CI provides errors as riskmetric assessments

[llrs-roche]

Currently the riskmetric assessment provided by the CI (in inst/assessments) has this tally:

d_riskmetric <- readRDS("inst/assessments/dplyr.rds")
sort(table(sapply(d_riskmetric, is)), TRUE)
## 
##                pkg_metric_error                   pkg_metric_na 
##                              11                               4 
##         pkg_metric_dependencies        pkg_metric_downloads_1yr 
##                               1                               1 
## pkg_metric_reverse_dependencies                 pkg_metric_todo 
##                               1                               1 

There are only 3 useful data points. It would be great if the CI could reduce the number of metrics that are errors.

[llrs-roche]

@Gotfrid Do you think that the assessment generated by the CI could reduce the number of errors? The current file provided has very little information to write a good report. I got more results running riskmetric locally.

[Gotfrid]

@llrs-roche thank you for highlighting this! after some testing i was able to deduce that this is not a problem of CI, but rather the problem of the repository that we are using.

Likely, you use the default CRAN repo when you test things locally, but in the CI we use https://github.com/r-hub/repos. Riskmetric fails to generate scores for a package that was taken from this repo - I suppose due to the fact that this is a "special" (not real) cran-like repo, as we discussed on one of the meetings.

I tried using https://cran.r-universe.dev and it seems to be fine. I remember it was mentioned once that we should consider r-universe as our baseline repo. The problem that we are discussing here might very well be a strong reason to do it.

[llrs-roche]

I use the canon CRAN repository to check packages locally.
This could also be fixed on riskmetric, as I found on several issues with how it read packages sources (pharmaR/riskmetric#357, pharmaR/riskmetric#358) .
For the moment I will add an assessment generated locally from CRAN (or one you generate with the CI from cran.r-univers.dev), but, this will need to be discussed about on the meeting.

[llrs-roche]

Running the assessment locally work better but there are still 3 NA:

d_riskmetric <- readRDS("inst/assessments/dplyr_local.rds")
sort(table(sapply(d_riskmetric, is)), TRUE)
## 
##                   pkg_metric_na         pkg_metric_dependencies        pkg_metric_downloads_1yr          pkg_metric_export_help   pkg_metric_exported_namespace 
##                               3                               1                               1                               1                               1 
##  pkg_metric_has_bug_reports_url         pkg_metric_has_examples       pkg_metric_has_maintainer             pkg_metric_has_news   pkg_metric_has_source_control 
##                               1                               1                               1                               1                               1 
##        pkg_metric_has_vignettes          pkg_metric_has_website  pkg_metric_last_30_bugs_status              pkg_metric_license         pkg_metric_news_current 
##                               1                               1                               1                               1                               1 
## pkg_metric_reverse_dependencies        pkg_metric_size_codebase 
##                               1                               1