Limit image size, fixes #4464.
[librarian.git] / src / librarian / document.py
index 1bd249d..7780b61 100644 (file)
@@ -1,16 +1,30 @@
+# This file is part of Librarian, licensed under GNU Affero GPLv3 or later.
+# Copyright © Fundacja Wolne Lektury. See NOTICE for more information.
+#
+from collections import Counter
+import gettext
+import os
+import re
+import urllib.request
 from lxml import etree
 from lxml import etree
-from .builders import get_builder_class
 from .parser import parser
 from .parser import parser
-from . import dcparser
+from . import dcparser, DCNS, RDFNS, DirDocProvider
+from .functions import lang_code_3to2
 
 
 class WLDocument:
 
 
 class WLDocument:
-    def __init__(self, tree=None, filename=None):
-        if filename is not None:
-            tree = etree.parse(filename, parser=parser)
+    def __init__(self, filename=None, url=None, provider=None):
+        source = filename or urllib.request.urlopen(url)
+        tree = etree.parse(source, parser=parser)
         self.tree = tree
         tree.getroot().document = self
         self.tree = tree
         tree.getroot().document = self
-        self.base_meta = dcparser.BookInfo({}, {}, validate_required=False)
+        self.base_meta = dcparser.BookInfo({}, {
+            DCNS('language'): ["pol"],
+        }, validate_required=False)
+
+        self.provider = provider if provider is not None else DirDocProvider('.')
+
+        self.tree.getroot().validate()
 
     @property
     def meta(self):
 
     @property
     def meta(self):
@@ -19,6 +33,158 @@ class WLDocument:
         return self.tree.getroot().meta
         return master.meta
 
         return self.tree.getroot().meta
         return master.meta
 
-    def build(self, builder_id, **kwargs):
-        return get_builder_class(builder_id)().build(self, **kwargs)
-        
+    @property
+    def children(self):
+        for part_uri in self.meta.parts or []:
+            with self.provider.by_slug(part_uri.slug) as f:
+                try:
+                    yield type(self)(filename=f, provider=self.provider)
+                except Exception as e:
+
+                    yield e
+
+    def build(self, builder, base_url=None, **kwargs):
+        return builder(base_url=base_url).build(self, **kwargs)
+
+    def assign_ids(self, existing=None):
+        # Find all existing IDs.
+        existing = existing or set()
+        que = [self.tree.getroot()]
+        while que:
+            item = que.pop(0)
+            try:
+                item.normalize_insides()
+            except AttributeError:
+                pass
+            existing.add(item.attrib.get('id'))
+            que.extend(item)
+
+        i = 1
+        que = [self.tree.getroot()]
+        while que:
+            item = que.pop(0)
+            que.extend(item)
+            if item.attrib.get('id'):
+                continue
+            if not getattr(item, 'SHOULD_HAVE_ID', False):
+                continue
+            while f'e{i}' in existing:
+                i += 1
+            item.attrib['id'] = f'e{i}'
+            i += 1
+
+    def _compat_assign_ordered_ids(self):
+        """
+        Compatibility: ids in document order, to be roughly compatible with legacy
+        footnote ids. Just for testing consistency, change to some sane identifiers
+        at convenience.
+        """
+        EXPR = re.compile(r'/\s', re.MULTILINE | re.UNICODE)
+        def _compat_assign_ordered_ids_in_elem(elem, i):
+            if isinstance(elem, etree._Comment): return i
+            elem.attrib['_compat_ordered_id'] = str(i)
+            i += 1
+            if getattr(elem, 'HTML_CLASS', None) == 'stanza':
+                if elem.text:
+                    i += len(EXPR.split(elem.text)) - 1
+                for sub in elem:
+                    i = _compat_assign_ordered_ids_in_elem(sub, i)
+                    if sub.tail:
+                        i += len(EXPR.split(sub.tail)) - 1
+            else:
+                if elem.tag in ('uwaga', 'extra'):
+                    return i
+                for sub in elem:
+                    i = _compat_assign_ordered_ids_in_elem(sub, i)
+            return i
+
+        _compat_assign_ordered_ids_in_elem(self.tree.getroot(), 4)
+
+    def _compat_assign_section_ids(self):
+        """
+        Ids in master-section order. These need to be compatible with the
+        #secN anchors used by WL search results page to link to fragments.
+        """
+        def _compat_assigns_section_ids_in_elem(elem, prefix='sec'):
+            for i, child in enumerate(elem):
+                idfier = '{}{}'.format(prefix, i + 1)
+                try:
+                    child.attrib['_compat_section_id'] = idfier
+                except:
+                    pass
+                _compat_assigns_section_ids_in_elem(child, idfier + '-')
+        _compat_assigns_section_ids_in_elem(self.tree.getroot().master)
+
+
+    def editors(self):
+        persons = set(self.meta.editors
+                      + self.meta.technical_editors)
+        for child in self.children:
+            persons.update(child.editors())
+        if None in persons:
+            persons.remove(None)
+        return persons
+
+    def references(self):
+        return self.tree.findall('.//ref')
+
+    def get_statistics(self):
+        def count_text(text, counter, in_fn=False, stanza=False):
+            if text:
+                text = re.sub(r'\s+', ' ', text)
+
+                chars = len(text) if text.strip() else 0
+                words = len(text.split()) if text.strip() else 0
+
+                counter['chars_with_fn'] += chars
+                counter['words_with_fn'] += words
+                if not in_fn:
+                    counter['chars'] += chars
+                    counter['words'] += words
+                if not stanza:
+                    counter['chars_out_verse_with_fn'] += chars
+                    if not in_fn:
+                        counter['chars_out_verse'] += chars
+
+        def count(elem, counter, in_fn=False, stanza=False):
+            if elem.tag in (RDFNS('RDF'), 'nota_red', 'abstrakt', 'uwaga', 'ekstra'):
+                return
+            if not in_fn and elem.tag in ('pa', 'pe', 'pr', 'pt', 'motyw'):
+                in_fn = True
+            if elem.tag == 'strofa':
+                # count verses now
+                #verses = len(elem.findall('.//br')) + 1
+                verses = list(elem.get_verses())
+                counter['verses_with_fn'] += len(verses)
+                if not in_fn:
+                    counter['verses'] += len(verses)
+                stanza = True
+
+                for child in verses:
+                    count(child, counter, in_fn=in_fn, stanza=True)
+            else:
+                count_text(elem.text, counter, in_fn=in_fn, stanza=stanza)
+                for child in elem:
+                    count(child, counter, in_fn=in_fn, stanza=stanza)
+                    count_text(child.tail, counter, in_fn=in_fn, stanza=stanza)
+
+        data = {
+            "self": Counter(),
+            "parts": [],
+            "total": {
+            }
+        }
+
+        count(self.tree.getroot(), data['self'])
+        for k, v in data['self'].items():
+            data['total'][k] = v
+
+        for part in self.children:
+            if isinstance(part, Exception):
+                data['parts'].append((None, {'error': part}))
+            else:
+                data['parts'].append((part, part.get_statistics()))
+                for k, v in data['parts'][-1][1]['total'].items():
+                    data['total'][k] = data['total'].get(k, 0) + v
+
+        return data