On Mon, 21 Oct 2013, aaron.sto...@us.mahle.com wrote:
Buffer size change (we have no idea what the 33 was for, but it was
probably required in the old version)
-#define WRITEBUFFERSIZE (1024*32)-33
+#define WRITEBUFFERSIZE 1024*128
Ah, you're talking about the buffer size you pass in to a libssh2 function.
libssh2_sftp_write() I take it.
Behavior wasn't reproducible when I moved the code, it instead 'hung' at a
different part of the file transfer (in terms of bytes). Haven't tried
putting it in a separate application yet. As it won't reproducible
independently, we haven't been able to test if there is some magic file size
that causes the problem.
But if the problem can't be repeated independently, how can you then be sure
it is due to the buffer size change? If it truly is a problem with a 32K
buffer, then surely it won't work if you write a small stand-alone program
using a 32K upload buffer? Could it be that the -33 is causing the problem?
We're using Windows, server is Tectia on Windows. The application was
compiled on a 32 bit Windows XP machine using Builder XE.
We also switched from openssl to libgcrypt when we migrated from 1.2.7 to
1.4.3
And you decided that could not be a reason for the new problem?
This happens across multiple client/server environments with differing
amounts of RTT, but here it is essentially negligible.
That's good information as SFTP is fairly sensitive to RTT and could
potentially show different behaviors.
--
/ daniel.haxx.se
_______________________________________________
libssh2-devel http://cool.haxx.se/cgi-bin/mailman/listinfo/libssh2-devel
