Fix pdf_combine() "too many open files" error

[trevorld]

• Use ClosedFileInputSource when combining PDFs so files are opened/closed on demand rather than held open throughout the write phase.
• Also fixed an inner loop variable shadowing (i → j)

closes #21

[jeroen]

Is there any reason this needs a new function, instead of changing the existing read_pdf_with_password?

[trevorld]

Is there any reason this needs a new function, instead of changing the existing read_pdf_with_password?

Do you mean:

1. Should we always use ClosedFileInputSource?

or

2. Should we add an extra argument to read_pdf_with_password() to switch between the two types of objects?

w.r.t. 1) in #21 (comment) it sounded like the ClosedFileInputSource get closed more aggressively and we might need to exert effort to keep them open in other functions

w.r.t. 2) I suppose we could do this if you'd like. The big difference between the two is one uses

pdf->processFile(infile, value.get_cstring());

and the other uses

auto cis = std::make_shared<ClosedFileInputSource>(infile);
pdf->processInputSource(cis, value.get_cstring());

so we could use an ifelse statement a couple of times in the code.

[trevorld]

Is there any reason this needs a new function, instead of changing the existing read_pdf_with_password?

• I've added a use_cfis flag to read_pdf_with_password(). If true we use ClosedFileInputSource
• I've also put a skip() before the reproducible example since on my machine it takes a few minutes and creates a 455Mb temporary file. I've added a simple 2 file case before the skip which is much faster.

[trevorld]

Although I still skip() it I changed the reproducible "too many files" example in the new tests to instead combine 10,000 blank pdf files produced by

grDevices::pdf(blank_file)
grid::grid.newpage()
invisible(grDevices::dev.off())

Although on my machine it still takes 40 seconds and produces a 30Mb temporary file this is an order of magnitude better than combining pdf-example-password.original.pdf 10,000 times.