Get tracecount from arg in create, not header

The cheeky trick of passing the number of traceheaders in the binary
header when creating new files, in a field that is not dedicated to
number-of-traces-per-file, doesn't work because it's only 16 bits and
overflows very fast, and SEG-Y allows for 32-bits of traces (governed by
traceno).

Instead, pass the override-number-of-traces argument to the internal
filehandle constructor explicitly, and only consider that number if the
binary header is also passed.

This fixes the issue reported in
equinor#235

python/segyio/create.py

@@ -135,19 +135,19 @@ def create(filename, spec):
     if not structured(spec):
         tracecount = spec.tracecount
     else:
-        tracecount    = len(spec.ilines) * len(spec.xlines) * len(spec.offsets)
+        tracecount = len(spec.ilines) * len(spec.xlines) * len(spec.offsets)
 
     ext_headers = spec.ext_headers if hasattr(spec, 'ext_headers') else 0
     samples = numpy.asarray(spec.samples, dtype = numpy.single)
 
     binary = bytearray(_segyio.binsize())
-    _segyio.putfield(binary, 3213, tracecount)
     _segyio.putfield(binary, 3217, 4000)
     _segyio.putfield(binary, 3221, len(samples))
     _segyio.putfield(binary, 3225, int(spec.format))
     _segyio.putfield(binary, 3505, int(ext_headers))
 
-    f = segyio.SegyFile(str(filename), "w+", binary = binary)
+    f = segyio.SegyFile(str(filename), "w+", tracecount = tracecount,
+                                             binary = binary)
 
     f._il            = int(spec.iline)
     f._xl            = int(spec.xline)
@@ -161,7 +161,7 @@ def create(filename, spec):
         f._xlines        = numpy.copy(numpy.asarray(spec.xlines, dtype=numpy.intc))
 
         line_metrics = _segyio.line_metrics(f.sorting,
-                                            f.tracecount,
+                                            tracecount,
                                             len(f.ilines),
                                             len(f.xlines),
                                             len(f.offsets))

python/segyio/segy.py

@@ -36,7 +36,7 @@ class SegyFile(object):
 
     _unstructured_errmsg = "File opened in unstructured mode."
 
-    def __init__(self, filename, mode, iline=189, xline=193, binary=None):
+    def __init__(self, filename, mode, iline=189, xline=193, tracecount=0, binary=None):
         """
         Constructor, internal.
         """
@@ -64,7 +64,7 @@ def __init__(self, filename, mode, iline=189, xline=193, binary=None):
         self._xline = None
         self._gather = None
 
-        self.xfd = _segyio.segyiofd(filename, mode, binary)
+        self.xfd = _segyio.segyiofd(filename, mode, tracecount, binary)
         metrics = self.xfd.metrics()
         self._fmt = metrics['format']
         self._tracecount = metrics['tracecount']

python/segyio/segyio.cpp

@@ -221,9 +221,13 @@ namespace fd {
 int init( segyiofd* self, PyObject* args, PyObject* ) {
     char* filename = NULL;
     char* mode = NULL;
+    int tracecount = 0;
     buffer_guard buffer;
 
-    if( !PyArg_ParseTuple( args, "ss|z*", &filename, &mode, &buffer ) )
+    if( !PyArg_ParseTuple( args, "ss|iz*", &filename,
+                                           &mode,
+                                           &tracecount,
+                                           &buffer ) )
         return -1;
 
     const char* binary = buffer.buf< const char >();
@@ -261,7 +265,6 @@ int init( segyiofd* self, PyObject* args, PyObject* ) {
         return -1;
     }
 
-    int tracecount = 0;
     char bin[ SEGY_BINARY_HEADER_SIZE ] = {};
 
     if( !binary ) {
@@ -273,13 +276,6 @@ int init( segyiofd* self, PyObject* args, PyObject* ) {
         }
 
         binary = bin;
-    } else {
-        /*
-         * usually we wouldn't trust the bin-traces in the binary header, but
-         * this code path is only taken with create() (or similarly
-         * pre-verified header), in which case we can just read the tracecount.
-         */
-        segy_get_bfield( binary, SEGY_BIN_TRACES, &tracecount );
     }
 
     const long trace0 = segy_trace0( binary );

python/test/segy.py

@@ -1065,4 +1065,25 @@ def value(x, y):
         for index, depth_slice in enumerate(f.depth_slice):
             assert np.allclose(depth_slice, buf * index)
 
-
+def test_no_16bit_overflow_tracecount(tmpdir):
+    spec = segyio.spec()
+    spec.format = 1
+    spec.sorting = 2
+    spec.samples = np.arange(501)
+    spec.ilines = np.arange(345)
+    spec.xlines = np.arange(250)
+
+    # build a file with more than 65k traces, which would cause a 16bit int to
+    # overflow.
+    # see https://github.com/Statoil/segyio/issues/235
+    ones = np.ones(len(spec.samples), dtype = np.single)
+    with segyio.create(tmpdir / 'foo.sgy', spec) as f:
+        assert f.tracecount > 0
+        assert f.tracecount > 2**16 - 1
+        for i in range(f.tracecount):
+            f.trace[i] = ones
+            f.header[i] = {
+                    segyio.TraceField.INLINE_3D: i,
+                    segyio.TraceField.CROSSLINE_3D: i,
+                    segyio.TraceField.offset: 1,
+            }

python/test/segyio_c.py

@@ -107,7 +107,11 @@ def write_text_header(f, mmap):
     f.close()
 
 
-def get_instance_segyiofd(tmpdir, file_name="small.sgy", mode="r+", binary=None):
+def get_instance_segyiofd(tmpdir,
+                          file_name="small.sgy",
+                          mode="r+",
+                          tracecount=0,
+                          binary=None):
     path = str(tmpdir)
     f = os.path.join(path, file_name)
     if binary is not None:
@@ -424,18 +428,20 @@ def mkempty():
 @tmpfiles("test-data/small.sgy")
 def test_read_and_write_trace_mmap(tmpdir):
     binary = bytearray(_segyio.binsize())
-    _segyio.putfield(binary, 3213, 100)
     _segyio.putfield(binary, 3221, 25)
-    f = get_instance_segyiofd(tmpdir, "trace-wrt.sgy", "w+", binary)
+    f = get_instance_segyiofd(tmpdir, "trace-wrt.sgy", "w+",
+                              tracecount=100,
+                              binary=binary)
     read_and_write_trace(f, True)
 
 
 @tmpfiles("test-data/small.sgy")
 def test_read_and_write_trace(tmpdir):
     binary = bytearray(_segyio.binsize())
-    _segyio.putfield(binary, 3213, 100)
     _segyio.putfield(binary, 3221, 25)
-    f = get_instance_segyiofd(tmpdir, "trace-wrt.sgy", "w+", binary)
+    f = get_instance_segyiofd(tmpdir, "trace-wrt.sgy", "w+",
+                              tracecount=100,
+                              binary=binary)
     read_and_write_trace(f, False)