Problem creating new BAM records (SAM=Segfault,BAM=silent failure)

[ebioman]

Hi
I am currently trying to create uBAM (unmapped BAM) files and running into problems creating non-mapped entries from
any kind of given sequence (e.g. FASTA, FASTQ,...).
Essentially I can create a new BAM default record and BAM::Writer will write it but then I get either:

1. for SAM output a segfault
2. for BAM output no error at all, but the generated BAM is truncated

To understand better, lets take a BAM input and generate a BAM output file

use rust_htslib::{bam, bam::Read};

fn ubam2ubam (input: &str, output: &str) {

  let mut reader = bam::Reader::from_path(input).expect("ERROR: could not open input file");
  let header       = bam::Header::from_template(reader.header());
  let mut writer  = bam::Writer::from_path(output, &header, bam::Format::Sam ).expect("ERROR: could not open output file");
  
  for r in reader.records() {
          let record = r.expect("ERROR: could not read BAM entry!");
          // now generating an empty record
          let mut out_record = bam::Record::default();
          // populating with some features from input BAM
          out_record.set(
              record.qname(), //qname
              None, //Cigar
              &record.seq().as_bytes(), //sequence 
              record.qual() //qual
          );
          writer.write(&out_record).expect("ERROR: could not write out record!");
      }
}

This compiles but fails then with a Segmentation fault (core dumped) for Sam.
If I write the original record instead of the new empty one everything works fine.
I guess there is some combination of feature or rather missing feature which breaks it but I cant find which one.

[jch-13]

Hello! Can you please try to explicitly unset contig and position right before writing the Record to the output file?

out_record.set_tid(-1);
out_record.set_pos(-1);
out_record.set_mpos(-1);
out_record.set_mtid(-1);
// Perhaps also
out_record.set_unmapped();

Does the error still occur?

[ebioman]

@jch-13 thanks, I tried that already before - but unfortunately does not fix the situation.
A few things I tried unsuccessfully so far:

• out_record.set_tid(-1);
• out_record.set_flags(4);
• out_record.set_pos(-1);
• let mapq = "*".as_bytes();
  out_record.set_mapq(mapq[0]);
• out_record.set_mpos(-1);
• out_record.set_insert_size(0);

My suspicion is currently either the CIGAR or the header somehow.
For Cigar I tried the following as there is no record.set_Cigar function:

let cigar_str = CigarString::try_from("".as_bytes()).expect("ERROR: could not create empty CIGAR string!");
out_record.set(
              record.qname(), //qname
              Some(&cigar_str), //Cigar
              &record.seq().as_bytes(), //sequence 
              record.qual() //qual
          );

But it does not fix the issue neither. For the header I am wondering is it mandatory to set maybe set_header for each entry?

[jch-13]

I can't reproduce it here (the code is working just fine when I add the lines above), so the culprit might indeed be the header of the input BAM file.

[ebioman]

Thanks, the following fixed it:

out_record.set_mtid(-1);

Which I did not try and which you suggested later in the edit. This was not on my radar at all as I am working with single reads and not paired ones.

I would though currently hesitate to close the ticket as I think that this would maybe need a fix in the bam::Record::default(); or some kind of other structure to assure there is no silent failure in Bam case and that the Sam failure is easier to understand ?

[jch-13]

Ah yes, sorry, I should have made the edit more clear. And I think you are right; this is more of a work-around than a proper fix.

I wonder if MaybeUninit::zeroed().assume_init() in here:

rust-htslib/src/bam/record.rs

Line 117 in a21aff2

inner: unsafe { MaybeUninit::zeroed().assume_init() },

should be replaced by *htslib::bam_init1().

Edit: No, it turns out that this does not help. This seems to be htslib's default behaviour, so I'm not sure whether we should default to creating unmapped records in rust-htslib or just document it carefully?

[ebioman]

Personally I would advocate that the default of generating a new record should be a valid un-mapped single read first, with no pair defined. That would be safe, meaning one could generate a default new header with a default new record and it would generate a valid BAM file.
If we'd go rather for paired reads as default (maybe people prefer that) at least it should then generate a valid default record for the 2nd read by default, too.
The way it is currently, that a new default record ends in an error is obviously not favorable as it is not easily understandable for the user where the error stems from.

[jch-13]

Looking at the headers over at https://github.com/samtools/htslib/blob/9045785cef02e7cd97b0b4caecbdcaf6217a986a/htslib/sam.h#L1007-L1036, it seems htslib's bam_set1() method sets all necessary fields, whereas rust-htslib's bam::record::set() only sets the variable length data. So in rust-htslib, we don't actually have a method to create a valid BAM record where all necessary fields are populated.
I believe it would make sense if bam::record::new() took all necessary fields as arguments. The Default impl could take fewer arguments, if any (by calling new() with reasonable defaults). What do you think?

[ebioman]

Hello @jch-13 , sorry for the late reply.
I think this makes definitely sense in my opinion