On 7 November 2023 at 14:58, Andreas Tille wrote:
| Do you see any way to answer the question that is discussed in this
| thread by r2u how to know whether new Bioconductor packages might have
| new dependencies not yet packaged for Debian?
"Kinda. Sorta. Not fully." I have written related code doing most of this
during the many attempt for 'turning CRAN into .deb packages'.
I.e. when I recompile BioC packages in r2u as I did this weekend I start from
all BioC packages I have already built within r2u (same for you here for a
'within Debian' check), use available.packages() etc to get the package
database (in the R sense) and use that to map out dependencies. In my case I
sort strip off CRAN (already built) and base R packages to get a count of
'pure BioC depends'. I then sort and first build all of these with a
dependency count of zero, refresh the index so that these become available,
then all with a count of one and so. (Max count this weekend was 41.)
The one step I did not do (as I didn't need it) was to check 'is package X
already available'. When it wasn't I just built it :) But you can do all that
from either shell into apt-cache, or R via my RcppAPT package, or via
python-apt and friends.
My code is in R with use of data.table for the mangling so it is somewhat
'internal'. It is based on R's own 'tools::package_dependencies()'. There
must also be suitable code in R itself which I never pulled out because R can
run a package's reverse dependencies. But anyway here is a minimal sketch
using R and its data.table package.
> AP <- suppressMessages( data.table(available.packages(repos =
> BiocManager::repositories())) )
> AP[, lcpkg := tolower(Package)]
> basePkgs <- c("base", "class", "codetools", "datasets", "graphics", "grid",
> "lattice",
+ "Matrix", "mgcv", "nnet", "rpart", "splines", "stats4", "tcltk",
"translations",
+ "boot", "cluster", "compiler", "foreign", "grDevices", "KernSmooth", "MASS",
+ "methods", "nlme", "parallel", "spatial", "stats", "survival", "tools",
"utils")
> cranPkgs <- AP[Repository=="https://cloud.r-project.org/src/contrib";, Package]
> biocPkgs <- AP[Repository!="https://cloud.r-project.org/src/contrib";, Package]
>
> pkg <- "SingleCellExperiment"
> deps <- tools::package_dependencies(pkg, AP, recursive=TRUE)[[1]]
> nAll <- length(deps)
> nBase <- length(intersect(deps, basePkgs))
> nCran <- length(intersect(deps, cranPkgs))
> nBioc <- length(intersect(deps, biocPkgs))
>
> intersect(deps, biocPkgs)
[1] "SummarizedExperiment" "S4Vectors" "BiocGenerics"
"GenomicRanges"
[5] "DelayedArray" "MatrixGenerics" "IRanges"
"S4Arrays"
[9] "GenomeInfoDb" "XVector" "Biobase"
"GenomeInfoDbData"
[13] "zlibbioc"
>
So for all packages you are interested in (here I look just at
'SingleCellExperiment') you construct the BioC (or maybe CRAN and BioC)
dependencies, and then create an aggregate list of the unique
combination. Those are the packages you need and apt-cache and related will
tell you if they exist.
Dirk
--
dirk.eddelbuettel.com | @eddelbuettel | e...@debian.org
