[satyaorz]

Disclosing up front: this was put together with help from Claude (Anthropic's coding agent). I reproduced and verified everything below before opening the PR.

writeMSA writes SELEX, Stockholm, CLUSTAL and PIR files that parseMSA cannot read back, or reads back with different labels or sequences. On Python 3 the CLUSTAL and PIR writers do not work at all. Before → after, writing a small MSA and parsing it back:

case	master	this PR
.aln	TypeError: 'float' object cannot be interpreted as an integer (range(n/60))	round-trips
.ali	same TypeError	round-trips
.slx with a label longer than 31 characters	label and sequence are spliced together	round-trips
.sth with a 500-character label	same, or a segfault in parseSelex for some label sets	round-trips
.slx with 12000 columns	IOError: failed to parse SELEX/Stockholm file at line 2 (line longer than the fixed buffer)	round-trips

SELEX / Stockholm (msaio.c)

• parseSelex. Previously it took the column where the first sequence starts and read every line at that column. That broke for labels of different lengths, labels longer than the padding, and lines longer than SELEXLINELEN. Now each line is split into a label (everything before the last whitespace-separated item) and a sequence. Lines are read whole, growing the buffer as needed. Alignments split into blocks separated by blank lines are joined, and a block whose labels or row count differ from the first block's raises IOError. Markup lines (#) are skipped, and nothing after // is read.
• writeSelex. It now always leaves at least one space between a label and its sequence. It also no longer writes labels longer than the padding into a fixed-size buffer, and the label's bytes are released only after they are written (they were Py_DECREFed before use).
• Labels. Labels are decoded as UTF-8 with replace, so the C path and MSAFile agree on non-ASCII labels.

Python readers and writers (msafile.py)

• MSAFile. It now reads multi-block SELEX/Stockholm the same way as the C parser, and also reads CLUSTAL and PIR. It can write CLUSTAL (written on close(), since CLUSTAL interleaves sequences) and PIR. A Stockholm writer on a stream now writes the closing // too.
• parseClustal.
  • Handles blocks, conservation lines and the residue counts clustalw 2 writes.
  • Accepts the headers Biopython accepts (MUSCLE, PROBCONS, MSAPROBS, Kalign).
  • Allows labels that contain spaces.
• writeClustal. Pads labels to the longest one instead of a fixed 16 characters, and writes a CLUSTAL W header that Biopython reads.
• parsePIR. Takes the line after each >P1; line as the description, whatever its first field, and no longer keeps the terminal * as a residue.
• writePIR. Writes the full label by default, so it can be read back.
• splitSeqLabel. In a label containing /, only a trailing /start-end is split off, so labels like sp/inner/4-11 or x/1/2 keep their full identifier.

Tests

• New file prody/tests/sequence/test_msa_formats.py. It has round trips for all four formats through writeMSA/parseMSA and through MSAFile, on plain and .gz files and on streams. It also reads fixtures written the way HMMER/Pfam, clustalw 1/2 and MODELLER write them, and checks that mismatched blocks raise IOError. On master these tests fail, and the run segfaults in parseSelex partway through.
• Fixtures are added as prody/tests/datafiles/msa_formats_*, which the existing datafiles/msa* package-data pattern already covers.
• Results.
  • prody/tests/sequence: 254 passed. The one failure is testBuildMSAclustalw, which needs the clustalw binary I don't have installed locally.
  • Full suite: no other failures beyond missing optional dependencies (openmm, pdbfixer, mmtf, requests, clustalw).
  • Valgrind is clean on the C parser over a set of malformed inputs (empty files, CRLF, NUL bytes, very long lines and labels, ragged blocks).
• Interop. Biopython's AlignIO reads the Stockholm, CLUSTAL and PIR files written here, and ProDy reads Biopython's.