> Am 06.02.2019 um 18:34 schrieb Gregor Thalhammer > <gregor.thalham...@gmail.com>: > > > >> Am 06.02.2019 um 16:09 schrieb Andreas Kloeckner <li...@informa.tiker.net>: >> >> Hi Jerome, >> >> Jerome Kieffer <jerome.kief...@esrf.fr> writes: >>> We encounter a bug in pyFAI which looks related to pyopencl & pybind11 ... >>> https://github.com/silx-kit/pyFAI/issues/1131 >>> >>> But this is a "rare" bug so we are struggling in reproduce it. >> >> Gregor Thalhammer (cc'd) wrote last October to report a similar >> phenomenon (and even had a notebook that appeared to reproduce his >> issue). Unfortunately, I haven't had much time to debug, and I haven't >> encountered this in my work. Here's his message: >> >> https://lists.tiker.net/hyperkitty/list/pyopencl@tiker.net/message/3TWCVHBB4MZGC42WZTOQ6NPBVWBIG34X/ > > Very short update on this issue I also encountered: > * I am almost sure this is a bug in pybind11, not of pyopencl itself, related > to deallocating objects by the garbage collector > * Possibly pybind11 is overly cautious and triggers an program exit where it > should just gracefully ignore double reallocation > * workaround (quick and dirty): disable the garbage collector > > Best > Gregor
Another update, I managed to create a script (not a notebook) that crashes often. This should be easier to debug than a notebook kernel. Important ingredients appear to be * more than one thread * garbage collection https://gist.github.com/geggo/d9b4016ff5b77dab262be7b30a98d862 best Gregor >> >> Could you create an issue on Github to collect evidence and related info? >> I'd also be grateful for any help in debugging. >> >> Andreas
_______________________________________________ PyOpenCL mailing list -- pyopencl@tiker.net To unsubscribe send an email to pyopencl-le...@tiker.net
