fixed bug with 1-bit encoded genotypes

setup.py

@@ -123,7 +123,7 @@ def build_zstd():
     description='Package for loading data from bgen files',
     long_description=io.open('README.md', encoding='utf-8').read(),
     long_description_content_type='text/markdown',
-    version='1.6.0',
+    version='1.6.1',
     author='Jeremy McRae',
     author_email='[email protected]',
     license="MIT",

src/writer.cpp

@@ -320,7 +320,7 @@ std::uint32_t encode_unphased(std::vector<std::uint8_t> &encoded,
       bit_idx += bit_depth;
     }
   }
-  return genotype_offset + (bit_idx / 8);
+  return genotype_offset + (bit_idx / 8) + (std::uint32_t)((bit_idx % 8) > 0);
 }
 
 std::uint32_t encode_phased(std::vector<std::uint8_t> &encoded,
@@ -383,7 +383,7 @@ std::uint32_t encode_phased(std::vector<std::uint8_t> &encoded,
     i += (max_probs * max_ploidy) - (n_probs * _ploid);
     sample_idx += 1;
   }
-  return genotype_offset + (bit_idx / 8);
+  return genotype_offset + (bit_idx / 8) + (std::uint32_t)((bit_idx % 8) > 0);
 }
 
 std::vector<std::uint8_t> encode_layout2(

tests/test_bgenwriter.py

@@ -398,3 +398,48 @@ def test_multiple_read_writes(self):
                 as_integers = (probs * max_val).round()
                 self.assertTrue(probs_close(geno, probs, bit_depth=bit_depth))
                 self.assertTrue((integer_values == as_integers).all())
+
+    def test_all_possible_genotypes(self):
+        ''' check all possible values in the range available to the bit depths
+        '''
+        max_samples = 10000000
+        for bit_depth in range(1, 24):
+            print(bit_depth)
+            first_path = self.tmpdir / f'temp_{bit_depth}.v1.bgen'
+            second_path = self.tmpdir / f'temp2_{bit_depth}.v2.bgen'
+
+            max_val = (2 ** bit_depth) - 1
+            increment = 1
+            if (max_val / 2) > max_samples:
+                increment = ((max_val + 1) / 2) / max_samples
+
+            integer_values = np.arange(0, max_val + 1, increment)
+
+            remainder = max_val - (integer_values[::2] + integer_values[::-2])
+
+            integer_values = np.array([integer_values[::2], integer_values[::-2],
+                                       remainder], dtype=np.uint32)
+            integer_values = np.ascontiguousarray(integer_values.T)
+            geno = integer_values / max_val
+
+            # write first round
+            with BgenWriter(first_path, n_samples=len(geno)) as bfile:
+                bfile.add_variant('var1', 'rs1', 'chr1', 10, ['A', 'C'], geno,
+                                  bit_depth=bit_depth)
+
+            # check the first write
+            with BgenWriter(second_path, n_samples=len(geno)) as out_bfile:
+                with BgenReader(first_path, delay_parsing=True) as bfile:
+                    probs = next(bfile).probabilities
+                    as_integers = (probs * max_val).round()
+                    self.assertTrue(probs_close(geno, probs, bit_depth=bit_depth))
+                    self.assertTrue((integer_values == as_integers).all())
+                    out_bfile.add_variant(
+                        'var1', 'rs1', 'chr1', 10, ['A', 'C'], probs, bit_depth=bit_depth)
+
+            # check the re-written bgen
+            with BgenReader(second_path, delay_parsing=True) as bfile:
+                    probs = next(bfile).probabilities
+                    as_integers = (probs * max_val).round()
+                    self.assertTrue(probs_close(geno, probs, bit_depth=bit_depth))
+                    self.assertTrue((integer_values == as_integers).all())