X-Git-Url: https://git.mdrn.pl/librarian.git/blobdiff_plain/73858688886ec363048373f7e8253a33497fe3cd..261a3308d235fc252f30ae03603d0e964bb60223:/src/librarian/html.py?ds=inline diff --git a/src/librarian/html.py b/src/librarian/html.py index d262198..f0f11db 100644 --- a/src/librarian/html.py +++ b/src/librarian/html.py @@ -51,32 +51,48 @@ def transform_abstrakt(abstrakt_element): return re.sub(']*>', '', html) -def add_image_sizes(tree, gallery_path, gallery_url): - widths = [360, 600, 1200, 1800] - for ilustr in tree.findall('//ilustr'): +def add_image_sizes(tree, gallery_path, gallery_url, base_url): + widths = [360, 600, 1200, 1800, 2400] + + for i, ilustr in enumerate(tree.findall('//ilustr')): rel_path = ilustr.attrib['src'] - img = Image.open(gallery_path + rel_path) + img_url = six.moves.urllib.parse.urljoin(base_url, rel_path) + + f = six.moves.urllib.request.urlopen(img_url) + img = Image.open(f) + ext = {'GIF': 'gif', 'PNG': 'png'}.get(img.format, 'jpg') + srcset = [] + # Needed widths: predefined and original, limited by + # whichever is smaller. + img_widths = [ + w for w in + sorted( + set(widths + [img.size[0]]) + ) + if w <= min(widths[-1], img.size[0]) + ] + largest = None for w in widths: - if w < img.size[0]: + fname = '%d.W%d.%s' % (i, w, ext) + fpath = gallery_path + fname + if not os.path.exists(fpath): height = round(img.size[1] * w / img.size[0]) th = img.resize((w, height)) - - fname = ('.W%d.' % w).join(rel_path.rsplit('.', 1)) - th.save(gallery_path + fname) - srcset.append(" ".join(( - gallery_url + fname, - '%dw' % w - ))) - srcset.append(" ".join(( - gallery_url + rel_path, - '%dw' % img.size[0] - ))) + th.save(fpath) + th_url = gallery_url + fname + srcset.append(" ".join(( + th_url, + '%dw' % w + ))) + largest_url = th_url ilustr.attrib['srcset'] = ", ".join(srcset) - ilustr.attrib['src'] = gallery_url + rel_path + ilustr.attrib['src'] = largest_url + + f.close() -def transform(wldoc, stylesheet='legacy', options=None, flags=None, css=None, gallery_path='img/', gallery_url='img/'): +def transform(wldoc, stylesheet='legacy', options=None, flags=None, css=None, gallery_path='img/', gallery_url='img/', base_url='file://./'): """Transforms the WL document to XHTML. If output_filename is None, returns an XML, @@ -102,7 +118,12 @@ def transform(wldoc, stylesheet='legacy', options=None, flags=None, css=None, ga if not options: options = {} - add_image_sizes(document.edoc, gallery_path, gallery_url) + try: + os.makedirs(gallery_path) + except OSError: + pass + + add_image_sizes(document.edoc, gallery_path, gallery_url, base_url) css = ( css