fix(render): match real dj-root/dj-view attribute, not substring, when picking VDOM source (#1746) by johnrtipton · Pull Request #1747 · djust-org/djust

johnrtipton · 2026-06-07T02:29:57Z

Root cause

python/djust/mixins/template.py:161-165 (get_template()) chose the VDOM source with a naive substring check:

vdom_source = (
    template_source                                   # the CHILD template
    if ("dj-root" in template_source or "dj-view" in template_source)
    else resolved                                     # the full resolved (base+child) template
)

"dj-root" in template_source matches the token anywhere — including documentation/example code that merely displays dj-root/dj-view, or another word containing it as a substring (adj-view). When the real <div dj-root> lives in the base template (a common layout) and the child only mentions the tokens in text, this wrongly picks the child as vdom_source; _extract_liveview_root_with_wrapper finds no real dj-root there → wrong VDOM template → render_full_template's shell/dj-root replacement nests the whole page (two <!DOCTYPE>/two <footer>, truncated <script>, broken dj-navigate).

Fix

Replace the substring check with the anchored-attribute regexes already defined in the module (_DJ_ROOT_RE / _DJ_VIEW_RE), which require a REAL <div ... dj-root/dj-view ...> tag:

vdom_source = (
    template_source
    if (_DJ_ROOT_RE.search(template_source) or _DJ_VIEW_RE.search(template_source))
    else resolved
)

Both regexes are module-level and already in scope at line 161. A child that merely displays the token in text no longer matches; the legitimate "child declares its own real <div dj-root>" case (the reason the heuristic exists) still picks template_source.

Variant repro table (regression test `test_render_full_template_djroot_substring_1746.py`)

child content	`<!DOCTYPE>` before	after
`<p>hello</p>` (control)	1	1
`<p>dj-view</p>` (plain text)	2	1
`<p>dj-root</p>` (plain text)	2	1
`<p><code>dj-view</code> and <code>dj-root</code></p>`	2	1
`<p>adj-view</p>` (substring)	2	1
`<p>dj-click</p>` (other dj-* attr, control)	1	1

Plus a control proving the good path still works: a child that DOES declare its own real <div dj-root> still renders exactly one <!DOCTYPE>/<footer>.

Gate-off (#254)

Reverting the predicate to the naive substring (if ("dj-root" in ... or "dj-view" in ...)) makes the 4 behavior-meaningful cases (dj-view/dj-root/code/adj-view) fail again (2 DOCTYPEs); restoring the regex makes them pass. The 3 control cases pass either way — confirming the 4 are non-tautological.

Symbol-migration grep (canon)

grep -rn '"dj-root" in\|"dj-view" in' python/djust/ found one other production instance: python/djust/checks.py:3341-3342 (_check_view_root_same_element, T005). It is legitimately fine — it operates on text already matched to be inside an HTML tag (tag_re extracts <...> tags first, then checks substring within each tag), so it cannot match displayed text the way the render path did. Not in scope (#1079). The remaining matches are test assertions (test_mcp.py).

Test results

New regression file: 7/7 pass (RED→GREEN confirmed).
test_ssr_render_normalization_1737.py + test_template_inheritance.py + test_context_processors_in_includes_1722.py + test_model_fk_rendering.py: 30/30 pass (the Initial SSR render skips the comment/whitespace normalization that render_with_diff applies → first-hydration morphChildren re-render (flash) #1737 parity tests stay green).
python/djust/tests/: 3099 passed, 3 skipped.
python/tests/ + tests/unit/: 4206 passed; 4 pre-existing cross-dir-ordering pollution failures in tests/unit/test_demo_views.py (urlconf registration) — verified pre-existing (pass in isolation, and fail identically on a clean origin/main tree with my changes stashed). Unrelated to this fix.

Closes #1746

…n picking VDOM source (#1746) get_template() chose the VDOM source with a naive substring check ('dj-root' in template_source or 'dj-view' in template_source). When the real <div dj-root> lives in the BASE template and the CHILD only mentions the tokens as displayed text (e.g. a code example) — or as a substring of another word like adj-view — the check wrongly picked the child as the VDOM source. _extract_liveview_root_with_wrapper found no real dj-root there, so render_full_template's shell/dj-root replacement nested the whole page (two <!DOCTYPE>/two <footer>, truncated <script>, broken dj-navigate morphs). Replace the substring check with the anchored-attribute regexes already defined in the module (_DJ_ROOT_RE / _DJ_VIEW_RE), which require a REAL <div ... dj-root/dj-view ...> tag — so a child that merely displays the token in text won't match. The legitimate 'child declares its own real <div dj-root>' case (the reason the heuristic exists) still picks the child. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

…er Unreleased

johnrtipton · 2026-06-07T02:39:49Z

Stage-11 Adversarial Review — PR #1747 (fix #1746)

Verdict: 🟢 APPROVE. The change is correct, well-targeted, non-tautological, and — the key concern of this review — narrowing the predicate from a substring check to the <div>-anchored regexes does not break any legitimate root-declaration shape. In fact the <body dj-view> shape it was feared to regress is one the fix also fixes.

Pre-review gates

Stale-base (Full HTML update path mangles page rendering #250): PASS. BEHIND=0 vs origin/main. git log origin/main..HEAD shows exactly the 2 PR commits.
Two-commit shape: PASS. 25c15909 = fix (mixins/template.py) + test (test_render_full_template_djroot_substring_1746.py), no CHANGELOG. 26756b1a = CHANGELOG-only, well-formed ### Fixed entry under [Unreleased].

Empirical verification (all run)

New test passes: 7/7 green on HEAD.
Gate-off (non-tautological): reverting only the predicate back to the substring in check → exactly 4 fail (dj_view_text, dj_root_text, dj_root_view_code, substring_adj_view, all 2 <!DOCTYPE>), 3 controls pass (control_plain, control_dj_click, child-own-<div dj-root>). The tests genuinely exercise the change.
No regression: test_ssr_render_normalization_1737.py + test_template_inheritance.py + test_context_processors_in_includes_1722.py → 26 passed.

THE KEY EDGE CASE — does div-anchoring break non-div / `<body dj-view>` roots? No — verified empirically and structurally.

The predicate is just a gate in front of _extract_liveview_root_with_wrapper, which is the function that actually consumes the chosen vdom_source. That extractor (template.py:739-741) is already div-only: <div\s+[^>]*dj-root[^>]*> / <div\s+[^>]*dj-view[^>]*>. So is the streaming chunker (_DJ_ROOT_RE at template.py:434) and the shell-replacement path (template.py:972). The entire Python render/extraction pipeline has always been div-only. The new gate is now consistent with the extractor it feeds — the old substring gate was the lone outlier that could select a non-div child the extractor couldn't then use.

Empirical harness (base owns real <div dj-root>; child shapes vary), NEW code vs OLD substring code:

child root shape	NEW (`<div>`-regex)	OLD (substring)
`<body dj-view="...">` in child	DOCTYPE=1 ✅	DOCTYPE=2 ❌ (double-render)
`<section dj-root>` in child	DOCTYPE=1 ✅	DOCTYPE=1 ✅
`<div dj-root>` in child (control)	DOCTYPE=1 ✅	DOCTYPE=1 ✅

The <body dj-view> case was broken under the old code (the substring matched dj-view, selected the child as vdom_source, extractor found no <div...dj-view>, page double-rendered) and is fixed by this PR. <section dj-root> worked under both because the substring matched but the extractor's div-regex found nothing and fell through to the resolved full-template path. Conclusion: the framework's dj-root is always <div dj-root>; non-div roots are uniformly unsupported across extraction + streaming, and this fix removes a gate inconsistency rather than introducing one. (The Rust VDOM find_liveview_root at crates/djust_vdom/src/parser.rs:379-396 is element-agnostic, but it runs on already-rendered/extracted HTML, not this selection path, so it's not in scope.)

Regex on template source (item 3): correct

template_source = template.template.source is the raw template with {% %}/{{ }} tags. Probed:

<div dj-root>{{ count }}</div> → match ✅
{% if x %}<div dj-root>{% endif %} → match ✅
<div data-dj-root> / <p>adj-view</p> / displayed-text / <span dj-root> → correctly no-match ✅
newline-separated attrs → match ✅

🟡 One minor, contrived narrowing (non-blocking)

<div {% if a %}dj-root{% endif %}> (dj-root attached conditionally via a template tag with no trailing space) → the OLD substring matched, the NEW regex does not (the { of {% endif %} immediately follows dj-root, failing the (?=[\s=>/]) lookahead). This is a non-idiomatic shape — djust authors write a literal <div dj-root> — and a trailing space (dj-root {% endif %}) makes it match again. Worth a one-line awareness note; not a real-world blocker.

Symbol-migration grep (item 4): complete

Only other production hit is checks.py:3341-3342 ("dj-view" in tag / "dj-root" in tag). Implementer's "fine" judgment is correct: tag is match.group(0) from tag_re = <[a-zA-Z][^>]*>, i.e. already a real extracted tag — displayed text / adj-view can't reach it. It's the T005 system-check linter, not a render path. Re-grepped both quote styles across python/djust/ — no other render/extraction-path instance missed.

Pre-existing failures (item 5): confirmed pre-existing, NOT caused by this PR

tests/unit/test_demo_views.py 4 failures (test_pwa_view_in_urlconf, test_tenant_view_in_urlconf, test_service_worker_url_resolves, test_manifest_url_resolves):

Pass 43/43 in isolation on PR HEAD.
Appear only in cross-dir invocation (pytest python/ tests/unit/test_demo_views.py): 4 failed / 5615 passed on HEAD.
Identical 4 failures on origin/main with the PR removed: 4 failed / 5608 passed. URLconf-pollution from sibling dirs, unrelated to this fix.

Counts

New test: 7 passed. Gate-off: 4 failed / 3 passed (correct). No-regression suite: 26 passed. Cross-dir: 4 pre-existing fails on both HEAD and main.

Working tree restored to clean PR-branch HEAD (26756b1a); pre-existing untracked scripts/antigravity_pipeline_run.py left alone.

johnrtipton · 2026-06-07T02:40:56Z

Retrospective — PR #1747 (#1746)

Task: fix render_full_template double-render when a LiveView child template contains literal dj-root/dj-view text.

Quality: 5/5

One-line root-cause fix with a full variant repro, gate-off, and an edge-case review that turned up a bonus fix.

What went well

Root cause was nailed before coding (template.py:161-165 naive "dj-root" in template_source substring gate) via the downstream-driven investigation on djust.org, with a minimal in-framework repro (the variant table) filed in render_full_template double-renders the whole page when a LiveView's child template contains the literal text "dj-root"/"dj-view" (e.g. in a code example) #1746. Implementer verified the line + that _DJ_ROOT_RE/_DJ_VIEW_RE were in scope, then made the one-line change.
The fix is consistency, not just a patch: the entire Python render/extraction pipeline (_extract_liveview_root_with_wrapper, the streaming chunker, the shell-replacement) was already <div>-only; the substring gate was the lone outlier that could pick a non-div child the extractor couldn't use. Anchoring the gate to _DJ_ROOT_RE/_DJ_VIEW_RE makes it consistent with everything downstream.
Stage-11 edge-case hunt found a bonus: narrowing to the div-regex doesn't just avoid breaking <body dj-view> children — it fixes them (they double-rendered under the old substring gate too). No legitimate root shape regressed.
Gate-off (Add remaining 13 Django built-in template filters #254): reverting the predicate fails exactly the 4 behavior-meaningful cases (dj-view/dj-root/code/adj-view = 2 DOCTYPEs); 3 controls pass either way — non-tautological.
Symbol-migration grep (canon): found the only other "dj-root" in/"dj-view" in production hit (checks.py:3341, T005) and correctly judged it safe (operates on extracted tags, not displayed text).

What didn't

Nothing blocking. One 🟡: a contrived <div {% if a %}dj-root{% endif %}> template shape no longer matches the lookahead — non-idiomatic, non-blocking.
4 pre-existing tests/unit/test_demo_views.py cross-dir URLconf-pollution failures (pass in isolation; identical on clean main) — unrelated; same class as tech-debt: fix 2 pre-existing test-pollution failures in test_checks.py + audit module-level caches for test-reset #1741.

Process finding

A downstream consumer (djust.org) surfaced a core-framework render bug that the framework's own tests + demo didn't catch — because no test rendered a LiveView whose reactive content displays djust attribute names. The minimal in-framework repro (filed in #1746 before the fix) is what made this a clean one-line fix instead of a guess. Reinforces: when a downstream app hits a framework bug, distill it to a minimal in-framework repro first.

Verified

New regression test 7/7 (variant table: before dj-view/dj-root/code/adj-view=2 DOCTYPE, after all=1); Initial SSR render skips the comment/whitespace normalization that render_with_diff applies → first-hydration morphChildren re-render (flash) #1737 + inheritance + Context-processor theme vars ({{ theme_panel }} / {{ theme_head }}) are empty inside nested {% include %} (incomplete #233) #1722 render tests 26/26 (no core-render regression); python/djust/tests/ 3099 passed. All CI green (py3.12, demo-checks, rust-tests). Two-commit shape intact.

RETRO_COMPLETE

…dening) Drains #1751 (close-side </div> + scanner consolidation) and #1752 items 2-3 into v1.0.3; records #1747/#1750/#1748/#1753 already merged toward it. Audit-bypass-reason: docs-only ROADMAP update via pipeline-drain skill (no retro needed)

johnrtipton and others added 2 commits June 6, 2026 22:26

docs(changelog): add #1746 render_full_template double-render fix und…

26756b1

…er Unreleased

johnrtipton merged commit 0503818 into main Jun 7, 2026
13 checks passed

johnrtipton deleted the fix/render-full-template-naive-djroot-substring-1746 branch June 7, 2026 02:40

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

fix(render): match real dj-root/dj-view attribute, not substring, when picking VDOM source (#1746)#1747

fix(render): match real dj-root/dj-view attribute, not substring, when picking VDOM source (#1746)#1747
johnrtipton merged 2 commits into
mainfrom
fix/render-full-template-naive-djroot-substring-1746

johnrtipton commented Jun 7, 2026

Uh oh!

johnrtipton commented Jun 7, 2026

Uh oh!

Uh oh!

johnrtipton commented Jun 7, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Conversation

johnrtipton commented Jun 7, 2026

Root cause

Fix

Variant repro table (regression test test_render_full_template_djroot_substring_1746.py)

Gate-off (#254)

Symbol-migration grep (canon)

Test results

Uh oh!

johnrtipton commented Jun 7, 2026

Stage-11 Adversarial Review — PR #1747 (fix #1746)

Pre-review gates

Empirical verification (all run)

THE KEY EDGE CASE — does div-anchoring break non-div / <body dj-view> roots? No — verified empirically and structurally.

Regex on template source (item 3): correct

🟡 One minor, contrived narrowing (non-blocking)

Symbol-migration grep (item 4): complete

Pre-existing failures (item 5): confirmed pre-existing, NOT caused by this PR

Counts

Uh oh!

Uh oh!

johnrtipton commented Jun 7, 2026

Retrospective — PR #1747 (#1746)

Quality: 5/5

What went well

What didn't

Process finding

Verified

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Variant repro table (regression test `test_render_full_template_djroot_substring_1746.py`)

THE KEY EDGE CASE — does div-anchoring break non-div / `<body dj-view>` roots? No — verified empirically and structurally.