Invalid memory use and segfaults

[otsaw]

After porting some code to dplyr's left_join, group_by+summarize and rbind_*, we keep getting segfaults often at rather random places. Assuming these segfaults were preceded by some kind of memory corruption, I ran R with valgrind (R -d valgrind &> foo.log < foo.R) and got the following output.

https://gist.github.com/otsaw/19be4bcff7690e8e75c8
https://gist.github.com/otsaw/8c368d552a1f378bae75

I have also managed to catch one relevant looking segfault with a backtrace.

https://gist.github.com/otsaw/f15a68e841e78187e79f

These problems occur in a large pile of stochastic simulation code. I haven't been able to isolate the problem. I can reproduce it, but not consistently. If I run the code with one core and valgrind, it might take days to reproduce. As far as I can tell, there shouldn't be anything weird about the data frames fed to dplyr functions, just normal data frames with more than zero rows and columns and basic data types.

I'm not familiar with debugging compiled code or memory errors. Does that output help at all? Is there something besides valgrind that I should try? Should dplyr's valgrind output be clean or can these be false positives? Based on the output, there appears to be some problems converting back to data frame from tbl_df; as_regular_df and ungroup_grouped_df?

[romainfrancois]

It is going to be really hard to help if you can't give us some reproducible. The valgrind output is not really useful in isolation.

[otsaw]

Yes, I figured so. Feel free to close this, I got no clues remaining, I can't isolate the bug.