bug#48005: low performance when building profile of already built packages

[Ricardo Wurmus]

Here is a manifest that isn’t unusual for a bioinformatics 
workflow:

--8<---------------cut here---------------start------------->8---
(define packages
 (list "bash-minimal"
       "fastqc"
       "gzip"
       "htseq"
       "multiqc"
       "r-corrplot"
       "r-crosstalk"
       "r-data-table"
       "r-deseq2"
       "r-dt"
       "r-genomicalignments"
       "r-genomicranges"
       "r-ggplot2"
       "r-ggrepel"
       "r-gprofiler"
       "r-knitr"
       "r-pheatmap"
       "r-plotly"
       "r-reshape2"
       "r-rmarkdown"
       "r-rsamtools"
       "r-rtracklayer"
       "r-s4vectors"
       "r-scales"
       "r-summarizedexperiment"
       "r-tximport"
       "samtools"
       "sed"
       "trim-galore"))

(specifications->manifest packages)
--8<---------------cut here---------------end--------------->8---

When all these packages already exist and nothing is to be done 
when Guix is asked to instantiate this manifest, more than 13 
seconds pass.

--8<---------------cut here---------------start------------->8---
$ [env] export GUIX_PROFILING="object-cache 
add-data-to-store-cache rpc"
$ [env] time guix package -m manifest-test.scm -p /tmp/foo
The following packages will be installed:
  bash-minimal           5.0.16
  fastqc                 0.11.5
  gzip                   1.10
  htseq                  0.9.1
  multiqc                1.5
  r-corrplot             0.84
  r-crosstalk            1.1.1
  r-data-table           1.14.0
  r-deseq2               1.30.1
  r-dt                   0.17
  r-genomicalignments    1.26.0
  r-genomicranges        1.42.0
  r-ggplot2              3.3.3
  r-ggrepel              0.9.1
  r-gprofiler            0.7.0
  r-knitr                1.31
  r-pheatmap             1.0.12
  r-plotly               4.9.3
  r-reshape2             1.4.4
  r-rmarkdown            2.7
  r-rsamtools            2.6.0
  r-rtracklayer          1.50.0
  r-s4vectors            0.28.1
  r-scales               1.1.1
  r-summarizedexperiment 1.20.0
  r-tximport             1.18.0
  samtools               1.11
  sed                    4.8
  trim-galore            0.6.6
[[most of the time passes here]]

nothing to be done
Store object cache:
 fresh caches:     2
 lookups:      45493
 hits:         43943 (96.6%)
 cache size:    1549 entries

'add-data-to-store' cache:
 lookups:       7755
 hits:          3411 (44.0%)
 .drv files:    4357 (56.2%)
 Scheme files:  3343 (43.1%)
Remote procedure call summary: 5283 RPCs
 built-in-builders              ...     1
 add-to-store/tree              ...    27
 add-to-store                   ...   213
 query-references               ...   698
 add-text-to-store              ...  4344

real    0m13.379s
user    0m16.414s
sys     0m0.253s
--8<---------------cut here---------------end--------------->8---

Some of these R packages have a lot of transitive inputs, such as 
r-deseq2 with 81 transitive propagated inputs.  So there’s a lot 
of propagation here, which I suspect contributes to things being 
so unexpectedly slow.

This report may be related to performance issues reported for 
“guix environment” in https://issues.guix.gnu.org/41702, which 
also involves R packages (such as r-learnr with 46 transitive 
propagated inputs).

--
Ricardo