[Mono-dev] [PATCH]System.IO.File.Replace

[Paolo Molaro]

On 08/17/07 Bill Holmes wrote:
> Here is an updated version of my patch.  I apologize in advance if my
> attachment is a "binary blob" again.  I am not quite sure what that
> meant or what I should do about it.  This time I am sending the mail
> from SUSE so maybe that will help.

It looks good now:) Binary blob means that the patch you sent was marked
as an attachment of mime-type octect-stream (octect-stream is another
name for binary blob) instead of the text file it is.

> Index: mono/mono/io-layer/io.c
> ===================================================================
> --- mono/mono/io-layer/io.c	(revision 84326)
> +++ mono/mono/io-layer/io.c	(working copy)
> @@ -1835,6 +1835,55 @@
>  	return(ret);
>  }
>  
> +gboolean write_file (int src_fd, int dest_fd, struct stat *st_src, gboolean report_errors)
> +{
> +	int remain, n;
> +	char *buf;
> +	int buf_size = st_src->st_blksize;
> +	gboolean ret = FALSE;
> +
> +	buf_size = buf_size < 8192 ? 8192 : (buf_size > 65536 ? 65536 : buf_size);
> +	buf = (char *) malloc (buf_size);
> +
> +	for (;;) {
> +		remain = read (src_fd, buf, buf_size);
> +		if (remain < 0) {
> +			if (errno == EINTR && !_wapi_thread_cur_apc_pending ())
> +				continue;
> +
> +			if (report_errors)
> +				_wapi_set_last_error_from_errno ();
> +
> +			goto write_file_cleanup;
> +		}
> +		if (remain == 0) {
> +			break;
> +		}
> +
> +		while (remain > 0) {
> +			if ((n = write (dest_fd, buf, remain)) < 0) {
> +				if (errno == EINTR && !_wapi_thread_cur_apc_pending ())
> +					continue;
> +
> +				if (report_errors)
> +					_wapi_set_last_error_from_errno ();
> +#ifdef DEBUG
> +				g_message ("%s: write failed.", __func__);
> +#endif
> +				goto write_file_cleanup;
> +			}
> +
> +			remain -= n;
> +		}
> +	}
> +
> +	ret = TRUE;
> +
> +write_file_cleanup:
> +	free (buf);
> +	return ret ;

In this function I wouldn't use goto (I don't think I asked for this
change): the cleanup code is very simple.
Anyway, this change made me incpect the code more and there seems to be
a bug: in the case of a short write, we try to repeatedly write always
from the start of the buffer, instead of advancing into it. This causes
corruption and needs fixing.

> +gchar* convert_arg_to_utf8 (const gunichar2 *arg, const gchar *arg_name)

Here and in the other places, put the function name at the start of a
line, with the return type and storage type in their own line.

Thanks!
lupus

-- 
-----------------------------------------------------------------
lupus at debian.org                                     debian/rules
lupus at ximian.com                             Monkeys do it better