* Added support for obtaining input reads directly from the Sequence Read Archive, via NCBI's [NGS language bindings](https://github.com/ncbi/ngs). This is activated via the [`--sra-acc`](manual.shtml#bowtie2-options-sra-acc) option. This implementation is based on Daehwan Kim's in [HISAT2](https://ccb.jhu.edu/software/hisat2). Supports both unpaired and paired-end inputs.
* Bowtie 2 now compiles on ARM architectures (via [#216](https://github.com/BenLangmead/bowtie2/pull/216))
* `--interleaved` can now be combined with FASTA inputs (worked only with FASTQ before)
* Fixed issue whereby large indexes were not successfully found in the `$BOWTIE2_INDEXES` directory
* Fixed input from FIFOs (e.g. via process substitution) to distinguish gzip-compressed versus uncompressed input
* Fixed issue whereby arguments containing `bz2` `lz4` were misinterpretted as files
* Fixed several compiler warnings
* Fixed issue whereby both ends of a paired-end read could have negative TLEN if they exactly coincided
* Fixed issue whereby `bowtie2-build` would hang on end-of-file (via [#228](https://github.com/BenLangmead/bowtie2/pull/228))
* Fixed issue whereby wrapper script would sometimes create zombie processes (via [#51](https://github.com/BenLangmead/bowtie2/pull/51))
* Fixed issue whereby `bowtie2-build` and `bowtie2-inspect` wrappers would fail on some versions of Python/PyPy
* Replaced old, unhelpful `README.md` in the project with a version that includes badges, links and some highlights from the manual
* Note: BAM input support and CMake build support both remain experimental, but we expect to finalize them in the next release
