| IRC logs at
Set by AlexDaniel on 12 June 2018.
00:03 patrickz joined 00:06 patrickb left 00:24 Kaiepi left 00:25 Kaiepi joined 00:40 patrickz left 01:01 ZzZombo joined 02:23 lucasb left
colomon Is there any current-ish Raku literate programming module? Perl6::Literate claims to not be very serious and is years out of date, the better one that README references doesn’t look like it’s been touched in a decade… 04:16
05:40 ZzZombo_ joined 05:42 ZzZombo__ joined, ZzZombo__ left 05:43 ZzZombo left, ZzZombo__ joined 05:45 ZzZombo_ left 06:09 ZzZombo__ is now known as ZzZombo 07:02 domidumont joined 07:11 domidumont left 07:35 domidumont joined 08:08 Kaiepi left 08:16 Kaiepi joined 08:20 patrickb joined
nine jnthn: what you described was the original bug. I could easily see that in rr which showed me that deopt was clearing spesh_cand and effective_spesh_slots. This time the current frame never had those. 08:22
tellable6 2019-12-11T23:54:00Z #raku-dev <patrickb> nine Never mind.
hey nine, you have a message:
08:30 zakharyas joined 09:48 sena_kun joined 10:11 zakharyas left 10:22 zakharyas joined 10:54 sena_kun left 11:08 sena_kun joined 11:15 domidumont left 12:12 zakharyas left 12:50 domidumont joined 12:51 domidumont left, domidumont joined 12:54 sena_kun left 13:09 sena_kun joined 13:41 zakharyas joined 14:06 lucasb joined 14:54 sena_kun left 15:08 sena_kun joined 15:15 nine left 15:16 nine joined 15:19 camelia joined 15:57 domidumont left 15:59 domidumont joined 16:16 patrickb left 16:22 zakharyas left 16:23 zakharyas joined 16:31 domidumont left 16:43 ZzZombo_ joined 16:44 ZzZombo left, ZzZombo_ is now known as ZzZombo 16:48 robertle joined 16:53 sena_kun left 16:59 ZzZombo left 17:00 ZzZombo joined 17:08 sena_kun joined 18:19 domidumont joined 18:29 domidumont left 18:34 domidumont joined 18:39 zakharyas left 18:42 patrickb joined 18:54 sena_kun left 19:09 sena_kun joined 19:14 domidumont left 19:37 domidumont joined 19:50 domidumont left 20:42 robertle left
nine Oh, the frame got moved by the GC! That's why the watch point for its spesh_cand doesn't trigger in reverse-cont 20:44
And yes, it's deopt all again. So it seems like I need to do similar fixes for JIT compiled frames as I've already done for interpreted ones 20:46
20:55 sena_kun left
nine So, how does this work in the first place when we're in a JIT compiled frame and we invoked something and that causes a deopt all. How does the JIT compiled code know to exit to the interp? 20:59
21:08 sena_kun joined
nine Well, a deoptall on an inactive frame may only happen during an invoke which already has a lot of special handling in the JIT. Handling which nativecallinvoke lacks. 21:26
So I guess what's needed is for the JIT to generate code that after a nativecallinvoke checks if the frame actually still has a spesh_cand and exit to the interp if not, i.e. a somewhat manual deopt. 21:28
I really wonder if the same issue exists with nativeinvoke_o. It should 21:31
I still don't know how exactly the JIT returns to the interpreter after a deopt in an invoke. But at least I have an idea on what's missing, even if I don't know yet how exactly that needs to look like. 21:36
21:48 sena_kun left 23:08 vrurg left, vrurg joined