diff options
author | Michał Górny <mgorny@gentoo.org> | 2020-04-13 11:26:40 +0200 |
---|---|---|
committer | Michał Górny <mgorny@gentoo.org> | 2020-04-13 12:32:14 +0200 |
commit | d2046bcd67c169466279d561c5a69006f2faa74d (patch) | |
tree | 569d06f466aad24914ece43eb28a76058eadd5cc /dev-python/nltk-data | |
parent | dev-python/nltk: Remove old (drop to ~arch) (diff) | |
download | gentoo-d2046bcd67c169466279d561c5a69006f2faa74d.tar.gz gentoo-d2046bcd67c169466279d561c5a69006f2faa74d.tar.bz2 gentoo-d2046bcd67c169466279d561c5a69006f2faa74d.zip |
dev-python/nltk-data: Install corpora/pl196x unconditionally
Install corpora/pl196x unconditionally as it is required
for dev-python/nltk-3.5 tests.
Signed-off-by: Michał Górny <mgorny@gentoo.org>
Diffstat (limited to 'dev-python/nltk-data')
-rw-r--r-- | dev-python/nltk-data/nltk-data-20200312-r1.ebuild | 184 |
1 files changed, 184 insertions, 0 deletions
diff --git a/dev-python/nltk-data/nltk-data-20200312-r1.ebuild b/dev-python/nltk-data/nltk-data-20200312-r1.ebuild new file mode 100644 index 000000000000..c8f12c580e60 --- /dev/null +++ b/dev-python/nltk-data/nltk-data-20200312-r1.ebuild @@ -0,0 +1,184 @@ +# Copyright 2020 Gentoo Authors +# Distributed under the terms of the GNU General Public License v2 + +EAPI=7 + +inherit check-reqs + +DESCRIPTION="Data files for NLTK" +HOMEPAGE="https://www.nltk.org/nltk_data/" + +# at least some of the files have poorly documented licenses +# TODO: create a USE flag for free-ish subset +LICENSE="all-rights-reserved" +SLOT="0" +KEYWORDS="~amd64 ~x86" +IUSE="extra" +RESTRICT="bindist mirror" + +BDEPEND="app-arch/unzip" + +PACKAGES_ZIP=( + # wget -O - https://www.nltk.org/nltk_data/ | xml sel -t -m '//package[@unzip=0]' -v @subdir -o "/" -v @id -n - | sort + corpora/comtrans + corpora/conll2007 + corpora/jeita + corpora/knbc + corpora/machado + corpora/masc_tagged + corpora/nombank.1.0 + corpora/panlex_swadesh + corpora/propbank + corpora/reuters + corpora/semcor + corpora/universal_treebanks_v20 + sentiment/vader_lexicon + stemmers/snowball_data +) + +PACKAGES_UNPACK=( + # wget -O - https://www.nltk.org/nltk_data/ | xml sel -t -m '//package[@unzip=1]' -v @subdir -o "/" -v @id -n - | sort + corpora/abc + corpora/alpino + corpora/brown + corpora/cess_cat + corpora/cess_esp + corpora/chat80 + corpora/city_database + corpora/cmudict + corpora/comparative_sentences + corpora/conll2000 + corpora/conll2002 + corpora/crubadan + corpora/dependency_treebank + corpora/dolch + corpora/europarl_raw + corpora/floresta + corpora/framenet_v15 + corpora/framenet_v17 + corpora/gazetteers + corpora/genesis + corpora/gutenberg + corpora/ieer + corpora/inaugural + corpora/indian + corpora/lin_thesaurus + corpora/mac_morpho + corpora/movie_reviews + corpora/mte_teip5 + corpora/names + corpora/nonbreaking_prefixes + corpora/nps_chat + corpora/omw + corpora/opinion_lexicon + corpora/pl196x + corpora/ppattach + corpora/product_reviews_1 + corpora/product_reviews_2 + corpora/pros_cons + corpora/ptb + corpora/qc + corpora/rte + corpora/senseval + corpora/sentence_polarity + corpora/sentiwordnet + corpora/shakespeare + corpora/sinica_treebank + corpora/state_union + corpora/stopwords + corpora/subjectivity + corpora/swadesh + corpora/switchboard + corpora/timit + corpora/toolbox + corpora/treebank + corpora/twitter_samples + corpora/udhr + corpora/udhr2 + corpora/verbnet + corpora/webtext + corpora/wordnet + corpora/wordnet_ic + corpora/words + grammars/book_grammars + grammars/large_grammars + grammars/sample_grammars + misc/perluniprops + models/bllip_wsj_no_aux + models/moses_sample + models/wmt15_eval + models/word2vec_sample + stemmers/porter_test + stemmers/rslp + taggers/averaged_perceptron_tagger + taggers/averaged_perceptron_tagger_ru + taggers/universal_tagset + tokenizers/punkt +) + +PACKAGES_UNPACK_EXTRA=( + chunkers/maxent_ne_chunker + corpora/biocreative_ppi + corpora/brown_tei + corpora/kimmo + corpora/paradigms + corpora/pe08 + corpora/pil + corpora/problem_reports + corpora/smultron + corpora/unicode_samples + corpora/verbnet3 + corpora/ycoe + grammars/basque_grammars + grammars/spanish_grammars + help/tagsets + misc/mwa_ppdb + taggers/maxent_treebank_pos_tagger +) + +add_data() { + local x + for x; do + SRC_URI+=" + https://raw.githubusercontent.com/nltk/nltk_data/gh-pages/packages/${x}.zip + -> nltk-${x#*/}-${PV}.zip" + done +} + +add_data "${PACKAGES_ZIP[@]}" "${PACKAGES_UNPACK[@]}" +SRC_URI+=" + extra? (" +add_data "${PACKAGES_UNPACK_EXTRA[@]}" +SRC_URI+=" + )" + +CHECKREQS_DISK_USR=3G +CHECKREQS_DISK_BUILD=${CHECKREQS_DISK_USR} + +src_unpack() { + local x + local to_unpack=( "${PACKAGES_UNPACK[@]}" ) + use extra && to_unpack+=( "${PACKAGES_UNPACK_EXTRA[@]}" ) + for x in "${to_unpack[@]}"; do + local cat=${x%/*} + local pkg=${x#*/} + + mkdir -p "${S}/${cat}" || die + cd "${S}/${cat}" || die + unpack "nltk-${pkg}-${PV}.zip" + done +} + +src_install() { + dodir /usr/share/nltk_data + mv * "${ED}/usr/share/nltk_data/" || die + + local x + for x in "${PACKAGES_ZIP[@]}"; do + local cat=${x%/*} + local pkg=${x#*/} + + insinto "/usr/share/nltk_data/${cat}" + newins "${DISTDIR}/nltk-${pkg}-${PV}.zip" "${pkg}.zip" + done +} |