epiread.c

[vpbrendel]

line 706
// reference is a G
if (bsstrand && rb == 'G' && rpos+j-1 >= rs->beg) {

I think you need also "&& j>0" to prevent a core dump for some input.

Volker

[jamorrison]

I looked through the code block associated with that statement, and I think the code is okay as written. rpos is 1-based, so it's smallest value would be 1, and the rpos+j-1 >= rs->beg check will make sure there isn't a core dump from fetching a base outside of the cached reference window.

However, if you find a case where it actually does produce a core dump, feel free to reopen this issue with the test case and I'll look into it more then.

[jamorrison]

I looked into this further and found there is a specific case where the core dump occurs. When the read is from the OB/CTOB strand and the first base is a G in a CG, then as the code was written, it would produce a core dump. The implemented fix isn't ideal (see below for a copy of the in-code comment), but none of the fixes I thought through were ideal.

/* This is a tough case that can only happen when calling '-5 0' in
 * the invocation. But it's also not a rare edge case.
 *
 * When the first base of a OB/CTOB read is a G in a CG, then you
 * can get the methylation status from that base. However, all CGs
 * have the methylation status placed on the C for consistency
 * across strands and ease for downstream analysis. Because the C
 * doesn't occur in the read itself, this can cause issues for how
 * to handle this methylation status. You can't put it on the C,
 * because it's not there, and you can't put it on the G, since that
 * doesn't fit how all the other reads are handled.
 *
 * For now, the least worst option (or the one I'm going with) is to
 * filter this base when it occurs. The occurrences will be counted
 * and printed out at the end so that the user knows it occurs. If
 * running in verbose mode, then a message will be printed to the
 * screen as well.
 *
 * This is an issue for both NOMe- and BS-seq.
 */