Max Schwarz

Interestingly, just running latexrun again fixes the problem for me - even though it does not execute latex again (`--verbose-cmds` does not print anything)

After running latexrun once, `latex.out/xyz.pdf` is correct, but `xyz.pdf` is some intermediate version, not the final one... I uploaded a log here: https://gist.github.com/xqms/610a018e95251bcbc9b80c5927a900f5 The command I used was `latexrun --debug...

> So, given that the extra GL contexts should be windowless, with either a trivial (32x32, as GLX/EGL does) or a completely nonexistent (CGL) default framebuffer, I'd say one way...

> TIL about nvJPEG, thanks! Too bad it's tied to CUDA... I wonder how would that compare to having ASTC-encoded images as the input instead, for example. Hmm, nice idea!...

Hm, I guess leaving the flag there even if it has no impact right now is OK. But I don't have a strong opinion on this :sweat_smile:

I also like the modularization and customization aspects of the current setup! (but I have also seen how complex it is to people new to Magnum). > Have magnum repos...

> @xqms you said you have CUDA interop implemented on your side as well, how do you do that? Or was it just a side-effect of the PyTorch integration? :)...

@gonultasbu The question is if it is possible to load meshes directly from memory, without opening a file. AFAIK it's possible to populate ASSIMP structures "by hand". So a patch...

> The shutting_down_mutex_ is protecting the access to the shutting_down_ flag. And in the creation methods as well as the shutdown method it keeps that lock until the method has...

Okay, that's what I expected. We have the same situation on the other side in `TransportTCP` then. A comment on the mutex about what it is actually protecting might be...