Hi. Based on the last email from Andreas, it seems that upgrading to 2.6.0 will not be as simple as I thought.
This is what I tried without success: I tried dropping the current debian/patches/numpy2.patch and using the attached patch instead, which is based on upstream's commit regarding NumPy. Those two patches overlap to some extent, but I'm not 100% sure that the second one should work as a drop-in replacement of the first. Thanks.
From: Nathaniel Echols <[email protected]> Subject: [SL-11761] fix numpy 2.x compatibility Bug-Debian: https://bugs.debian.org/1114717 --- python-pbcore-2.1.2+dfsg.orig/pbcore/io/align/BamAlignment.py +++ python-pbcore-2.1.2+dfsg/pbcore/io/align/BamAlignment.py @@ -532,7 +532,10 @@ class BamAlignment(AlignmentRecordMixin) if data.dtype == np.int8: gapCode = ord("-") else: - gapCode = data.dtype.type(-1) + try: + gapCode = data.dtype.type(-1) + except OverflowError: + gapCode = data.dtype.type(np.iinfo(data.dtype).max) uc = self.unrolledCigar(orientation=orientation) alnData = np.repeat(np.array(gapCode, dtype=data.dtype), len(uc)) gapMask = (uc == gapOp) --- python-pbcore-2.1.2+dfsg.orig/pbcore/io/align/_BamSupport.py +++ python-pbcore-2.1.2+dfsg/pbcore/io/align/_BamSupport.py @@ -1,5 +1,7 @@ # Author: David Alexander +import re + import numpy as np @@ -63,7 +65,13 @@ BAM_CDIFF = 8 # qId calculation from RG ID string # def rgAsInt(rgIdString): - return np.int32(int(rgIdString.split("/")[0], 16)) + numericId = int(rgIdString.split("/")[0].split("-")[0], 16) + # Identifiers may exceed the 32-bit range, so compensate manually + # the overflow, otherwise numpy 2 and later raises OverflowError. + overflownId = (numericId + np.iinfo(np.int32).min) \ + % (2 * (np.iinfo(np.int32).max + 1)) \ + + np.iinfo(np.int32).min + return np.int32(overflownId) # # Kinetics: decode the scheme we are using to encode approximate frame --- python-pbcore-2.1.2+dfsg.orig/pbcore/io/dataset/DataSetMembers.py +++ python-pbcore-2.1.2+dfsg/pbcore/io/dataset/DataSetMembers.py @@ -827,7 +827,7 @@ class Filters(RecordWrapper): 'mapqv': (lambda x: x.mapQV), 'accuracy': ( lambda x: (np.ones(len(x.nMM), dtype='f4') - - (x.nMM + x.nIns + x.nDel).astype(float) / + (x.nMM + x.nIns + x.nDel).astype(float32) / (x.nM + x.nMM + x.nIns))) } base = self._pbiVecAccMap()
