bswap16 short header words in field_forall

bswap32 cannot be indiscriminatly applied to the int32 temporary, as it
essentially is a byte-reorder and left-shift when applies to an int16.

The symptom is that certain header fields extracted this way show up in
the absurd range (some 65 million), when normal values are in the few
thousands, such as dt/sample-interval.

Addresses #368

#368

lib/src/segy.c

@@ -687,6 +687,14 @@ static int slicelength( int start, int stop, int step ) {
     return (stop - start - 1) / step + 1;
 }
 
+static int32_t bswap_header_word(int32_t f, int word_size) {
+    if (word_size == 4)
+        return bswap32(f);
+
+    int16_t shortf = f;
+    return bswap16(shortf);
+}
+
 int segy_field_forall( segy_file* fp,
                        int field,
                        int start,
@@ -702,6 +710,10 @@ int segy_field_forall( segy_file* fp,
     err = segy_get_field( header, field, &f );
     if( err != SEGY_OK ) return SEGY_INVALID_ARGS;
 
+    // since segy_get_field didn't fail earlier, it is safe to look up the word
+    // size
+    const int word_size = field_size[field];
+
     int slicelen = slicelength( start, stop, step );
 
     // check *once* that we don't look past the end-of-file
@@ -719,7 +731,7 @@ int segy_field_forall( segy_file* fp,
         for( int i = start; slicelen > 0; i += step, ++buf, --slicelen ) {
             segy_seek( fp, i, trace0, trace_bsize );
             get_field( fp->cur, field_size, field, &f );
-            if( lsb ) f = bswap32(f);
+            if (lsb) f = bswap_header_word(f, word_size);
             *buf = f;
         }
 
@@ -744,7 +756,7 @@ int segy_field_forall( segy_file* fp,
         if( readc != 1 ) return SEGY_FREAD_ERROR;
 
         segy_get_field( header, field, &f );
-        if( lsb ) f = bswap32(f);
+        if (lsb) f = bswap_header_word(f, word_size);
         *buf = f;
     }
 

python/test/segy.py

@@ -1643,6 +1643,17 @@ def test_readtrace_int16():
         assert list(tr[40:19:-5]) == [-888, -2213, 5198, -1170, 0]
         assert list(tr[53:50:-1]) == [-2609, -2625, 681]
 
+def test_attributes_shortword_little_endian():
+    f3msb = 'test-data/f3.sgy'
+    f3lsb = 'test-data/f3-lsb.sgy'
+    word = segyio.su.dt
+    # this test (in particular) is a pretty good candidate for fuzzing
+    with segyio.open(f3msb) as msb:
+        with segyio.open(f3lsb, endian = 'little') as lsb:
+            msba = msb.attributes(word)
+            lsba = lsb.attributes(word)
+            npt.assert_array_equal(msba[:], lsba[:])
+
 
 @tmpfiles('test-data/f3.sgy')
 def test_writetrace_int16(tmpdir):