Accept binary bytes on the PDF header line

[rth]

I have a pdf vittel_rdc.pdf that fails to load in lopdf with

Error: Parse(invalid file header)

but otherwise opens fine in other pdf viewers.

After some investigations it looks like it has a header with something like

%PDF-1.3 \xb0\x9f\x92\x9c\x9f\xd4\xe0\xce\xd0\xd0\xd0\r

and was produced by ImageMill Imaging Library v3.000.

If I understand correctly, the pdf spec (ISO 32000, §7.5.2) would requires a newline before the binary comment line and that's what lopdf implements. However looking at what other pdf libraries implement they are a bit less strict,

• pypdf: Only validates the 5-byte %PDF- prefix, doesn't parse the rest of the line
• pdf.js (Mozilla): Reads version chars until whitespace (byte ≤ 0x20) or 7 chars max, ignores the rest
• qpdf: Uses regex [0-9]+\.[0-9]+ to extract version digits, ignores trailing bytes

so I would propose to be a bit less strict about this header parsing, to not error on such files. There should be little downside.

Made a PR in #481

[J-F-Liu]

pypdf can set a parameter: strict=False/True, lopdf should also support this.

[rth]

Thanks for the feedback! I'll add it to the PR.

Do you want the default to be strict=true or false? In pypdf it looks like they switched the default from true to false eventually.

[rth]

And also API wise @J-F-Liu if I try to add a strict flag in #481. Would you prefer,

  let doc = Document::load_with_options("file.pdf", &LoadOptions::lenient())?;
  let doc = Document::load_mem_with_options(&bytes, &LoadOptions::lenient())?;

or

Document::load_lenient("f.pdf")?;
// load remains the strict version

or some other pattern?

Do you want the default to be strict=true or false?

I imagine strict by default for backward compatibility?

[J-F-Liu]

I suggest keep Document::load, add Document::load_with_options to replace load_with_password and load_filtered.
Default to strict=false is ok.