Fix: Decoding filters that didn't detect an EOD in the source data still

tried to look 1 byte ahead, occasionally causing obscure errors.  The fix
unfortunately required adding a new architectural requirement to the
implementation of such filters.  (fix from igor@artifex.com)

git-svn-id: http://svn.ghostscript.com/ghostscript/trunk@851 a1074d23-0009-0410-80fe-cf8c14f379e6

14 files changed, 32 insertions, 12 deletions

diff --git a/gs/doc/News.htm b/gs/doc/News.htm
index 48d989688..26ba9fc45 100644
--- a/gs/doc/News.htm
+++ b/gs/doc/News.htm
@@ -75,6 +75,9 @@ metrics computed from FontBBox rather than the horizontal metrics.
 (PDF interpreter)
 	- The interpreter now skips to the next EOL after reading the data
 following the ID operator.
+(Streams)
+	- All decoding filters that recognize an EOD in the source data
+must now have an init procedure that sets state->min_left = 1.
 
 </pre>
 
diff --git a/gs/src/sa85d.h b/gs/src/sa85d.h
index e3ccaa9f3..63dc76a5a 100644
--- a/gs/src/sa85d.h
+++ b/gs/src/sa85d.h
@@ -36,7 +36,7 @@ typedef struct stream_A85D_state_s {
 /* We define the initialization procedure here, so that the scanner */
 /* can avoid a procedure call. */
 #define s_A85D_init_inline(ss)\
-  ((ss)->word = 0, (ss)->odd = 0)
+  ((ss)->min_left = 1, (ss)->word = 0, (ss)->odd = 0)
 extern const stream_template s_A85D_template;
 
 #endif /* sa85d_INCLUDED */
diff --git a/gs/src/sbhc.c b/gs/src/sbhc.c
index 342c42c88..763a2282c 100644
--- a/gs/src/sbhc.c
+++ b/gs/src/sbhc.c
@@ -183,13 +183,14 @@ s_BHCD_init(register stream_state * st)
     min(hcd_initial_bits, ss->definition.num_counts);
     uint dsize = hc_sizeof_decoding(&ss->definition, initial_bits);
     hcd_code *decode = ss->decode.codes =
-    (hcd_code *) gs_alloc_byte_array(st->memory, dsize,
-				     sizeof(hcd_code), "BHCD decode");
+	(hcd_code *) gs_alloc_byte_array(st->memory, dsize,
+					 sizeof(hcd_code), "BHCD decode");
 
     if (decode == 0)
 	return ERRC;
 /****** WRONG ******/
     hc_make_decoding(decode, &ss->definition, initial_bits);
+    st->min_left = 1;
     return s_BHCD_reinit(st);
 }
 
diff --git a/gs/src/scfd.c b/gs/src/scfd.c
index c06389fe1..8b3cff7f0 100644
--- a/gs/src/scfd.c
+++ b/gs/src/scfd.c
@@ -75,6 +75,7 @@ s_CFD_init(stream_state * st)
     ss->rpos = ss->wpos = raster - 1;
     ss->eol_count = 0;
     ss->invert = white;
+    ss->min_left = 1;
     return 0;
 }
 
diff --git a/gs/src/scommon.h b/gs/src/scommon.h
index 7dd0318c9..897fcef5b 100644
--- a/gs/src/scommon.h
+++ b/gs/src/scommon.h
@@ -157,6 +157,7 @@ stream_proc_report_error(s_no_report_error);
 	const stream_template *template;\
 	gs_memory_t *memory;\
 	stream_proc_report_error((*report_error));\
+        int min_left; /* required bytes for lookahead */ \
 	char error_string[STREAM_MAX_ERROR_STRING + 1]
 struct stream_state_s {
     stream_state_common;
diff --git a/gs/src/sfilter1.c b/gs/src/sfilter1.c
index 9bd802f20..c2d86748f 100644
--- a/gs/src/sfilter1.c
+++ b/gs/src/sfilter1.c
@@ -169,6 +169,8 @@ s_SFD_init(stream_state * st)
 
     ss->match = 0;
     ss->copy_count = 0;
+    ss->min_left = (ss->eod.size != 0);
+
     return 0;
 }
 
diff --git a/gs/src/slzwd.c b/gs/src/slzwd.c
index 0e1b264a6..1cb79f5d9 100644
--- a/gs/src/slzwd.c
+++ b/gs/src/slzwd.c
@@ -84,6 +84,7 @@ s_LZWD_init(stream_state * st)
 	return ERRC;
 /****** WRONG ******/
     ss->table.decode = dc;
+    ss->min_left = 1;
     return s_LZWD_reset(st);
 }
 
diff --git a/gs/src/srlx.h b/gs/src/srlx.h
index 74579bb19..5937392e2 100644
--- a/gs/src/srlx.h
+++ b/gs/src/srlx.h
@@ -66,7 +66,7 @@ typedef struct stream_RLD_state_s {
 #define s_RLD_set_defaults_inline(ss)\
   ((ss)->EndOfData = true)
 #define s_RLD_init_inline(ss)\
-  ((ss)->copy_left = 0)
+  ((ss)->min_left = ((ss)->EndOfData ? 1 : 0), (ss)->copy_left = 0)
 extern const stream_template s_RLD_template;
 
 #endif /* srlx_INCLUDED */
diff --git a/gs/src/sstring.h b/gs/src/sstring.h
index 79f557ca9..a2168c26a 100644
--- a/gs/src/sstring.h
+++ b/gs/src/sstring.h
@@ -48,7 +48,7 @@ typedef struct stream_AXD_state_s {
   gs_private_st_simple(st_AXD_state, stream_AXD_state,\
     "ASCIIHexDecode state")
 #define s_AXD_init_inline(ss)\
-  ((ss)->odd = -1, 0)
+  ((ss)->min_left = 1, (ss)->odd = -1, 0)
 extern const stream_template s_AXD_template;
 
 /* PSStringDecode */
diff --git a/gs/src/stream.c b/gs/src/stream.c
index 6d5a7f059..af31172d5 100644
--- a/gs/src/stream.c
+++ b/gs/src/stream.c
@@ -106,6 +106,7 @@ s_init(stream *s, gs_memory_t * mem)
 {
     s->memory = mem;
     s->report_error = s_no_report_error;
+    s->min_left = 0;
     s->error_string[0] = 0;
     s->prev = s->next = 0;	/* clean for GC */
     s->file_name.data = 0;	/* ibid. */
@@ -133,6 +134,7 @@ s_init_state(stream_state *st, const stream_template *template,
     st->template = template;
     st->memory = mem;
     st->report_error = s_no_report_error;
+    st->min_left = 0;
 }
 stream_state *
 s_alloc_state(gs_memory_t * mem, gs_memory_type_ptr_t stype,
@@ -420,15 +422,14 @@ sclose(register stream * s)
 /*
  * Define the minimum amount of data that must be left in an input buffer
  * after a read operation to handle filter read-ahead.  This is 1 byte for
- * filters (including procedure data sources) that haven't reached EOD,
- * 0 for files.
+ * filters (including procedure data sources) that require EOD but
+ * haven't reached EOD, 0 for files and for filters that have no EOD.
  */
 int
 sbuf_min_left(const stream *s)
 {
-    return
-	(s->strm == 0 ? (s->end_status != CALLC ? 0 : 1) :
-	 s->end_status == EOFC || s->end_status == ERRC ? 0 : 1);
+    return 
+      (s->end_status == EOFC || s->end_status == ERRC ? 0 : s->state->min_left);
 }
 
 /*
diff --git a/gs/src/strimpl.h b/gs/src/strimpl.h
index d4f894832..45ba3485c 100644
--- a/gs/src/strimpl.h
+++ b/gs/src/strimpl.h
@@ -109,6 +109,14 @@
  */
 
 /*
+ * Note that all decoding filters that require an explicit EOD in the
+ * source data must have an init procedure that sets min_left = 1.
+ * This effectively provides a 1-byte lookahead in the source data,
+ * which is required so that the stream can close itself "after reading
+ * the last byte of data" (per Adobe specification), as noted above.
+ */
+
+/*
  * Define a template for creating a stream.
  *
  * The meaning of min_in_size and min_out_size is the following:
diff --git a/gs/src/szlibd.c b/gs/src/szlibd.c
index ea5e3f28a..2007fa900 100644
--- a/gs/src/szlibd.c
+++ b/gs/src/szlibd.c
@@ -40,6 +40,7 @@ s_zlibD_init(stream_state * st)
 	s_zlib_free_dynamic_state(ss);
 	return ERRC;	/****** WRONG ******/
     }
+    st->min_left=1;
     return 0;
 }
 
diff --git a/gs/src/zfile.c b/gs/src/zfile.c
index 33fd3b483..951b9ff55 100644
--- a/gs/src/zfile.c
+++ b/gs/src/zfile.c
@@ -848,9 +848,9 @@ filter_open(const char *file_access, uint buffer_size, ref * pfile,
     } else if (st != 0)		/* might not have client parameters */
 	memcpy(sst, st, ssize);
     s->state = sst;
-    sst->template = template;
-    sst->memory = mem;
+    s_init_state(sst, template, mem);
     sst->report_error = filter_report_error;
+
     if (template->init != 0) {
 	code = (*template->init)(sst);
 	if (code < 0) {
diff --git a/gs/src/zfproc.c b/gs/src/zfproc.c
index 0b3c46a7f..3cc3c4104 100644
--- a/gs/src/zfproc.c
+++ b/gs/src/zfproc.c
@@ -80,6 +80,7 @@ s_proc_init(ref * sop, stream ** psstrm, uint mode,
     sstrm->procs.process = temp->process;
     state->template = temp;
     state->memory = mem;
+    state->min_left = 1;
     state->eof = 0;
     state->proc = *sop;
     make_empty_string(&state->data, a_all);