Too Many Items error when reading large CSV

Summary:

When reading in posterior samples from a CSV with more than INT_MAX total elements (2,147,483,647), the CSV cannot be loaded.

Description:

Due to Rstan's reliance on the base R's scan function to load the CSV. It is limited by the R vector size.

This was also documented here:
https://groups.google.com/forum/#!topic/stan-users/XvroDYe_yJc

Even though R supports long vectors, they are not yet implemented in the scan function. I created a patch to the scan function here, which fixes the issue.

https://gist.github.com/aaronjg/f39e5966687ca004dab5a10e7655c648

There may also be a way to fix this without patching base R using read.table rather than scan, since that doesn't require loading everything into a single vector.

Reproducible Steps:

Generate a stan output CSV with more than INT_MAX total values, and try to load into stan using the read_stan_csv function.

Current Output:

Error in scan(file, what, nmax, sep, dec, quote, skip, nlines, na.strings, :
too many items
Calls: stan -> sampling -> sampling -> .local -> scan

Expected Output:

RStan should read this correctly

RStan Version:

2.17.2

R Version:

3.5.0

Operating System:

Ubuntu, 64 bit.

Start at read_stan_csv and trace the sampling -> .local -> scan call shown in the error. Compare the existing scan-based loading with the suggested read.table approach or the linked base R scan patch, then reproduce with a CSV exceeding INT_MAX total elements. Done means RStan can load that CSV without the “too many items” error.