You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
@wds15 I was thinking about that, but usually rstan lags behind and boost 1.82 has been out for a minute. My guess is that BH will have updated by the time rstan is ready to use this version. Does that make sense?
I'm fine with just bumping to 1.81, but feels like we will end up bumping to 1.82 soon after
If you have a good reason to go to 1.82 ... then I'd be with you. Otherwise just go with 1.81 is what I would say. From interacting with the maintainer of BH I understood that he intends to do once per year updates only. This is due to the large effort of upgrading (such that my bet would be at 1.83 or even 1.84 to land on CRAN as BH). In any case, I do understand that rstan is so much behind such that it is almost hard to argue for compatibility with it. Given that you have already invested in an 1.82 PR maybe we should just go with it and simply discourage usage of 1.82 only features?
Given that you have already invested in an 1.82 PR
I just let your script run and made one fix so this was very low effort on my part
@wds15@WardBrian the unordered flat map is in 1.81, but 1.82 has a major update to the unordered containers which says
Fixed bug in internal emplace impl where stack-local types were not properly constructed using the allocator of the container which breaks uses-allocator construction.
Which feels very necessary for our use case (I need to unordered container in the reverse pass of the multi index assignment.)
Activity
WardBrian commentedon May 8, 2023
Supersedes #2856, I suppose. Are we concerned at all with the BH library in R?
SteveBronder commentedon May 8, 2023
I think by the time we are up to this version of the math library BH will at least be at 1.82
wds15 commentedon May 9, 2023
In the past we aligned boots with the CRAN r version of bh…which is 1.81 for now.
SteveBronder commentedon May 9, 2023
@wds15 I was thinking about that, but usually rstan lags behind and boost 1.82 has been out for a minute. My guess is that BH will have updated by the time rstan is ready to use this version. Does that make sense?
I'm fine with just bumping to 1.81, but feels like we will end up bumping to 1.82 soon after
wds15 commentedon May 9, 2023
If you have a good reason to go to 1.82 ... then I'd be with you. Otherwise just go with 1.81 is what I would say. From interacting with the maintainer of BH I understood that he intends to do once per year updates only. This is due to the large effort of upgrading (such that my bet would be at 1.83 or even 1.84 to land on CRAN as BH). In any case, I do understand that rstan is so much behind such that it is almost hard to argue for compatibility with it. Given that you have already invested in an 1.82 PR maybe we should just go with it and simply discourage usage of 1.82 only features?
WardBrian commentedon May 9, 2023
I believe stan-dev/stan#3188 would be changed to use some Boost 1.82 features if they were avaible
SteveBronder commentedon May 9, 2023
I just let your script run and made one fix so this was very low effort on my part
@wds15 @WardBrian the unordered flat map is in 1.81, but 1.82 has a major update to the unordered containers which says
Which feels very necessary for our use case (I need to unordered container in the reverse pass of the multi index assignment.)
(ctrl+f for 'Unordered' in the release notes below)
https://www.boost.org/users/history/version_1_82_0.html
Though tbh I'm fine with not having a flat unordered map for right now and just waiting while we do a 1.81 upgrade for now
SteveBronder commentedon May 11, 2023
Actually I'm going to close this for now