On Wed, Feb 09, 2011 at 05:14:21PM +0000, corvid wrote:

Jeremy wrote:

...

Valgrind is reporting uninitialised data in DLItem::update() in downloads.cc[1]. Here's the code with the offending lines (705, 706) marked.

/* Rate */ if (csec >= 2) { tsec = (float) (curr_time - twosec_time); ! rate = ((float)(curr_bytesize-twosec_bytesize) / 1024) / tsec; ! snprintf(str, 64, (rate < 100) ? "%.1fK/s" : "%.0fK/s", rate); prRate->copy_label(str); prRate->redraw_label(); }

I'm wondering what is happening here. DLItem::update() is scheduled to be run every second. I think that twosec_time is only set to an initialised value when the function has been run at least twice before. Also, csec is calculated from the difference between the current time and the initial time. Is it possible that the "csec >= 2" could be true even when the function has been run only once before? After all, the timing of callbacks is never exact. Perhaps the callback should explicitly count how many times it has been called rather than attempt to work it out from the timing?

(The list software never did send me a copy of your msg...)

Strange. Although I remember the wearlab was down a few days ago...

Would something like the following work?

--- a/dpi/downloads.cc Tue Jan 11 01:18:28 2011 +0000 +++ b/dpi/downloads.cc Sun Feb 06 20:40:53 2011 +0000 @@ -318,6 +318,8 @@ // Init value. Reset later, upon the first data bytes arrival init_time = time(NULL);

+ twosec_time = onesec_time = init_time; + // BUG:? test a URL with ' inside. /* escape "'" character for the shell. Is it necessary? */ esc_url = Escape_uri_str(url, "'");

Most probably. I have a vague memory of noticing the issue when coding it long ago, but not caring much as the following second the display would be updated with correct data. -- Cheers Jorge.-