append error (bug?)

[petersudmant]

I am running into an issue with append that results in either a core dump or a faulty getctg.

I am using this agc provided by @lh3 https://github.com/lh3/OpenHGL. To this agc (human579.agc) I am attempting to append two genomes (2 haplotypes of an assembly), HG02554.1.chr.fasta and HG02554.2.chr.fasta. If I try to append them both (as below) I get a core dump:

#this core dumps
~/local_modules_sw/agc-3.2_x64_linux/agc append human579.agc input/HG02554.1.chr.fasta input/HG02554.2.chr.fasta >./test/human579_plusHG02554.agc
3198569 Segmentation fault      (core dumped)

I CAN append EITHER single genome without a core dump - however, if I then try to getctg I get strange behavior. For example - this fetch returns nothing:

~/local_modules_sw/agc-3.2_x64_linux/agc getctg ./test/human579_plusHG02554.1.agc NC_060928.1@HG02554.1.chr:145315365-146249433
>NC_060928.1 HG02554.1:145315365-146249433

I have created a reproducible example of this where I extracted the first 10Mb of HG02554.1.chr.fasta and try to append it to the agc. You can see that only the first 3350706 bases are able to be extracted from the agc.

wget https://raw.githubusercontent.com/petersudmant/agc_bugfix/refs/heads/main/HG02554.1_10M.chr.fasta
wc HG02554.1_10M.chr.fasta
#add to agc
~/local_modules_sw/agc-3.2_x64_linux/agc append human579.agc HG02554.1_10M.chr.fasta >./test/human579_plus_10M.agc
#fetch it
~/local_modules_sw/agc-3.2_x64_linux/agc getctg ./test/human579_plus_10M.agc NC_060928.1:1-10000000@HG02554.1_10M.chr:1-10000000 | wc

Thanks!

[lh3]

I have seen a segfault when appending ~100 APR genomes to a previous build human472.agc. In my case, I didn't run getctg. The segfault happened during append. I later decided to replace a couple of genomes in human472, so I rebuilt from scratch and didn't investigate what was causing the segfault.

[sebastiandeorowicz]

Sorry for the bug and thank you for the report. We are investigating the problem.

[lh3]

No worries. AGC has been working well otherwise. It has completely changed how we work with large human genome collections.

[petersudmant]

Completely agree! It's fantastic! Let me know if I can assist at all w/ tracking down anything.

[agnieszkadanek]

Thank you for reporting this issue — the bug with append and getctg used after append has been fixed in AGC 3.2.2, which is now also available on Bioconda.

A few more notes regarding obtaining a specified contig range (":from-to") with getctg:

• The output contig header will have the ":from-to" suffix added.
• The default line length is 80 characters. This can be changed with the -l switch.
• Indexing is 0-based, meaning the first base has index 0.

For your example, this will result in almost the same file as input (with only the header suffix added):
agc getctg -l 60 ./test/human579_plus_10M.agc NC_060928.1:1-10000000@HG02554.1_10M.chr:0-9999999 | wc