ENG-9414: Remove hardcoded docs urls

[carlosabadia]

No description provided.

[linear]

ENG-9414 Remove hardcoded docs urls

[codspeed-hq]

Merging this PR will not alter performance

✅ 9 untouched benchmarks

---

_{Comparing carlos/bring-back-docs-links (6869b0d) with main (44271c5)}

[greptile-apps]

Greptile Summary

This PR replaces ~230 hardcoded /docs/... URL strings across 68 documentation markdown files with dynamic {namespace.path} expressions, backed by a new _resolve_link_target helper that uses re.sub + eval to expand brace expressions against the doc execution environment. The core rendering change converts _render_spans from a free function to an instance method so link resolution can reach self.env, and a fix in __init__.py ensures top-level page namespaces retain their .path attribute after being shadowed by subpage SimpleNamespace objects.

Confidence Score: 4/5

Safe to merge; changes are confined to internal documentation infrastructure with no runtime product impact.

Only P2 findings: a soft eval sandbox (mitigated by the fully repo-controlled input) and missing error context for NameError/AttributeError in the link resolver. No logic bugs or data-path issues identified.

docs/app/reflex_docs/docgen_pipeline.py — the _resolve_link_target eval call and error handling.

Security Review

• eval with {"__builtins__": {}} in _resolve_link_target (docgen_pipeline.py) provides only a soft sandbox. Expressions come exclusively from repo-controlled markdown files today, but objects in env (e.g. the rx module) expose full Python object graphs that can be used to escape the restriction via attribute traversal. If inputs ever became user-supplied this would be a code-execution risk; for the current use-case it is a low-severity concern.

Important Files Changed

Filename	Overview
docs/app/reflex_docs/docgen_pipeline.py	Converted _render_spans from free function to instance method; added _resolve_link_target using regex + eval to expand {expr} hrefs; updated all call sites. Two P2 concerns: soft eval sandbox and missing error context for NameError/AttributeError.
docs/app/reflex_docs/pages/docs/init.py	Copies .path attribute from top-level page objects onto the shadowed SimpleNamespace objects so markdown can reference {library.path} and {custom_components.path} correctly.
docs/app/tests/test_docgen_link_targets.py	New test file covering plain URL passthrough, brace expr substitution, and anchor composition. No issues found.
docs/getting_started/basics.md	All hardcoded /docs/... URLs replaced with {namespace.path} expressions; corresponding imports added to the python exec block.
docs/enterprise/overview.md	All hardcoded enterprise component links replaced with dynamic path references; enterprise namespace imported in exec block.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A["Markdown source\n[text]({ns.attr.path})"] --> B["_exec_code\nbuilds env namespace"]
    B --> C["ReflexDocTransformer\n(holds self.env)"]
    C --> D["self._render_spans(spans)"]
    D --> E{LinkSpan?}
    E -- yes --> F["_resolve_link_target(target, self.env)"]
    F --> G["_BRACE_RE.sub(_sub, target)"]
    G --> H["eval(expr, {__builtins__:{}}, env)\nresolves ns.attr.path → str"]
    H --> I["doclink2(text, href=resolved)"]
    E -- no --> J["other span types\n(text, bold, code, img...)"]
    I --> K["rx.Component tree"]
    J --> K

Loading

_{Reviews (1): Last reviewed commit: "ENG-9414: Remove hardcoded docs urls" | Re-trigger Greptile}

[adhami3310]

i would prefer if we didn't, it makes less digestable to AI

[carlosabadia]

i would prefer if we didn't, it makes less digestable to AI

that's true indeed

see this @Alek99

[adhami3310]

this should instead be a CI/check step that all internal doc links are valid

[masenf]

one thing we need to solve with this is getting the actual url to be rendered when the user accesses the .md link. right now, it's just showing the placeholder, which makes the link unusable.

there's another issue i think we should address #6436, which is indicating the markdown files served via .md are utf-8. on my mac, both safari and chrome are defaulting to latin-1 or something and it's garbling some characters.