fix(parser): recover root-bearing linearized xrefs

[yzxcj797]

Problem

Issue #345's scanned PDF reports page_count: 0 and rejects conversion with OCR is required. The file is linearized and has more than 512 bytes of padding after its final %%EOF.

The document has two classic xref tables:

• an early table whose trailer contains /Root and /Prev;
• the final table, which completes the object list but has no /Root in its own trailer.

lopdf cannot find the final EOF through the padding, so container repair runs. The old repair selected the newest classic xref without inspecting its trailer. That loads the rootless final table, leaves the catalog unreachable, and produces a zero-page result even though the root-bearing table and its /Prev chain are intact.

Change

• Enumerate validated classic xref starts from newest to oldest.
• Prefer the newest table whose own trailer names /Root, which also recovers the rest of the chain through /Prev.
• Keep the newest valid table as a fallback for unusual trailer layouts.
• Add a synthetic, stdlib-only linearized PDF regression with two xref tables and 600 trailing NUL bytes. It fails on main with page_count == 0 and now returns one text page containing Hello World.

The sample from the issue was also checked locally: it now reports four scanned pages (still requiring OCR) instead of zero pages.

Tests

• cargo test --lib — 987 passed.
• cargo test --test integration_tests test_process_pdf_repairs_padded_linearized_xref_chain -- --nocapture — 1 passed.
• cargo clippy --lib -- -D warnings — clean.
• cargo fmt — clean.

Fixes #345

---

Summary by cubic

Recover linearized PDFs with padded EOF by preferring a root-bearing classic xref and validating /Prev direction. Previously repair always chose the newest classic xref and could pick a rootless final table, producing page_count: 0; now it selects the newest xref whose trailer has /Root, or keeps the newest rootless table only when its /Prev points backward, preserving revisions and restoring a reachable catalog.

• Scans valid classic xref tables newest→oldest; bounds trailer searches to before the next xref; skips comments while locating the standalone trailer keyword; parses trailer dictionaries for top‑level keys only (ignores comments, strings, nested keys).
• Preserves amended /Prev chains by retaining the newest rootless table only if /Prev points backward; rejects forward /Prev and falls back to the newest root-bearing table.
• Keeps the newest valid table as a fallback when no root-bearing trailer exists; guards short inputs.
• Adds a stdlib‑only regression PDF with 600 trailing NULs and unit tests for commented “trailer”, nested/commented /Root, and backward/forward /Prev chains; the sample in Could not convert this file: unsupported input: PDF has no extractable text (Scanned, 0 pages): OCR is required #345 now reports four scanned pages instead of zero.

Fixes #345.

^{Written for commit bc1e19e. Summary will update on new commits.}

[cubic-dev-ai]

All reported issues were addressed across 2 files

<sub>Shadow auto-approve: would not auto-approve because issues were found.

Fix all with cubic | Re-trigger cubic</sub>

[cubic-dev-ai]

All reported issues were addressed across 1 file (changes from recent commits).

<sub>Shadow auto-approve: would not auto-approve because issues were found.
Tip: Review your code locally with the cubic CLI to iterate faster.

Fix all with cubic | Re-trigger cubic</sub>

[cubic-dev-ai]

1 issue found across 1 file (changes from recent commits).

Prompt for AI agents (unresolved issues)

Check if these issues are valid — if so, understand the root cause of each and fix them. If appropriate, use sub-agents to investigate and fix each issue separately.


<file name="src/lib.rs">

<violation number="1" location="src/lib.rs:4229">
P3: The new `find_standalone_keyword` reimplements the exact whitespace/`%`-comment skipping that `skip_pdf_whitespace_and_comments` (defined a few lines above, in the same file) already provides, with identical match arms (`is_ascii_whitespace`, then `%` → skip to `\n`/`\r`). Two separate copies of the PDF comment semantics now have to be kept in sync. The standalone keyword scan can reuse the existing helper: since whitespace/comments are skipped before inspecting the next byte, the preceding byte is always whitespace (or `pos == 0`), so the `before_ok` check becomes implied and the loop simplifies.</violation>
</file>

<sub>Shadow auto-approve: would not auto-approve because issues were found.
Tip: Review your code locally with the cubic CLI to iterate faster.

Fix all with cubic | Re-trigger cubic</sub>

[cubic-dev-ai]

P3: The new find_standalone_keyword reimplements the exact whitespace/%-comment skipping that skip_pdf_whitespace_and_comments (defined a few lines above, in the same file) already provides, with identical match arms (is_ascii_whitespace, then % → skip to \n/\r). Two separate copies of the PDF comment semantics now have to be kept in sync. The standalone keyword scan can reuse the existing helper: since whitespace/comments are skipped before inspecting the next byte, the preceding byte is always whitespace (or pos == 0), so the before_ok check becomes implied and the loop simplifies.

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. At src/lib.rs, line 4229:

<comment>The new `find_standalone_keyword` reimplements the exact whitespace/`%`-comment skipping that `skip_pdf_whitespace_and_comments` (defined a few lines above, in the same file) already provides, with identical match arms (`is_ascii_whitespace`, then `%` → skip to `\n`/`\r`). Two separate copies of the PDF comment semantics now have to be kept in sync. The standalone keyword scan can reuse the existing helper: since whitespace/comments are skipped before inspecting the next byte, the preceding byte is always whitespace (or `pos == 0`), so the `before_ok` check becomes implied and the loop simplifies.</comment>

<file context>
@@ -4222,16 +4222,29 @@ fn find_standalone_keyword(buf: &[u8], start: usize, end: usize, keyword: &[u8])
+    while pos < end {
+        match buf[pos] {
+            byte if byte.is_ascii_whitespace() => pos += 1,
+            b'%' => {
+                while pos < end && !matches!(buf[pos], b'\n' | b'\r') {
+                    pos += 1;
</file context>