Why don't mapping programs map directly into BAM format?

[oiiio]

Is this historical contingency? Or is there a functional reason why mapping programs like BWA or Stampy don't map directly to BAM?

[lh3]

1) because writing BAM is non-trivial and no one wants to add a dependency only for writing BAM

2) because you can easily generate BAM by piping the output to samtools

[adaptivegenome]

But assuming it was trivial, is there a good reason to output to SAM. Seems like all downstream operations would be BAM-based anyways?

[simonandrews]

When mapping sequences each sequence is handled independently and the mapping program just wants to write this to the output file and move on (even more so when running multiple threads).

Many people who want to use BAM files want to have indexed files which allow for efficient random access. Because the files need to be sorted the output needs to be post-processed after all of the mapping is done, you can't just write them out as you go along. Mapping programs would therefore still have to write out SAM files as they went along and then convert these to BAM at the end. Since you can easily do this with a samtools command the mapping programs generally don't bother to do this and leave it up to the user.

[maubp]

Indeed. Or to put it another way,

Output of unsorted SAM is easy.

Output of pre-sorted SAM/BAM is hard.

Since you (or the tool) will have to do a SAM/BAM coordinate sorting step anyway, you might as well do the unsorted SAM to sorted BAM in one go with samtools.