Remove writing locally

[MKLau]

Writing to disk as a dump of provenance out of memory is not a long-term solution for provenance storage, as memory limits and the handling of provenance by storage or analysis applications. Also, local writing may block CRAN acceptance.

[blernermhc]

R provides a function called tempdir() that returns the path to the per-session temporary directory. According to CRAN rules, this is an ok place to put files. If we keep snapshots, we could put them there and only move them to a more permanent place if the user did some sort of save command, at the same time that the json itself becomes more permanent.

[blernermhc]

There are a couple of other places where RDT writes to files:

• It makes a copy of the script itself, input data files, output files and plots. This ensures that they are consistent with the provenance collected. We will lose assurance of this consistency if they are not saved when the provenance is captured.
• We also save plots that are displayed on the screen but not written to a file.

[MKLau]

Thanks Barb, I agree that the writing to the users file system is not necessarily forbidden by CRAN policy. The tempdir function may be a part of a future solution, but I am still not convinced that this mode of provenance storage via writing to disk as it is currently implemented is the longer-term solution. Verifying the provenance via the saving of the script and the graphics also seems like a short term solution.

I still think taking this feature out and designing the long term feature is the way to go.

[MKLau]

Oh, I just read through this thread (see link below) on the tmpdir function, and it looks like this is not a good way to go anyway. The tmp directory is automatically purged at intervals that depend on the OS.

https://serverfault.com/questions/377348/when-does-tmp-get-cleared

This is a reminder to me that developing in R is often not intuitive and seemingly easy paths are often trickier than the first pass suggests!