perf: deduplicate zh site_libs and images trees against root copies

.github/scripts/prune-zh-duplicates.py

@@ -0,0 +1,59 @@
+#!/usr/bin/env python3
+"""Delete the zh duplicate site_libs/ and images/ trees after refs were
+rewritten to the root trees by rewrite-zh-site-refs.py.
+
+Safety: aborts if any zh HTML reference still resolves inside the
+duplicate trees (resolution-based, so refs pointing at the root trees
+do not trigger a false abort).
+"""
+import os
+import re
+import shutil
+import sys
+
+site = sys.argv[1] if len(sys.argv) > 1 else "_site"
+zh_root = os.path.join(site, "zh")
+
+if not os.path.isdir(zh_root):
+    print(f"No zh directory at {zh_root}; nothing to prune")
+    sys.exit(0)
+
+# Any relative site_libs/ or images/ reference that resolves INSIDE the zh
+# tree means the rewrite step did not cover it — refuse to delete.
+ref_re = re.compile(r'["\'(=]\s*((?:\.\./)*)(?:site_libs|images)/[^"\'()\s>]+')
+leftover = 0
+for dirpath, _dirnames, filenames in os.walk(zh_root):
+    for fname in filenames:
+        if not fname.endswith(".html"):
+            continue
+        fpath = os.path.join(dirpath, fname)
+        with open(fpath, encoding="utf-8") as fh:
+            content = fh.read()
+        for match in ref_re.finditer(content):
+            full_ref = match.group(0).lstrip("\"'(= ")
+            resolved = os.path.normpath(
+                os.path.join(os.path.dirname(fpath), match.group(1) + full_ref)
+            )
+            if os.path.commonpath([resolved, zh_root]) == zh_root:
+                leftover += 1
+                print(f"LEFTOVER: {fpath} -> {full_ref}")
+                break
+
+if leftover:
+    print(f"ERROR: {leftover} zh page(s) still reference the duplicate trees; refusing to delete")
+    sys.exit(1)
+
+removed = 0
+for tree in ("site_libs", "images"):
+    tree_path = os.path.join(zh_root, tree)
+    if os.path.isdir(tree_path):
+        size_mb = sum(
+            os.path.getsize(os.path.join(dp, fn))
+            for dp, _, fns in os.walk(tree_path)
+            for fn in fns
+        ) // (1024 * 1024)
+        print(f"Deleting {tree} ({size_mb} MB)")
+        shutil.rmtree(tree_path)
+        removed += 1
+
+print(f"Pruned {removed} duplicate tree(s) from zh/")

.github/scripts/rewrite-zh-site-refs.py

@@ -0,0 +1,74 @@
+#!/usr/bin/env python3
+"""Rewrite zh-page asset references to share the root site trees.
+
+Background: babelquarto renders the zh pass into _site/zh/ and copies the
+shared site_libs/ and images/ trees into _site/zh/ as byte-identical
+duplicates of the root copies (_site/site_libs/, _site/images/), roughly
+94 MB per deploy.  This script rewrites every zh HTML reference that
+resolves under zh/site_libs/ or zh/images/ to the root-tree equivalent,
+so both zh duplicate trees can be deleted afterwards.
+
+Only references whose root-tree target exists are rewritten; anything
+else is left untouched and reported.  Run BEFORE deleting the duplicate
+trees (deletion is done by prune-zh-duplicates.sh).
+"""
+import os
+import re
+import sys
+
+SITE = sys.argv[1] if len(sys.argv) > 1 else "_site"
+ZH_ROOT = os.path.join(SITE, "zh")
+# Matches a relative reference whose first segment is site_libs/ or images/
+# (optionally preceded by ../).  Absolute, scheme-relative and srcset-less
+# attributes only; url(...) inside CSS also covered via the ( prefix.
+REF_RE = re.compile(
+    r'(?P<prefix>["\'(=])\s*(?P<dots>(?:\.\./)*)?(?P<path>(?:site_libs|images)/[^"\'()\s>]+)'
+)
+
+rewritten = 0
+missing = 0
+files_touched = 0
+
+
+def repl(match, rel_dir):
+    global rewritten, missing
+    dots = match.group("dots") or ""
+    path = match.group("path")
+    # Resolve the reference against the page directory (zh-relative).
+    zh_target = os.path.normpath(os.path.join(SITE, rel_dir, dots + path))
+    rel_to_zh_root = os.path.relpath(zh_target, ZH_ROOT)
+    if rel_to_zh_root.startswith(os.pardir):
+        missing += 1
+        return match.group(0)
+    root_target = os.path.join(SITE, rel_to_zh_root)
+    if not os.path.isfile(root_target):
+        missing += 1
+        return match.group(0)
+    fixed = os.path.relpath(root_target, os.path.join(SITE, rel_dir))
+    rewritten += 1
+    return match.group("prefix") + fixed
+
+
+for dirpath, _dirnames, filenames in os.walk(ZH_ROOT):
+    for fname in filenames:
+        if not fname.endswith(".html"):
+            continue
+        fpath = os.path.join(dirpath, fname)
+        with open(fpath, encoding="utf-8") as fh:
+            content = fh.read()
+        # Directory of the zh page, relative to _site, e.g. "zh" or "zh/Omics"
+        rel_dir = os.path.relpath(dirpath, SITE)
+
+        def sub(match):
+            return repl(match, rel_dir)
+
+        new_content = REF_RE.sub(sub, content)
+        if new_content != content:
+            files_touched += 1
+            with open(fpath, "w", encoding="utf-8") as fh:
+                fh.write(new_content)
+
+print(
+    f"Rewrote {rewritten} reference(s) across {files_touched} zh page(s); "
+    f"{missing} unresolved left as-is"
+)

.github/workflows/quarto-publish-update-cache.yml

@@ -524,6 +524,16 @@ jobs:
       - name: Point zh figure references at shared English figures
         run: python3 .github/scripts/rewrite-zh-figure-refs.py _site
 
+      # ── Deduplicate zh site_libs/ and images/ trees ──────────
+      # See quarto-publish.yml for full explanation.  babelquarto copies
+      # the shared site_libs/ and images/ trees into _site/zh/ as
+      # byte-identical duplicates of the root copies (~94 MB).  Rewrite zh
+      # HTML references to the root trees, then delete the duplicates.
+      - name: Deduplicate zh site_libs and images against root trees
+        run: |
+          python3 .github/scripts/rewrite-zh-site-refs.py _site
+          python3 .github/scripts/prune-zh-duplicates.py _site
+
       # ── Lazy-load images ─────────────────────────────────────
       # See quarto-publish.yml for full explanation. Native
       # loading="lazy" defers offscreen figure fetches, with

.github/workflows/quarto-publish.yml

@@ -717,6 +717,18 @@ jobs:
       - name: Point zh figure references at shared English figures
         run: python3 .github/scripts/rewrite-zh-figure-refs.py _site
 
+      # ── Deduplicate zh site_libs/ and images/ trees ──────────
+      # babelquarto copies the shared site_libs/ and images/ trees into
+      # _site/zh/ as byte-identical duplicates of the root copies (~94 MB
+      # per deploy).  Rewrite every zh HTML reference that resolves under
+      # zh/site_libs/ or zh/images/ to the root-tree equivalent, then
+      # delete the duplicate trees.  The prune script refuses to delete if
+      # any zh page still resolves a reference inside the zh trees.
+      - name: Deduplicate zh site_libs and images against root trees
+        run: |
+          python3 .github/scripts/rewrite-zh-site-refs.py _site
+          python3 .github/scripts/prune-zh-duplicates.py _site
+
       # ── Lazy-load images ─────────────────────────────────────
       # Figures dominate every page's payload. Native loading="lazy"
       # defers offscreen fetches (no JS, no layout shift - Quarto emits

Omics/CellChatCirclePlot.zh.qmd

@@ -112,7 +112,11 @@ cellchat <- aggregateNet(cellchat)
 
 整体圈图汇总所有推断的通讯交互。每个节点代表一种细胞类型；节点大小与该细胞类型参与的交互数量成正比。弧宽编码通讯强度。
 
-```{r fig1-overall-circle, fig.width=9, fig.height=5, warning=FALSE}
+```{r}
+#| label: fig-1-overall-circle
+#| fig-width: 9
+#| fig-height: 5
+#| warning: false
 #| fig-cap: "整体细胞-细胞通讯圈图"
 #| out.width: "95%"
 
@@ -143,7 +147,11 @@ par(mfrow = c(1, 1))
 
 逐行对权重矩阵进行子集化，分别展示每种细胞类型的输出信号。
 
-```{r fig2-individual-circle, fig.width=10, fig.height=10, warning=FALSE}
+```{r}
+#| label: fig-2-individual-circle
+#| fig-width: 10
+#| fig-height: 10
+#| warning: false
 #| fig-cap: "各细胞类型输出通讯圈图"
 #| out.width: "95%"
 
@@ -174,7 +182,11 @@ par(mfrow = c(1, 1))
 
 弦图使用 `circlize` 提供了另一种圆形视图，清晰展示细胞类型对之间的双向通讯量。
 
-```{r fig3-chord, fig.width=7, fig.height=7, warning=FALSE}
+```{r}
+#| label: fig-3-chord
+#| fig-width: 7
+#| fig-height: 7
+#| warning: false
 #| fig-cap: "细胞-细胞通讯弦图"
 #| out.width: "80%"
 
@@ -209,7 +221,11 @@ circos.clear()
 
 热图视图可快速识别主要的发送者-接收者细胞对及不对称通讯模式。
 
-```{r fig4-heatmap, fig.width=7, fig.height=6, warning=FALSE}
+```{r}
+#| label: fig-4-heatmap
+#| fig-width: 7
+#| fig-height: 6
+#| warning: false
 #| fig-cap: "细胞-细胞通讯热图"
 #| out.width: "80%"
 

Omics/ManhattanPlot.zh.qmd

@@ -208,7 +208,7 @@ ggplot()+
 由于ggplot2绘图是将一张张图叠加起来，因此可以将筛选后的SNP用geom_point叠加到原图上：
 
 ```{r}
-#| label: fig-2.3Hlight
+#| label: fig-2.3Highlight
 #| fig-cap: "高亮或标记snp"
 #| out.width: "95%"
 #| warning: false