Re: view growbuf read until() suggestion

[Roland Illig <roland illig gmx de>]

Pavel Tsekov wrote:
> Of course you might still be right in assuming that fread() may return less
> bytes than requested. I can imagine that it could happen with a fdopen()-ed
> FILE which has a file descriptor in non-blocking mode. This is of coure just
> speculation. Of course, calls to feof() and ferror() could be added to
> detect the situation where the read didn't fail but still returned less
> data.

ISO C99, 7.19.3p3:
[...] characters may be accumulated to or from the host environment as a 
block. [...] characters are intended to be transmitted [...] when a 
block is filled. [...]

The relevant part is the "may" and "are intended to". These are not 
strict requirements, so we cannot rely on it.

My patch has the side effect that mcview reads one byte more than 
strictly necessary in some situations. As it already reads data in 
blocks of 8k, this should not harm.

Roland

Index: view.c
===================================================================
RCS file: /cvsroot/mc/mc/src/view.c,v
retrieving revision 1.313
diff -u -p -r1.313 view.c
--- view.c	31 Jul 2005 20:29:35 -0000	1.313
+++ view.c	6 Aug 2005 01:49:16 -0000
@@ -376,13 +376,15 @@ view_growbuf_read_until (WView *view, of
     ssize_t nread;
     byte *p;
     size_t bytesfree;
+    gboolean short_read;
 
     assert (view->growbuf_in_use);
 
     if (view->growbuf_finished)
 	return;
 
-    while (view_growbuf_filesize (view) < ofs) {
+    short_read = FALSE;
+    while (view_growbuf_filesize (view) < ofs || short_read) {
 	if (view->growbuf_blocks == 0 || view->growbuf_lastindex == VIEW_PAGE_SIZE) {
 	    /* Append a new block to the growing buffer */
 	    byte *newblock = g_try_malloc (VIEW_PAGE_SIZE);
@@ -421,6 +423,7 @@ view_growbuf_read_until (WView *view, of
 		return;
 	    }
 	}
+	short_read = (nread < bytesfree);
 	view->growbuf_lastindex += nread;
     }
 }