Directory tree:

    Makefile        Builds and runs tests.
    include/        Public API.
    src/            Scripts, C implementation and internal headers.
        build/      Generated object files, executables etc.
    test/           Test files.
        generated/  Files generated by tests.

Suggested setup for testing:
    Checkout ghostpdl and mupdf into the same directory.
    Inside ghostpdl:
        ln -s ../mupdf/thirdparty/extract extract

    Then either:
        Inside ghostpdl:
            ./autogen.sh --with-extract-dir=extract
            make -j 8 debug DEBUGDIRPREFIX=debug-extract-
        Inside mupdf:
            make -j 8 debug
    or:
        make test-rebuild-dependent-binaries (for the first time)
	make test-build-dependent-binaries (for incremental builds)


    Then build and run tests from inside mupdf/thirdparty/extract
    as below.

Build and run tests with:
    make

Conventions:

    Errors:
    
        Functions return zero on success or -1 with errno set.

    Identifier/symbol names:

        All identifiers that can be seen by client code (generally things
        defined in include/) start with 'extract_'.
        
        Similarly global symbols in generated .o files all start with
        'extract_'; this is tested by target 'test-obj'.

        Other identifiers and symbols do not have an 'extract_' prefix - not
        necessary because client code cannot see these names.
        
        Header names in include/ start with 'extract_'.

    Allocation:

        Functions that free a data structure generally take a double pointer
        so that they can set the pointer to NULL before returning, which helps
        avoid stray invalid non-NULL pointers. E.g.:

            extract_span_free(extract_alloc_t* alloc, span_t** pspan);
            /* Frees a span_t, returning with *pspan set to NULL. */

        This double-pointer approach is also used for raw allocation - see
        include/extract_alloc.h.

    Lists:
        Lists of data items are generally implemented using an array of
        pointers and an int 'foo_num' entry, e.g.:

            line_t**    lines;
            int         lines_num;