@lark-apaas/coding-steering 0.1.18-dev.655b398 → 0.1.18-dev.7f786ca

package/steering/design-html/skills/pptx-style-extract/scripts/extract.py

@@ -11,6 +11,8 @@ as-is *and* re-encoded to a webp under ASSET_BUDGET_BYTES; without Pillow the
 transcode falls back to darwin `sips`, and failing that the row records
 transcode_blocked and extract.json records pillow_available: false.
 """
+import hashlib
+import io
 import json
 import os
 import re
@@ -162,6 +164,7 @@ def _webp_from_sips(part, raw, dst_base, budget):
 def export_media(pkg, images, outdir, pillow_ok, export_all=False):
     media_dir = os.path.join(outdir, 'media-out')
     os.makedirs(media_dir, exist_ok=True)
+    written = {}
     by_media = {i['media']: i for i in images}
     rows = []
     for part in pkg.media:
@@ -196,7 +199,15 @@ def export_media(pkg, images, outdir, pillow_ok, export_all=False):
             body = raw.decode('utf-8', 'replace')
             row['embedded_raster'] = 'data:image/' in body
         out_name = os.path.basename(part)
-        with open(os.path.join(media_dir, out_name), 'wb') as f:
+        dst_path = os.path.join(media_dir, out_name)
+        if dst_path in written:
+            # 两个 media part 落到同一个输出名。不覆盖——覆盖等于悄悄换掉一张图。
+            row['export_name_conflict'] = written[dst_path]
+            out_name = '%s~%d%s' % (os.path.splitext(out_name)[0], len(written),
+                                    os.path.splitext(out_name)[1])
+            dst_path = os.path.join(media_dir, out_name)
+        written[dst_path] = part
+        with open(dst_path, 'wb') as f:
             f.write(raw)
         row.update({'exported': True, 'out': 'media-out/' + out_name,
                     'out_bytes': len(raw), 'transcoded': False,
@@ -215,13 +226,11 @@ def export_media(pkg, images, outdir, pillow_ok, export_all=False):
         # rasterising it here would throw away the vector original.
         if needs and ext != 'svg':
             row['needs_transcode'] = needs
-            # The compressed file is only ever .webp (Pillow) or .jpg (sips). When
-            # the source already carries that extension the naive stem would point
-            # at the exported original and silently overwrite it.
-            stem = out_name.rsplit('.', 1)[0]
-            if ext in ('webp', 'jpg'):
-                stem += '-min'
-            base = os.path.join(media_dir, stem)
+            # 压缩产物必须用**完整原名**当前缀：只去掉扩展名的话，image4.png 的压缩版
+            # 就叫 image4.webp —— 而 ppt/media/image4.webp 往往是另一张真实存在的图，
+            # 两者抢同一个输出名，后写的覆盖先写的，且全程无人报错。实测某模板因此把
+            # 一张 109x109 的小图标当成了整页背景。
+            base = os.path.join(media_dir, out_name + '-min')
             done, why = None, 'pillow-unavailable' if not pillow_ok else None
             if pillow_ok:
                 try:
@@ -244,6 +253,214 @@ def export_media(pkg, images, outdir, pillow_ok, export_all=False):
     return rows
 
 
+# ---------------------------------------------------- rendered backgrounds
+def _is_full_canvas_picture(shape):
+    return (shape.get('kind') == 'pic' and shape.get('media')
+            and not shape.get('hidden')
+            and (shape.get('w_pct') or 0) >= 95
+            and (shape.get('h_pct') or 0) >= 95)
+
+
+def _picture_has_appearance(shape):
+    return bool(shape.get('flipH') or shape.get('flipV') or shape.get('rot')
+                or shape.get('crop')
+                or (shape.get('opacity') is not None and shape.get('opacity') != 1.0))
+
+
+def _effective_background(parts, bg_by_part):
+    background = None
+    for part in parts:
+        if bg_by_part.get(part):
+            background = bg_by_part[part]
+    return background
+
+
+def _draw_picture(canvas, shape, pkg, image_cache=None):
+    """Replay one p:pic's crop, flip, rotation and blip opacity with Pillow."""
+    from PIL import Image, ImageEnhance, ImageOps
+
+    media = shape.get('media')
+    if media not in pkg.names:
+        return False
+    image_cache = image_cache if image_cache is not None else {}
+    box = shape.get('box_unrotated') or shape.get('box') or {}
+    if media not in image_cache:
+        try:
+            source = Image.open(io.BytesIO(pkg.zip.read(media))).convert('RGBA')
+            source.load()
+            image_cache[media] = source
+        except Exception:
+            return False
+    picture = image_cache[media].copy()
+
+    crop = shape.get('crop') or {}
+    left = max(0.0, min(float(crop.get('l', 0) or 0), 100.0))
+    top = max(0.0, min(float(crop.get('t', 0) or 0), 100.0))
+    right = max(0.0, min(float(crop.get('r', 0) or 0), 100.0))
+    bottom = max(0.0, min(float(crop.get('b', 0) or 0), 100.0))
+    bounds = (
+        int(round(picture.width * left / 100.0)),
+        int(round(picture.height * top / 100.0)),
+        int(round(picture.width * (1.0 - right / 100.0))),
+        int(round(picture.height * (1.0 - bottom / 100.0))),
+    )
+    if bounds[2] > bounds[0] and bounds[3] > bounds[1]:
+        picture = picture.crop(bounds)
+
+    width = max(1, int(round(box.get('w') or 0)))
+    height = max(1, int(round(box.get('h') or 0)))
+    resampling = getattr(Image, 'Resampling', Image)
+    picture = picture.resize((width, height), resampling.LANCZOS)
+    if shape.get('flipH'):
+        picture = ImageOps.mirror(picture)
+    if shape.get('flipV'):
+        picture = ImageOps.flip(picture)
+    opacity = shape.get('opacity')
+    if opacity is not None:
+        picture.putalpha(ImageEnhance.Brightness(picture.getchannel('A')).enhance(
+            max(0.0, min(float(opacity), 1.0))))
+
+    rotation = float(shape.get('rot') or 0)
+    if rotation:
+        picture = picture.rotate(-rotation, resample=resampling.BICUBIC, expand=True)
+    center_x = float(box.get('x') or 0) + width / 2.0
+    center_y = float(box.get('y') or 0) + height / 2.0
+    x = int(round(center_x - picture.width / 2.0))
+    y = int(round(center_y - picture.height / 2.0))
+    layer = Image.new('RGBA', canvas.size)
+    layer.alpha_composite(picture, (x, y))
+    canvas.alpha_composite(layer)
+    return True
+
+
+def _draw_background(canvas, background, pkg, image_cache=None):
+    """Draw the effective p:bg underneath picture layers."""
+    from PIL import Image
+    from render_pages import _grad_image, _grad_stops, _rgba, background_css
+
+    css = background_css(background)
+    canvas.paste(_rgba(css, (255, 255, 255, 255)),
+                 (0, 0, canvas.width, canvas.height))
+    gradient = _grad_stops(css or '')
+    if gradient:
+        canvas.alpha_composite(
+            _grad_image(Image, canvas.width, canvas.height, gradient[0], gradient[1]))
+    if background and background.get('type') == 'image' and background.get('media'):
+        _draw_picture(canvas, {
+            'media': background['media'],
+            'box_unrotated': {'x': 0, 'y': 0, 'w': canvas.width, 'h': canvas.height},
+            'crop': background.get('crop') or {},
+        }, pkg, image_cache)
+
+
+def _save_background_webp(canvas, path):
+    """Prefer lossless output; fall back to a quality ladder when oversized."""
+    canvas.convert('RGB').save(path, 'WEBP', lossless=True, method=0)
+    if os.path.getsize(path) <= ASSET_BUDGET_BYTES:
+        return
+    rgb = canvas.convert('RGB')
+    for quality in (95, 90, 85, 75):
+        rgb.save(path, 'WEBP', quality=quality, method=6)
+        if os.path.getsize(path) <= ASSET_BUDGET_BYTES:
+            return
+
+
+def compose_backgrounds(pkg, graph, shapes, bg_by_part, units, outdir, pillow_ok):
+    """Flatten non-trivial full-canvas picture stacks into one web-safe asset."""
+    if not pillow_ok:
+        return {}, [], []
+    try:
+        from PIL import Image
+    except Exception:
+        return {}, [], []
+
+    by_part = defaultdict(list)
+    for shape in shapes:
+        by_part[shape['part']].append(shape)
+
+    chains = {}
+    for layout in pkg.layouts:
+        master = graph['master_of_layout'].get(layout)
+        show_master = pkg.xml(layout).get('showMasterSp', '1') != '0'
+        chains[layout] = [part for part in ((master if show_master else None), layout) if part]
+    for slide in pkg.slides:
+        layout = graph['layout_of_slide'].get(slide)
+        master = graph['master_of_layout'].get(layout)
+        show_master = pkg.xml(slide).get('showMasterSp', '1') != '0'
+        if layout:
+            show_master = show_master and pkg.xml(layout).get('showMasterSp', '1') != '0'
+        chains[slide] = [part for part in (
+            (master if show_master else None), layout, slide) if part]
+
+    media_dir = os.path.join(outdir, 'media-out')
+    os.makedirs(media_dir, exist_ok=True)
+    part_map, rows_by_media, images_by_media, rendered_specs = {}, {}, {}, {}
+    image_cache = {}
+    for target, chain in chains.items():
+        layers = [shape for part in chain for shape in by_part.get(part, [])
+                  if _is_full_canvas_picture(shape)]
+        if not layers or (len(layers) == 1 and not _picture_has_appearance(layers[0])):
+            continue
+        source_layers = [{
+            'part': shape['part'], 'media': shape['media'],
+            'crop': shape.get('crop'), 'flipH': bool(shape.get('flipH')),
+            'flipV': bool(shape.get('flipV')), 'rot': shape.get('rot') or 0,
+            'opacity': shape.get('opacity', 1.0),
+        } for shape in layers]
+        spec = json.dumps({
+            'background': _effective_background(chain, bg_by_part),
+            'layers': [{k: v for k, v in layer.items() if k != 'part'}
+                       for layer in source_layers],
+        }, ensure_ascii=False, sort_keys=True)
+        media = rendered_specs.get(spec)
+        if media is None:
+            canvas = Image.new('RGBA', (units.w, units.h))
+            _draw_background(
+                canvas, _effective_background(chain, bg_by_part), pkg, image_cache)
+            drawn = [shape for shape in layers
+                     if _draw_picture(canvas, shape, pkg, image_cache)]
+            if len(drawn) != len(layers):
+                continue
+            digest = hashlib.sha256(canvas.tobytes()).hexdigest()[:16]
+            media = 'generated/background/bg-composite-%s.webp' % digest
+            out = 'media-out/' + os.path.basename(media)
+            path = os.path.join(outdir, out)
+            if not os.path.exists(path):
+                _save_background_webp(canvas, path)
+            rendered_specs[spec] = media
+        else:
+            out = rows_by_media[media]['out']
+            path = os.path.join(outdir, out)
+        part_map[target] = media
+        row = rows_by_media.setdefault(media, {
+            'media': media, 'ext': 'webp', 'bytes': os.path.getsize(path),
+            'used_n': 0, 'reasons': ['background_composite'], 'candidate': True,
+            'exported': True, 'out': out, 'out_bytes': os.path.getsize(path),
+            'transcoded': False, 'generated': True, 'composited_from': source_layers,
+            'part_refs': [],
+        })
+        row['used_n'] += 1
+        row['part_refs'].append(target)
+        image = images_by_media.setdefault(media, {
+            'media': media, 'n': 0, 'boxes': [], 'exact_boxes': [],
+            'fullscreen': True, 'fullscreen_n': 0, 'fullscreen_top_cluster_n': 0,
+            'repeat_fixed': [], 'max_w_pct': 100.0, 'bleed': False,
+            'crop_variants': [], 'stitch_candidate': False,
+            'svg_companion': None, 'variant_group': [],
+        })
+        image['n'] += 1
+        image['fullscreen_n'] += 1
+        image['fullscreen_top_cluster_n'] += 1
+        image['boxes'].append({
+            'box': {'x': 0, 'y': 0, 'w': units.w, 'h': units.h},
+            'box_emu': {'x': 0, 'y': 0, 'cx': units.cx, 'cy': units.cy},
+            'count': 1, 'exact_count': 1, 'exact_variants': 1,
+            'w_pct': 100.0, 'h_pct': 100.0, 'parts': [target],
+            'layers': ['composite'],
+        })
+    return part_map, list(rows_by_media.values()), list(images_by_media.values())
+
+
 # ---------------------------------------------------------------- form hint
 # PowerPoint 出厂版式名（中英两套）。设计师起的名字是「这一页干什么用」，
 # 出厂名只是「这个占位符组合叫什么」——后者不算模板声明了页型。
@@ -466,6 +683,30 @@ REF_NOTES = """# ref/ 审计层说明（S12）
 """
 
 
+def dump_source(pkg, outdir):
+    """把 PPTX 解压后的原文原样落到 ref/source/。
+
+    普查是有损的：它按既定口径抽数，抽不到的、口径外的东西就没了。遇到判断不了的
+    情况（这个形状为什么这么摆、某个字段是什么意思），能直接翻原始 XML 比对着二手
+    数据猜可靠得多。只进中间产物，交付包里没有。
+    """
+    dst = os.path.abspath(os.path.join(outdir, 'ref', 'source'))
+    n = 0
+    for name in pkg.zip.namelist():
+        if name.endswith('/'):
+            continue
+        p = os.path.abspath(os.path.join(dst, *name.split('/')))
+        # zip 条目名是文件里写什么就是什么，带 ../ 就能写到 ref/source 外面去
+        # （pptx 是用户上传的，当不可信输入处理）。落在目录外的条目一律不落盘。
+        if not p.startswith(dst + os.sep):
+            continue
+        os.makedirs(os.path.dirname(p), exist_ok=True)
+        with open(p, 'wb') as f:
+            f.write(pkg.zip.read(name))
+        n += 1
+    return n
+
+
 def write_ref(outdir, payload, ref_data, units, parts_ordered, shapes, bg_by_part):
     ref = os.path.join(outdir, 'ref')
     os.makedirs(ref, exist_ok=True)
@@ -604,11 +845,16 @@ def extract(pptx, outdir, export_all=False):
     t = mark('derived_censuses', t)
 
     media_rows = export_media(pkg, images, outdir, pillow_ok, export_all)  # S9
+    background_composites, composite_media, composite_images = compose_backgrounds(
+        pkg, graph, shapes, bg_by_part, units, outdir, pillow_ok)
+    media_rows += composite_media
+    images += composite_images
     t = mark('S9_media_export', t)
 
     # S5b + S14 run after S9 because the palette only covers exported assets.
-    fps = media_fingerprints(pkg, [i['media'] for i in images], pillow_ok)
-    clusters, cluster_evidence = content_clusters(images, fps)
+    source_images = [i for i in images if i['media'] in pkg.names]
+    fps = media_fingerprints(pkg, [i['media'] for i in source_images], pillow_ok)
+    clusters, cluster_evidence = content_clusters(source_images, fps)
     exported = {m['media'] for m in media_rows if m.get('exported')}
     image_palette(images, fps, exported)
     t = mark('S5b_S14_content_palette', t)
@@ -680,6 +926,7 @@ def extract(pptx, outdir, export_all=False):
             'Pillow unavailable: byte-identical media only, no perceptual merging'),
         'palette_available': pillow_ok,
         'media': media_rows,
+        'background_composites': background_composites,
         # S4 shape-facts dominate this file (roughly half its bytes) and stage 2 reads
         # them only when a derived statistic needs backing evidence, so they live in
         # a sidecar and extract.json keeps just the pointer plus the derived censuses.
@@ -690,6 +937,7 @@ def extract(pptx, outdir, export_all=False):
             'slides': len(pkg.slides), 'layouts': len(pkg.layouts),
             'masters': len(pkg.masters), 'themes': len(pkg.themes),
             'media': len(pkg.media),
+            'background_composites': len(composite_media),
             'shapes_total': len(shapes),
             'shapes_kept': len(kept_shapes),
             'shapes_dropped_off_canvas': len(dropped_shapes),
@@ -699,8 +947,10 @@ def extract(pptx, outdir, export_all=False):
                                         if r.get('placement') == 'inherited'),
             'content_clusters': len(clusters),
             'content_clusters_multi_media': sum(1 for c in clusters if c['member_n'] > 1),
-            'media_exported': sum(1 for m in media_rows if m.get('exported')),
-            'media_transcoded': sum(1 for m in media_rows if m.get('transcoded')),
+            'media_exported': sum(1 for m in media_rows
+                                  if m.get('exported') and not m.get('generated')),
+            'media_transcoded': sum(1 for m in media_rows
+                                    if m.get('transcoded') and not m.get('generated')),
             'media_transcode_blocked': sum(1 for m in media_rows
                                            if m.get('transcode_blocked')),
             'media_over_budget': sum(1 for m in media_rows if m.get('over_budget')),
@@ -757,6 +1007,7 @@ def extract(pptx, outdir, export_all=False):
     with open(out_json, 'w') as f:
         json.dump(payload, f, ensure_ascii=False, indent=1)
     write_ref(outdir, payload, ref_data, units, parts_ordered, shapes, bg_by_part)
+    src_n = dump_source(pkg, outdir)
     perf['total'] = round(time.time() - t0, 3)
     with open(os.path.join(outdir, 'ref', 'perf.json'), 'w') as f:
         json.dump(perf, f, ensure_ascii=False, indent=1)
@@ -768,9 +1019,11 @@ def extract(pptx, outdir, export_all=False):
     print('  shapes %d kept / %d dropped off-canvas / %d bleed / %d clamped'
           % (payload['counts']['shapes_kept'], payload['counts']['shapes_dropped_off_canvas'],
              payload['counts']['shapes_bleed'], payload['counts']['shapes_clamped']))
-    print('  colors %d  fonts %d  images %d  media exported %d/%d  guides %d'
+    print('  colors %d  fonts %d  images %d  media exported %d/%d + %d composites  guides %d'
           % (len(color_freq), len(fonts), len(images),
-             payload['counts']['media_exported'], len(pkg.media), len(guides)))
+             payload['counts']['media_exported'], len(pkg.media),
+             len(composite_media), len(guides)))
+    print('  ref/source/ 原文 %d 个部件（判断不了时可直接翻）' % src_n)
     print('  extract.json %.1f KB  total %.2fs'
           % (os.path.getsize(out_json) / 1024.0, perf['total']))
     return payload
@@ -796,9 +1049,18 @@ def main(argv):
     if '--no-draft' not in flags:
         import subprocess
         d = os.path.join(os.path.dirname(os.path.abspath(__file__)), 'draft.py')
+        sys.stdout.flush()
         r = subprocess.run([sys.executable, d, args[1]])
         if r.returncode:
-            print('  ⚠ 草案生成失败，可单独重跑 draft.py 看报错')
+            # 普查产物已经齐了，缺的只是草案。重跑整条抽取会同样失败在这一步，
+            # 所以给一个区别于成功的终止哨兵，并指明只需重跑 draft.py。
+            print('EXTRACT_PARTIAL 普查产物齐全，草案生成失败：'
+                  'python3 -B scripts/draft.py %s 单独重跑看报错' % args[1])
+            sys.stdout.flush()
+            return 1
+    # 最后一行是终止哨兵：stdout 被截断时退出码仍可能是 0，两个哨兵都没有就是没跑完。
+    print('EXTRACT_OK %s' % os.path.join(args[1], 'l-out'))
+    sys.stdout.flush()
     return 0
 
 

package/steering/design-html/skills/pptx-style-extract/scripts/ooxml.py

@@ -519,8 +519,25 @@ def read_txbody(tx, ctx, kind='txBody'):
                 ins[k] = ctx.units.px(bp.get(k))
         if ins:
             b['insets_px'] = ins
-        if bp.find('a:normAutofit', NS) is not None:
+        na = bp.find('a:normAutofit', NS)
+        if na is not None:
             b['autofit'] = 'norm'
+            # normAutofit 用 fontScale / lnSpcReduction（单位 1/1000 %）把大字缩进小框——
+            # 章节大号数字就靠它让 160px 的字装进 144px 的框。只记 autofit 存在、丢掉
+            # fontScale，消费端就拿到未缩放字号 + 原始框高，字比框高，渐变裁切把溢出的
+            # 底部切成透明。缺省即 100%（无缩放）。
+            fs = na.get('fontScale')
+            if fs is not None:
+                try:
+                    b['font_scale'] = round(int(fs) / 100000.0, 4)
+                except (TypeError, ValueError):
+                    pass
+            lsr = na.get('lnSpcReduction')
+            if lsr is not None:
+                try:
+                    b['ln_spc_reduction'] = round(int(lsr) / 100000.0, 4)
+                except (TypeError, ValueError):
+                    pass
         elif bp.find('a:spAutoFit', NS) is not None:
             b['autofit'] = 'shape'
         if b: