library(tidyverse)

repos <- jsonlite::read_json("repos.json")

extract_licenses <- function(repo) {
  repo_name <- as.character(repo$name)
  path <- glue::glue("CRAN/{repo_name}.zip")

  files <- try(zip::zip_list(path), silent = TRUE)

  if (inherits(files, "try-error")) {
    return(NULL)
  }

  files <- files[endsWith(files$filename, "DESCRIPTION"), ]

  if (nrow(files) > 1) {
    files <- head(files, 1)
  }

  if (nrow(files) == 0) {
    return(NULL)
  }

  temp <- tempfile()
  on.exit(unlink(temp), add = TRUE)

  zip::unzip(path, files = files$filename, junkpaths = TRUE, exdir = temp)

  desc <- try(desc::desc(file = fs::path(temp, "DESCRIPTION")), silent = TRUE)

  if (inherits(desc, "try-error")) {
    return(NULL)
  }

  tibble(
    package = repo_name,
    license = desc$get("License")
  )
}

library(future)
plan(multisession, workers = 8)

# this lists all packages and the License as it's written in the DESCRIPTION
licenses <- furrr::future_map(
  repos,
  \(repo) extract_licenses(repo),
  .progress = TRUE
)

licenses <- bind_rows(licenses)

# Manually classified all possible licenses into 23 license codes.
# A tibble: 24 × 1
#    license_code
#    <chr>
#  1 ACM
#  2 AGPL
#  3 Apache
#  4 Artistic
#  5 BSD
#  6 BSD-Clause
#  7 BSL
#  8 CC-BY
#  9 CC-BY-NC
# 10 CC-BY-NC-ND
# 11 CC-BY-NC-SA
# 12 CC-BY-SA
# 13 CC0
# 14 CPL
# 15 CeCILL
# 16 EPL
# 17 EUPL
# 18 FreeBSD
# 19 GPL
# 20 LGPL
# 21 MIT
# 22 MPL
# 23 Unlimited
# 24 NA

# the manual classification ('classif' field) has to be parsed.
licenses_table <- readr::read_csv("licenses.csv")
# some licenses ended up duplicated (I think they have been trimed in excel or something)
licenses_table <- licenses_table %>% distinct()

parse_license <- function(license) {
  licenses <- license |>
    stringr::str_split(pattern = stringr::fixed("|"))
  licenses <- stringr::str_trim(licenses[[1]])

  license <- stringr::str_extract(licenses, "(.*?)(?=\\()|.*")
  license <- stringr::str_trim(license)

  version <- stringr::str_extract(licenses, "\\((.*?)\\)")
  tibble(
    license_code = license,
    version = version
  )
}

# this list is based in the licenses.json file from the stack dataset
# see: https://huggingface.co/datasets/bigcode/the-stack-dedup/blob/main/licenses.json

allowed_license_codes <- c("MIT", "Apache", "BSD", "CC-BY", "Artistic", "CC0",
                           "FreeBSD")


# we now make the indirection from allowed codes to allowed License as specified
# in the DESCRIPTION.
license_codes <- licenses_table %>%
  mutate(
    parsed = map(classif, parse_license)
  ) %>%
  unnest(parsed)

permissive_licenses <- license_codes%>%
  mutate(permissive = license_code %in% allowed_license_codes) %>%
  group_by(license) %>%
  summarise(permissive = any(permissive)) %>%
  filter(permissive)

# We then filter the allowed packages.
allowed_packages <- licenses %>%
  filter(license %in% permissive_licenses$license)

arrow::write_parquet(allowed_packages, "packages.parquet")

# the distribution of packages per license code
licenses %>%
  left_join(license_codes, by = "license", relationship = "many-to-many") %>%
  count(license_code) %>%
  mutate(n / nrow(licenses)) %>%
  arrange(desc(n)) %>%
  print(n = 100)

#  # A tibble: 24 × 3
#  license_code     n `n/nrow(licenses)`
#  <chr>        <int>              <dbl>
#  1 GPL          18440          0.726
#  2 MIT           4533          0.179
#  3 LGPL           556          0.0219
#  4 Apache         428          0.0169
#  5 BSD-Clause     343          0.0135
#  6 CC0            261          0.0103
#  7 AGPL           232          0.00914
#  8 Artistic       158          0.00622
#  9 NA              96          0.00378
# 10 Unlimited       76          0.00299
# 11 CC-BY           56          0.00221
# 12 MPL             52          0.00205
# 13 EUPL            35          0.00138
# 14 CC-BY-SA        34          0.00134
# 15 CeCILL          34          0.00134
# 16 BSD             33          0.00130
# 17 CC-BY-NC-SA     27          0.00106
# 18 FreeBSD         18          0.000709
# 19 BSL             10          0.000394
# 20 CC-BY-NC         6          0.000236
# 21 EPL              6          0.000236
# 22 ACM              5          0.000197
# 23 CPL              3          0.000118
# 24 CC-BY-NC-ND      1          0.0000394