survival: different results with R `v3.6.3` vs `v4.2.0`, but same `rstanarm` version and seed

Summary:

Same rstanarm version from feature/survival (version 2.21.2), same seed, different results between R 3.6.3 and 4.2.0.

Description:

As per the summary, I have the same version of rstanarm from feature/survival installed on R version 3.6.3 and 4.2.0. Even with the same seed, I get different draws form the posterior distribution.

The 2 versions of rstanarm were not installed at the same time. Could it be that some changes to the code potentially related to this issue were introduced, yet the version of rstanarm was not updated? Otherwise, do you have any idea about what the different results could be due to?

But above all: do you know I can obtain consistent results between R 3.6.3 and 4.2.0?

Output:

(reproducible code below)

R 3.6.3 + rstanarm 2.21.2

stan_surv
 baseline hazard: weibull
 formula:         Surv(futime, death) ~ mspike
 observations:    241
 events:          225 (93.4%)
 right censored:  16 (6.6%)
 delayed entry:   no
------
              Median    MAD_SD    exp(Median)
(Intercept)   -10.15258   0.62484        NA  
mspike         -0.12206   0.17364   0.88510  
weibull-shape   1.18821   0.06543        NA  

R 4.2.0 + rstanarm 2.21.2

stan_surv
 baseline hazard: weibull
 formula:         Surv(futime, death) ~ mspike
 observations:    241
 events:          225 (93.4%)
 right censored:  16 (6.6%)
 delayed entry:   no
------
              Median    MAD_SD    exp(Median)
(Intercept)   -10.14793   0.74071        NA  
mspike         -0.11830   0.18188   0.88843  
weibull-shape   1.19225   0.07253        NA 

Reproducible Code:

library(survival)
library(rstanarm)

surv <- stan_surv(Surv(futime, death) ~ mspike, 
                  data = mgus, 
                  basehaz = "weibull",
                  chains = 1, 
                  seed = 1234)

print(surv, digits = 5)

RStanARM Version:

2.21.2

R Version:

3.6.3 and 4.2.0

Operating System:

Mac OS 10.14.6

Start by running the reproducible stan_surv example with R 3.6.3 and R 4.2.0, using the stated rstanarm 2.21.2 version, seed, data, and settings. Compare the environments and posterior output to identify the source of the discrepancy; done means explaining the cause and documenting how to obtain consistent results.