i18n.py

"""Docutils transforms used by Sphinx when reading documents."""

from os import path
from re import DOTALL, match
from textwrap import indent
from typing import TYPE_CHECKING, Any, Dict, List, Optional, Tuple, Type, TypeVar

from docutils import nodes
from docutils.io import StringInput
from docutils.nodes import Element
from docutils.utils import relative_path

from sphinx import addnodes
from sphinx.config import Config
from sphinx.domains.std import make_glossary_term, split_term_classifiers
from sphinx.locale import __
from sphinx.locale import init as init_locale
from sphinx.transforms import SphinxTransform
from sphinx.util import get_filetype, logging, split_index_msg
from sphinx.util.i18n import docname_to_domain
from sphinx.util.nodes import (IMAGE_TYPE_NODES, LITERAL_TYPE_NODES, NodeMatcher,
                               extract_messages, is_pending_meta, traverse_translatable_index)

if TYPE_CHECKING:
    from sphinx.application import Sphinx


logger = logging.getLogger(__name__)

# The attributes not copied to the translated node
#
# * refexplict: For allow to give (or not to give) an explicit title
#               to the pending_xref on translation
EXCLUDED_PENDING_XREF_ATTRIBUTES = ('refexplicit',)


N = TypeVar('N', bound=nodes.Node)


def publish_msgstr(app: "Sphinx", source: str, source_path: str, source_line: int,
                   config: Config, settings: Any) -> Element:
    """Publish msgstr (single line) into docutils document

    :param sphinx.application.Sphinx app: sphinx application
    :param str source: source text
    :param str source_path: source path for warning indication
    :param source_line: source line for warning indication
    :param sphinx.config.Config config: sphinx config
    :param docutils.frontend.Values settings: docutils settings
    :return: document
    :rtype: docutils.nodes.document
    """
    try:
        # clear rst_prolog temporarily
        rst_prolog = config.rst_prolog
        config.rst_prolog = None  # type: ignore

        from sphinx.io import SphinxI18nReader
        reader = SphinxI18nReader()
        reader.setup(app)
        filetype = get_filetype(config.source_suffix, source_path)
        parser = app.registry.create_source_parser(app, filetype)
        doc = reader.read(
            source=StringInput(source=source,
                               source_path="%s:%s:<translated>" % (source_path, source_line)),
            parser=parser,
            settings=settings,
        )
        try:
            doc = doc[0]  # type: ignore
        except IndexError:  # empty node
            pass
        return doc
    finally:
        config.rst_prolog = rst_prolog  # type: ignore


def parse_noqa(source: str) -> Tuple[str, bool]:
    m = match(r"(.*)(?<!\\)#\s*noqa\s*$", source, DOTALL)
    if m:
        return m.group(1), True
    else:
        return source, False


class PreserveTranslatableMessages(SphinxTransform):
    """
    Preserve original translatable messages before translation
    """
    default_priority = 10  # this MUST be invoked before Locale transform

    def apply(self, **kwargs: Any) -> None:
        for node in self.document.findall(addnodes.translatable):
            node.preserve_original_messages()


class Locale(SphinxTransform):
    """
    Replace translatable nodes with their translated doctree.
    """
    default_priority = 20

    def apply(self, **kwargs: Any) -> None:
        settings, source = self.document.settings, self.document['source']
        msgstr = ''

        # XXX check if this is reliable
        assert source.startswith(self.env.srcdir)
        docname = path.splitext(relative_path(path.join(self.env.srcdir, 'dummy'),
                                              source))[0]
        textdomain = docname_to_domain(docname, self.config.gettext_compact)

        # fetch translations
        dirs = [path.join(self.env.srcdir, directory)
                for directory in self.config.locale_dirs]
        catalog, has_catalog = init_locale(dirs, self.config.language, textdomain)
        if not has_catalog:
            return

        # phase1: replace reference ids with translated names
        for node, msg in extract_messages(self.document):
            msgstr = catalog.gettext(msg)

            # There is no point in having #noqa on literal blocks because
            # they cannot contain references.  Recognizing it would just
            # completely prevent escaping the #noqa.  Outside of literal
            # blocks, one can always write \#noqa.
            if not isinstance(node, LITERAL_TYPE_NODES):
                msgstr, _ = parse_noqa(msgstr)

            # XXX add marker to untranslated parts
            if not msgstr or msgstr == msg or not msgstr.strip():
                # as-of-yet untranslated
                continue

            # Avoid "Literal block expected; none found." warnings.
            # If msgstr ends with '::' then it cause warning message at
            # parser.parse() processing.
            # literal-block-warning is only appear in avobe case.
            if msgstr.strip().endswith('::'):
                msgstr += '\n\n   dummy literal'
                # dummy literal node will discard by 'patch = patch[0]'

            # literalblock need literal block notation to avoid it become
            # paragraph.
            if isinstance(node, LITERAL_TYPE_NODES):
                msgstr = '::\n\n' + indent(msgstr, ' ' * 3)

            patch = publish_msgstr(self.app, msgstr, source,
                                   node.line, self.config, settings)
            # FIXME: no warnings about inconsistent references in this part
            # XXX doctest and other block markup
            if not isinstance(patch, nodes.paragraph):
                continue  # skip for now

            processed = False  # skip flag

            # update title(section) target name-id mapping
            if isinstance(node, nodes.title) and isinstance(node.parent, nodes.section):
                section_node = node.parent
                new_name = nodes.fully_normalize_name(patch.astext())
                old_name = nodes.fully_normalize_name(node.astext())

                if old_name != new_name:
                    # if name would be changed, replace node names and
                    # document nameids mapping with new name.
                    names = section_node.setdefault('names', [])
                    names.append(new_name)
                    # Original section name (reference target name) should be kept to refer
                    # from other nodes which is still not translated or uses explicit target
                    # name like "`text to display <explicit target name_>`_"..
                    # So, `old_name` is still exist in `names`.

                    _id = self.document.nameids.get(old_name, None)
                    explicit = self.document.nametypes.get(old_name, None)

                    # * if explicit: _id is label. title node need another id.
                    # * if not explicit:
                    #
                    #   * if _id is None:
                    #
                    #     _id is None means:
                    #
                    #     1. _id was not provided yet.
                    #
                    #     2. _id was duplicated.
                    #
                    #        old_name entry still exists in nameids and
                    #        nametypes for another duplicated entry.
                    #
                    #   * if _id is provided: below process
                    if _id:
                        if not explicit:
                            # _id was not duplicated.
                            # remove old_name entry from document ids database
                            # to reuse original _id.
                            self.document.nameids.pop(old_name, None)
                            self.document.nametypes.pop(old_name, None)
                            self.document.ids.pop(_id, None)

                        # re-entry with new named section node.
                        #
                        # Note: msgnode that is a second parameter of the
                        # `note_implicit_target` is not necessary here because
                        # section_node has been noted previously on rst parsing by
                        # `docutils.parsers.rst.states.RSTState.new_subsection()`
                        # and already has `system_message` if needed.
                        self.document.note_implicit_target(section_node)

                    # replace target's refname to new target name
                    matcher = NodeMatcher(nodes.target, refname=old_name)
                    for old_target in self.document.findall(matcher):  # type: nodes.target
                        old_target['refname'] = new_name

                    processed = True

            # glossary terms update refid
            if isinstance(node, nodes.term):
                for _id in node['ids']:
                    parts = split_term_classifiers(msgstr)
                    patch = publish_msgstr(self.app, parts[0], source,
                                           node.line, self.config, settings)
                    patch = make_glossary_term(self.env, patch, parts[1],
                                               source, node.line, _id,
                                               self.document)
                    processed = True

            # update leaves with processed nodes
            if processed:
                for child in patch.children:
                    child.parent = node
                node.children = patch.children
                node['translated'] = True  # to avoid double translation

        # phase2: translation
        for node, msg in extract_messages(self.document):
            if node.get('translated', False):  # to avoid double translation
                continue  # skip if the node is already translated by phase1

            msgstr = catalog.gettext(msg)

            # See above.
            if not isinstance(node, LITERAL_TYPE_NODES):
                msgstr, noqa = parse_noqa(msgstr)

            # XXX add marker to untranslated parts
            if not msgstr or msgstr == msg:  # as-of-yet untranslated
                continue

            # update translatable nodes
            if isinstance(node, addnodes.translatable):
                node.apply_translated_message(msg, msgstr)  # type: ignore
                continue

            # update meta nodes
            if isinstance(node, nodes.pending) and is_pending_meta(node):
                # docutils-0.17
                node.details['nodes'][0]['content'] = msgstr
                continue
            elif isinstance(node, addnodes.docutils_meta):
                # docutils-0.18+
                node['content'] = msgstr
                continue

            if isinstance(node, nodes.image) and node.get('alt') == msg:
                node['alt'] = msgstr
                continue

            # Avoid "Literal block expected; none found." warnings.
            # If msgstr ends with '::' then it cause warning message at
            # parser.parse() processing.
            # literal-block-warning is only appear in avobe case.
            if msgstr.strip().endswith('::'):
                msgstr += '\n\n   dummy literal'
                # dummy literal node will discard by 'patch = patch[0]'

            # literalblock need literal block notation to avoid it become
            # paragraph.
            if isinstance(node, LITERAL_TYPE_NODES):
                msgstr = '::\n\n' + indent(msgstr, ' ' * 3)

            # Structural Subelements phase1
            # There is a possibility that only the title node is created.
            # see: https://docutils.sourceforge.io/docs/ref/doctree.html#structural-subelements
            if isinstance(node, nodes.title):
                # This generates: <section ...><title>msgstr</title></section>
                msgstr = msgstr + '\n' + '=' * len(msgstr) * 2

            patch = publish_msgstr(self.app, msgstr, source,
                                   node.line, self.config, settings)
            # Structural Subelements phase2
            if isinstance(node, nodes.title):
                # get <title> node that placed as a first child
                patch = patch.next_node()

            # ignore unexpected markups in translation message
            unexpected: Tuple[Type[Element], ...] = (
                nodes.paragraph,    # expected form of translation
                nodes.title         # generated by above "Subelements phase2"
            )

            # following types are expected if
            # config.gettext_additional_targets is configured
            unexpected += LITERAL_TYPE_NODES
            unexpected += IMAGE_TYPE_NODES

            if not isinstance(patch, unexpected):
                continue  # skip

            # auto-numbered foot note reference should use original 'ids'.
            def list_replace_or_append(lst: List[N], old: N, new: N) -> None:
                if old in lst:
                    lst[lst.index(old)] = new
                else:
                    lst.append(new)

            is_autofootnote_ref = NodeMatcher(nodes.footnote_reference, auto=Any)
            old_foot_refs: List[nodes.footnote_reference] = list(node.findall(is_autofootnote_ref))  # NOQA
            new_foot_refs: List[nodes.footnote_reference] = list(patch.findall(is_autofootnote_ref))  # NOQA
            if not noqa and len(old_foot_refs) != len(new_foot_refs):
                old_foot_ref_rawsources = [ref.rawsource for ref in old_foot_refs]
                new_foot_ref_rawsources = [ref.rawsource for ref in new_foot_refs]
                logger.warning(__('inconsistent footnote references in translated message.' +
                                  ' original: {0}, translated: {1}')
                               .format(old_foot_ref_rawsources, new_foot_ref_rawsources),
                               location=node, type='i18n', subtype='inconsistent_references')
            old_foot_namerefs: Dict[str, List[nodes.footnote_reference]] = {}
            for r in old_foot_refs:
                old_foot_namerefs.setdefault(r.get('refname'), []).append(r)
            for newf in new_foot_refs:
                refname = newf.get('refname')
                refs = old_foot_namerefs.get(refname, [])
                if not refs:
                    newf.parent.remove(newf)
                    continue

                oldf = refs.pop(0)
                newf['ids'] = oldf['ids']
                for id in newf['ids']:
                    self.document.ids[id] = newf

                if newf['auto'] == 1:
                    # autofootnote_refs
                    list_replace_or_append(self.document.autofootnote_refs, oldf, newf)
                else:
                    # symbol_footnote_refs
                    list_replace_or_append(self.document.symbol_footnote_refs, oldf, newf)

                if refname:
                    footnote_refs = self.document.footnote_refs.setdefault(refname, [])
                    list_replace_or_append(footnote_refs, oldf, newf)

                    refnames = self.document.refnames.setdefault(refname, [])
                    list_replace_or_append(refnames, oldf, newf)

            # reference should use new (translated) 'refname'.
            # * reference target ".. _Python: ..." is not translatable.
            # * use translated refname for section refname.
            # * inline reference "`Python <...>`_" has no 'refname'.
            is_refnamed_ref = NodeMatcher(nodes.reference, refname=Any)
            old_refs: List[nodes.reference] = list(node.findall(is_refnamed_ref))
            new_refs: List[nodes.reference] = list(patch.findall(is_refnamed_ref))
            if not noqa and len(old_refs) != len(new_refs):
                old_ref_rawsources = [ref.rawsource for ref in old_refs]
                new_ref_rawsources = [ref.rawsource for ref in new_refs]
                logger.warning(__('inconsistent references in translated message.' +
                                  ' original: {0}, translated: {1}')
                               .format(old_ref_rawsources, new_ref_rawsources),
                               location=node, type='i18n', subtype='inconsistent_references')
            old_ref_names = [r['refname'] for r in old_refs]
            new_ref_names = [r['refname'] for r in new_refs]
            orphans = list(set(old_ref_names) - set(new_ref_names))
            for newr in new_refs:
                if not self.document.has_name(newr['refname']):
                    # Maybe refname is translated but target is not translated.
                    # Note: multiple translated refnames break link ordering.
                    if orphans:
                        newr['refname'] = orphans.pop(0)
                    else:
                        # orphan refnames is already empty!
                        # reference number is same in new_refs and old_refs.
                        pass

                self.document.note_refname(newr)

            # refnamed footnote should use original 'ids'.
            is_refnamed_footnote_ref = NodeMatcher(nodes.footnote_reference, refname=Any)
            old_foot_refs = list(node.findall(is_refnamed_footnote_ref))
            new_foot_refs = list(patch.findall(is_refnamed_footnote_ref))
            refname_ids_map: Dict[str, List[str]] = {}
            if not noqa and len(old_foot_refs) != len(new_foot_refs):
                old_foot_ref_rawsources = [ref.rawsource for ref in old_foot_refs]
                new_foot_ref_rawsources = [ref.rawsource for ref in new_foot_refs]
                logger.warning(__('inconsistent footnote references in translated message.' +
                                  ' original: {0}, translated: {1}')
                               .format(old_foot_ref_rawsources, new_foot_ref_rawsources),
                               location=node, type='i18n', subtype='inconsistent_references')
            for oldf in old_foot_refs:
                refname_ids_map.setdefault(oldf["refname"], []).append(oldf["ids"])
            for newf in new_foot_refs:
                refname = newf["refname"]
                if refname_ids_map.get(refname):
                    newf["ids"] = refname_ids_map[refname].pop(0)

            # citation should use original 'ids'.
            is_citation_ref = NodeMatcher(nodes.citation_reference, refname=Any)
            old_cite_refs: List[nodes.citation_reference] = list(node.findall(is_citation_ref))
            new_cite_refs: List[nodes.citation_reference] = list(patch.findall(is_citation_ref))  # NOQA
            refname_ids_map = {}
            if not noqa and len(old_cite_refs) != len(new_cite_refs):
                old_cite_ref_rawsources = [ref.rawsource for ref in old_cite_refs]
                new_cite_ref_rawsources = [ref.rawsource for ref in new_cite_refs]
                logger.warning(__('inconsistent citation references in translated message.' +
                                  ' original: {0}, translated: {1}')
                               .format(old_cite_ref_rawsources, new_cite_ref_rawsources),
                               location=node, type='i18n', subtype='inconsistent_references')
            for oldc in old_cite_refs:
                refname_ids_map.setdefault(oldc["refname"], []).append(oldc["ids"])
            for newc in new_cite_refs:
                refname = newc["refname"]
                if refname_ids_map.get(refname):
                    newc["ids"] = refname_ids_map[refname].pop()

            # Original pending_xref['reftarget'] contain not-translated
            # target name, new pending_xref must use original one.
            # This code restricts to change ref-targets in the translation.
            old_xrefs = list(node.findall(addnodes.pending_xref))
            new_xrefs = list(patch.findall(addnodes.pending_xref))
            xref_reftarget_map = {}
            if not noqa and len(old_xrefs) != len(new_xrefs):
                old_xref_rawsources = [xref.rawsource for xref in old_xrefs]
                new_xref_rawsources = [xref.rawsource for xref in new_xrefs]
                logger.warning(__('inconsistent term references in translated message.' +
                                  ' original: {0}, translated: {1}')
                               .format(old_xref_rawsources, new_xref_rawsources),
                               location=node, type='i18n', subtype='inconsistent_references')

            def get_ref_key(node: addnodes.pending_xref) -> Optional[Tuple[str, str, str]]:
                case = node["refdomain"], node["reftype"]
                if case == ('std', 'term'):
                    return None
                else:
                    return (
                        node["refdomain"],
                        node["reftype"],
                        node['reftarget'],)

            for old in old_xrefs:
                key = get_ref_key(old)
                if key:
                    xref_reftarget_map[key] = old.attributes
            for new in new_xrefs:
                key = get_ref_key(new)
                # Copy attributes to keep original node behavior. Especially
                # copying 'reftarget', 'py:module', 'py:class' are needed.
                for k, v in xref_reftarget_map.get(key, {}).items():
                    if k not in EXCLUDED_PENDING_XREF_ATTRIBUTES:
                        new[k] = v

            # update leaves
            for child in patch.children:
                child.parent = node
            node.children = patch.children

            # for highlighting that expects .rawsource and .astext() are same.
            if isinstance(node, LITERAL_TYPE_NODES):
                node.rawsource = node.astext()

            if isinstance(node, nodes.image) and node.get('alt') != msg:
                node['uri'] = patch['uri']
                continue  # do not mark translated

            node['translated'] = True  # to avoid double translation

        if 'index' in self.config.gettext_additional_targets:
            # Extract and translate messages for index entries.
            for node, entries in traverse_translatable_index(self.document):
                new_entries: List[Tuple[str, str, str, str, str]] = []
                for type, msg, tid, main, _key in entries:
                    msg_parts = split_index_msg(type, msg)
                    msgstr_parts = []
                    for part in msg_parts:
                        msgstr = catalog.gettext(part)
                        if not msgstr:
                            msgstr = part
                        msgstr_parts.append(msgstr)

                    new_entries.append((type, ';'.join(msgstr_parts), tid, main, None))

                node['raw_entries'] = entries
                node['entries'] = new_entries

        # remove translated attribute that is used for avoiding double translation.
        for translated in self.document.findall(NodeMatcher(translated=Any)):  # type: Element  # NOQA
            translated.delattr('translated')


class RemoveTranslatableInline(SphinxTransform):
    """
    Remove inline nodes used for translation as placeholders.
    """
    default_priority = 999

    def apply(self, **kwargs: Any) -> None:
        from sphinx.builders.gettext import MessageCatalogBuilder
        if isinstance(self.app.builder, MessageCatalogBuilder):
            return

        matcher = NodeMatcher(nodes.inline, translatable=Any)
        for inline in list(self.document.findall(matcher)):  # type: nodes.inline
            inline.parent.remove(inline)
            inline.parent += inline.children


def setup(app: "Sphinx") -> Dict[str, Any]:
    app.add_transform(PreserveTranslatableMessages)
    app.add_transform(Locale)
    app.add_transform(RemoveTranslatableInline)

    return {
        'version': 'builtin',
        'parallel_read_safe': True,
        'parallel_write_safe': True,
    }