This is the mail archive of the
mailing list for the Cygwin project.
Re: /dev/clipboard corrupted
Am 28.06.2012 10:20, schrieb Corinna Vinschen:
On Jun 28 00:16, Thomas Wolff wrote:
Not really reproducible (maybe later) but some more observations.
If the clipboard contains large data, the contents retrieved from
/dev/clipboard gets corrupted.
I compared the following in a few cases:
* cat /dev/clipboard or cp /dev/clipboard (which are equal)
* mouse-paste into mintty, read with cat
* read /dev/clipboard within application
and I found all three results to be different, the correct one
sometimes being cat /dev/clipboard but not always.
In today's case, the differences occured at byte 10240 and 65536
respectively, thus 10K-1 bytes or 64K-1 bytes being equal.
I can't reprocude this. There's also nothing in the /dev/clipboard
code which would rely on a 10K buffer or so. 64K, yes. But still,
I tried with wordpad, vi, mintty, cat, and cp with a text file of
about 90K. No problems. Do you have any reproducible testcase?
I made a small test program to read /dev/clipboard directly with
different buffer sizes.
I verified that this happens at borders of read() invocations.
It is more likely to happen if there are non-ASCII characters just at
this border, however, in one case it occurred after 2 blocks read where
the non-ASCII was between blocks 1 and 2. Also, in one case, a CRLF
newline at a border was just skipped (with non-ASCII somewhere before,
probably). Typically, I see 2 bytes skipped, sometimes 1 but sometimes
also ~100. Quite weird.
I looked at the fhandler_clipboard read() code and somehow suspect the
wcstomb although the observations are not strictly conclusive.
(It's the non-cygwin-specific branch, after filling the clipboard from
Windows, e.g. Firefox.)
Problem reports: http://cygwin.com/problems.html
Unsubscribe info: http://cygwin.com/ml/#unsubscribe-simple