Fread: possible segfault when reading a file with extra sep on out-of-sample row

[st-pasha]

Test file can be generated using the following python script:

nrows = 2100       # need >2000 in order to trigger nJumps==10
ncols = 1024 * 16  # "round" number increases probability that 
                   # `sizes` array will not be overallocated
row = ','.join(list('abcdefghijklmnop') * int(ncols / 16))
rows = [row] * nrows
rows[111] += ","
src = "\n".join(rows)
open("test.txt", "w").write(src)

Then

> for(b in 1:100){ fread("test.txt", header=F) -> DT }

 *** caught bus error ***
address 0x7fd839704e40, cause 'non-existent physical address'

The primary cause are the following lines in "TEPID" section of freadMain:

         else if (eol(&tch)) {
            int8_t thisSize = size[j];
            ((char **) targets)[thisSize] += thisSize;

Here j can be equal to ncol, which causes OOB access in array size, and even if that succeeds by accident, the retrieved value thisSize can be anything, causing OOB access on array targets.