I am tuning a data import script, and occasionally I find an approach puts too much into memory in one call (usually this is because I am writing inefficient code). The "failed to allocate" message is only sort of useful in that it tells you how much memory was needed (without an informative traceback) and only if the allocation fails. Regular profiling requires that enough memory be available for allocation (and contiguously placed), which changes depending on the circumstances under which the code is run, and is very slow.
Is there a function that simulates a call to see how much memory will be used, or otherwise efficiently profiles how much memory a line of R will need whether it succeeds or fails? Something that could wrap an existing line of code in a script like System.time
would be ideal.
Edit: lsos()
does not work for this because it only describes what is stored after a command is run. (see: Reserved memory of R is twice the size of an allocated array)