No Side Effects on Source Files

All content transformations SHALL operate on in-memory buffers. Source .adoc files SHALL never be modified by the extension. Processing the same file twice SHALL produce the same result.

Source: openspec/changes/archive/2026-07-28-add-pdf-output/specs/pdf-output/spec.md

Context

The extension processes .adoc files during the Antora build — parsing items, expanding macros, and substituting inline relationship links. The user’s source files are under version control and must never be modified by the build process.

Trigger

The Antora contentClassified event fires, triggering the extension’s content processing pipeline.

Acceptance Criteria

In-memory only: All content transformations (macro expansion, link substitution, title injection) occur on the in-memory content buffer — no .adoc file on disk is written to; source: strace / filesystem audit during build; horizon: every build.
Idempotent re-processing: Processing the same file twice produces the same in-memory result and does not modify the graph state (warns on duplicate IDs but does not crash); source: process() called twice on same content; horizon: every build.
Read-only file safety: The build succeeds even if .adoc files are read-only on the filesystem; source: integration test with chmod -w; horizon: every release.