Hi Alex, Alexander Lamaison wrote: > SFTP reading is completely broken for me in 1.4.0 (though it has been > for a while). The last release that works for me is 1.2.7.
Bad. We need to fix this. > Some revisions cause a read with a buffer larger than 65-130k to > block. Other revisions cause reads to return 0 prematurely indicating > the end of the file (usually around the 200k mark). Recent revisions > do the latter. > > Using my usual test script (attached) and git bisect I've narrowed > down the offending commits. Apparently, premature EOF is caused by > 03ca9020756a4e16f0294e5b35e9826ee6af2364 [1] and blocking is caused by > 90b4b4073f34919aa72deff61a5c9bc188c47c95 [2]. It's difficult to know > how accurate these are as there were several overlapping, interacting > bugs and attempts to fix them that kept changing the SFTP behaviour. > > This is all on Windows 7 64-bit. To try the script yourself, run it as: > libssh2_bigread ip_address username password path > If you run it with something like /dev/zero as the path, the expected > behaviour is that it will try reading from that path with increasing > buffer sizes up to about 8MB. > > Daniel, liuzl: we went through this process in September and you guys > made some fixes which I though had solved it. I've changed both the > Windows and Linux machines I'm testing on since then so I don't know > if that's why the problems are back. I don't understand enough about > the SFTP layer to suggest a fix but, as usual, I'll happily test any > patches. Could you test if this happens also from Linux to Linux? Would be good to test both on localhost on the server where Windows 7 shows this problem, and between two different Linux machines. Could you stop by IRC soonish for some more discussion and testing? //Peter _______________________________________________ libssh2-devel http://cool.haxx.se/cgi-bin/mailman/listinfo/libssh2-devel
