Repairing a Broken PDF in Rust — Rebuilding the XREF Table From Scratch

Source: Dev.to

The problem

Some PDFs won’t open, not because the content is missing, but because the index that tells readers where to find the content is corrupt.
That index is the XREF table, and it can be rebuilt.

What the XREF table looks

xref
0 6
0000000000 65535 f
0000000009 00000 n
0000000058 00000 n
0000000115 00000 n
0000000266 00000 n
0000000496 00000 n

When a reader opens a PDF, it reads this table first. If it’s missing or corrupt, the PDF “won’t open.” The content objects are still in the file; we just need to locate them and rebuild the index.

Rebuilding the XREF table in Rust

pub fn rebuild_xref(data: &[u8]) -> Result {
    // lopdf can attempt recovery on malformed files
    let doc = Document::load_mem(data)
        .or_else(|_| recover_document(data))?;
    Ok(doc)
}

Scanning for objects

pub fn recover_document(data: &[u8]) -> Result {
    // Scan the raw bytes for object markers
    // Pattern: "N 0 obj" where N is the object number
    let mut offsets: Vec = Vec::new();
    let obj_pattern = b" 0 obj";

    for (i, window) in data.windows(obj_pattern.len()).enumerate() {
        if window == obj_pattern {
            // Walk back to find the object number
            if let Some(num) = extract_obj_num(data, i) {
                offsets.push((num, 0, i - num.to_string().len()));
            }
        }
    }

    // Reconstruct document from found objects
    rebuild_from_offsets(data, offsets)
}

Typical scenarios where rebuilding helps

• PDFs truncated mid‑write (e.g., power loss during save)
• PDFs with incremental updates that broke the XREF chain
• Old files where the XREF was hand‑edited incorrectly
• Scanner output with malformed structure

If the content streams themselves are corrupt—the actual page data is gone—no amount of XREF rebuilding helps. Structural resurrection only works when the objects are present but the index is broken.

About 80 % of “won’t open” PDFs I’ve tested are XREF problems. The content is fine; they just need a new index.

Resources

• Hiyoko PDF Vault – https://hiyokoko.gumroad.com/l/HiyokoPDFVault
• Twitter: @hiyoyok