Chunker: fix wrong EOF assumption[1], check for return type[2]

[1]
This worked incidentally because OSes tend to return at least one page
worth of data when EOF is not reached. Increasing WINDOW_SIZE beyond
the page size might have lead to data loss.

[2]
If read() of the passed Python object returned something not-bytes,
PyBytes_Size returns -1 (ssize_t) which becomes a very larger number for
memcpy()s size_t.

borg/_chunker.c

@@ -174,6 +174,10 @@ chunker_fill(Chunker *c)
             return 0;
         }
         n = PyBytes_Size(data);
+        if(PyErr_Occurred()) {
+            // we wanted bytes(), but got something else
+            return 0;
+        }
         if(n) {
             memcpy(c->data + c->position + c->remaining, PyBytes_AsString(data), n);
             c->remaining += n;
@@ -205,7 +209,7 @@ chunker_process(Chunker *c)
             return NULL;
         }
     }
-    if(c->remaining < window_size) {
+    if(c->eof) {
         c->done = 1;
         if(c->remaining) {
             c->bytes_yielded += c->remaining;

borg/testsuite/chunker.py

@@ -1,7 +1,7 @@
 from io import BytesIO
 
 from ..chunker import Chunker, buzhash, buzhash_update
-from ..archive import CHUNK_MAX_EXP
+from ..archive import CHUNK_MAX_EXP, CHUNKER_PARAMS
 from . import BaseTestCase
 
 
@@ -29,3 +29,14 @@ def test_buzhash(self):
         self.assert_equal(buzhash(b'abcdefghijklmnop', 1), buzhash_update(buzhash(b'Xabcdefghijklmno', 1), ord('X'), ord('p'), 16, 1))
         # Test with more than 31 bytes to make sure our barrel_shift macro works correctly
         self.assert_equal(buzhash(b'abcdefghijklmnopqrstuvwxyzabcdefghijklmnopqrstuvwxyz', 0), 566521248)
+
+    def test_small_reads(self):
+        class SmallReadFile:
+            input = b'a' * (20 + 1)
+
+            def read(self, nbytes):
+                self.input = self.input[:-1]
+                return self.input[:1]
+
+        reconstructed = b''.join(Chunker(0, *CHUNKER_PARAMS).chunkify(SmallReadFile()))
+        assert reconstructed == b'a' * 20