New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
improve Matrix-ABI documentation #768
Comments
FWIW, the glmmTMB machinery seems a bit overcomplicated. Why not put
in |
Just to be clear: we should plan to send an updated The reason for all of the |
Yes, though we will ask CRAN to rebuild lme4 anyway. The version bump will primarily serve as a nudge for repositories other than CRAN.
At install time, files
in any |
Then your
where (Well, for Matrix you want to be storing and comparing ABI versions, not package versions. Anyway ...) |
preliminary attempt: 8be641b (eventually I will try to back-port this to |
Looks OK modulo my comment on the commit. Probably worth testing locally (if you haven't already) by installing lme4 under Matrix < 1.6-2, installing Matrix >= 1.6-2, and loading lme4. |
BTW: It was decided that Matrix 1.7-0 should be released with |
As you can see from this thread on r-package-devel, this is already causing minor headaches - 1.7.0 has been installed on win-builder without bumping the binary-dependent packages ... sigh ... |
It was wrong of me to say "not until April" because binary repositories have already published Martin submitted 1.7-0 to CRAN today, and the incoming checks just finished, so people out there could start seeing 1.7-0 binaries soon-ish. I can ping here when 1.7-0 is actually published. At any rate, an lme4 submission in the next week or so would make sense ... |
What are the risks that the releases won't be properly synchronized, i.e. that |
Nonzero unless you change to |
Ugh. I think this is upon us now. I asked CRAN maintainers for advice about staging releases and got the following:
I have a pretty strong preference not to make
I don't really understand this point — I guess the point is that lme4 has to be re-installed from source (but, I'm much more worried about supporting users who get binaries from other repositories/aren't up to installing compilation tools and re-installing from source themselves ..)
I'm really not sure what to do next. I guess sending a new version of Any thoughts or clarifications are welcome. |
Partly to mitigate the fallout of the ABI breakage for most users who will not see R 4.4 until after April 24. Partly because we plan to use BLAS and LAPACK routines not exposed in the R API until R 4.4.
You can cross your fingers and submit lme4 now without the stricter dependencies, with the reasonable hope that any repository building binaries for R 4.4 before April 24 will do so using the latest version of R-devel, including the latest versions of all recommended packages. Or you can withhold the submission until after April 24 and deal will the fallout in the mean time (e.g., #775). Or you can submit now with the stricter dependencies, which is maybe CRAN's preference ... I understand not wanting to bar for "no reason" users of recent/old versions of R from new/future versions of lme4. There actually is a reason for Matrix, which is that new versions of Matrix will use R 4.4 API. Martin may have more/other thoughts but he is on vacation. |
I think I'm going to cross my fingers and submit. Thanks for the input. |
New version on CRAN now, should propagate. Never did fix these things:
And, I think I'm going to leave this open for now because I think we could still do better a documenting - i.e., collect the information about when/why these updates are needed, give the clearest possible description of the solutions (install from r-universe? CRAN?), and put this on a man page that the warning message points to ... |
From @jaganmn:
We should be ready/get out in front of this.
glmmTMB
machinery as appropriate)predict
+se.fit
implementation:predict()
fails whense.fit = TRUE
andre.form = NULL
#753)scale()
continuous covariates? #731 seems too ambitious, allFit improvements #724 might be worth a little bit of workThe text was updated successfully, but these errors were encountered: