> session >> although |parseAndEvalToString| is used only (it beats me how this is >> possible). >> > > Well, I hope that threads work correctly. Given that independent > threads should execute fine. The problem is with simultanous > computations in main FriCAS thread and access to FriCAS functions > from the webserver.
Definitely. I'm expecting some irregularities sooner or later. Since I haven't a clear picture of the role parseAndEval... (in server.boot) plays in the main thread I probably have to dig deeper. The problems you addressed actually occur in the zmq kernel when it's loaded in interpreter mode (that's why it only works if fricas is used as a lisp library, so to say). > > I wrote that releated bugs are nasty. Main thing is that there > are timing dependencies. It may work fine on your machine and > fail on slightly different machine. Frequently probability > of hitting bug is quite low. You may be tempted to declare > that code is good enough -- far from ... > if there is a bug but nobody > notices does it matter? The trouble is that a priori you do > not know if and when bug will bite -- the amount of testing > needed to get any confidence is prohibitive. > Without doubt. If it failed early I would have kept my hands off, now it interests me. -- You received this message because you are subscribed to the Google Groups "FriCAS - computer algebra system" group. To unsubscribe from this group and stop receiving emails from it, send an email to fricas-devel+unsubscr...@googlegroups.com. To post to this group, send email to fricas-devel@googlegroups.com. Visit this group at https://groups.google.com/group/fricas-devel. For more options, visit https://groups.google.com/d/optout.
