segmentation fault

[ptitle]

For one of my input files, with release 4.0-beta7, I get:

Allocation failed in sample_tri_multi!  Proceeding.
Allocation failed in sample_tri_multi!  Proceeding.
Allocation failed in sample_tri_multi!  Proceeding.
[maybe 30 times]
[1]    25144 segmentation fault  bali-phy APOB_extracted_oneseq_onedirection_All.fasta -S gtr -n  -i 2000

I then tried building from source with the current master branch to see if this was something that had gotten resolved, and that gave me:

bali-phy: Error! evaluating reg # 8295 (unchangeable): case <175495> of {_ -> <175496>}

evaluating reg # 175495 (unchangeable): case <175496> of {_ -> <2404>}

evaluating reg # 175496 (unchangeable): MCMC:runMCMC 2000 0 0

evaluating reg # 471014 (unchangeable): case <86522> of {_ -> <86561>}

evaluating reg # 86522 (unchangeable): case <86561> of {_ -> <2404>}

evaluating reg # 86561 (unchangeable): MCMC:walk_tree_sample_alignments <6084> <46054> 0 0

std::bad_array_new_length

I can provide the input fasta file if that would be helpful. Any thoughts?

Thanks!

[bredelings]

Yeah, can you provide the input file?

[ptitle]

You can download it here.

It's a somewhat slow analysis, probably due to the couple of longer sequences relative to the rest.

[bredelings]

Initial thoughts:

• some of the sequences are really long. bali-phy can't handle really long sequences yet.
• some of the sequences are missing the conserved domain.
• you probably need to do some initial data curation before giving the sequences to bali-phy
• you need to chop off the non-homologous ends before giving the sequences to bali-phy.
• The Allocation failed messages are a result of the sequences that are too long.
• The SEGFAULT is bug.

APOB.fsa.fasta.gz

I have attached an alignment of your sequences with fsa. It is a relatively high-quality aligner that is good for data cleaning because it doesn't align non-homologous sequences. I think if you take a look at the fsa alignment in aliview, you will see what the issues are.

More coming.

Thoughts?

[ptitle]

Thanks! I'll have some time next week to look into this -- I'll report back.

[bredelings]

OK, I found and fixed the SEGFAULT. The original data set now runs under the github master version and seems to be giving a reasonable alignment. Obviously its going to run way faster if you remove the long flanking sequences.

[ptitle]

Thanks! I'll give it a try next week and confirm. Definitely seems to make sense to trim out those excessively long sequences.

[ptitle]

After doing some pre-filtering to remove long flanking sequences, and with the current main branch, bali-phy now works without throwing any errors. Thanks!