Thibault Monsel

Hello, Im back with some updates ! 1/ I used the pre-commit hooks but still have one small spurious change in `terms.py` :astonished: . 1/ `y0` and `y0_history` are disambiguated...

Latest commit does what you suggested - Reverted back to not touching the `state.tnext` and such. - Integrated your bissection code into the discontinuity detection. - Only search for discontinuities...

Latest commit has : - added a `discont_update` in the discontinuity checking part to make sure we correctly update `state.discontinuity` - added a `discont_check` argument in `diffeqsolve` to check for...

Latest commit updates some remarks of code comment after latest review https://github.com/patrick-kidger/diffrax/pull/169#pullrequestreview-1159470301. I've bundled together as you said the delays term together for an easier API. ```py class _Delays(eqx.Module): delays:...

> Okay, let's focus on this bit before we move on to discussing the discontinuity handling below. I'll leave you to make the changes already discussed here. Let me know...

Latest commit should have handle all of the issues mentionned above. - Not do an extra step after doing the implicit step by integrating de-facto the explicit step in the...

Ok, this makes sense, so i'll take back what I said in my bottleneck "Discussion/Bottleneck for implicit step", thanks for the clarification ! The `_pred` condition should be on the...

Relevant changes are made : - All the code is now managed with `PyTree` operations (i think) - Moved `Delays` and implicit step into a new file (but there is...

Great news ! Im not familiar with the wiring of the `implicit_jvp` up for some pointers. Other then that relevant changes were made

- Relevant changes were made for the last code block, everything is in `delays.py` now ! - Added 2 new attributes for `Delays` : `nb_sub_intervals` and `max_steps` - Lots of...