Developing
The document converter runs parsers for Memory in a separate process with bounded input and output. Use this page to inspect the process boundary and the parser dependencies.
How It Runs
Memory runs lnk-document-converter text show <pdf|docx> for each document, the
document on its input and the text, or why there's none, on its output.
It runs as a child of memory that holds none of memory's secrets and no
environment, requesting operating-system memory and CPU limits; one that crashes
or takes over 30 seconds fails that document alone. Memory finds it
beside its own program, and its sandbox reads it where it's installed.
Resource Limits
Memory clears the child’s environment, bounds its answer, and kills it
after a 30-second wall-clock deadline. It also requests a 1 GiB address
space limit and CPU limits from the operating system. A platform that
does not enforce a requested limit, including the memory limit on macOS,
still has the separate deadline. The deadline and the memory limit are
memory's, extract-deadline-secs and extract-memory-max-bytes in its
src/limits.toml.
Its Parsers
PDFs are read with lopdf, and Word documents with Link's own reader of
their text part, the zip unpacked with zip. An image-only PDF has no text layer to
extract; the converter performs no optical character recognition.