Re: /dev/clipboard sometimes fails to set CF_UNICODETEXT data.

[Takashi Yano <takashi.yano@nifty.ne.jp>]

On Tue, 5 Jul 2022 00:42:50 -0700
Mark Geisert wrote:
> Hi Takashi,
> 
> Takashi Yano wrote:
> > Hello,
> > 
> > In one of my PCs,
> > dd if=/dev/urandom count=10 | xxd > /dev/clipboard
> > sometimes fails to set CF_UNICODETEXT data.
> > As a result, pasting clipboard to notepad does not work.
> > 
> > Even in the case, cygnativeformat data is set correctly.
> > So, "cat /dev/clipboard" works.
> > 
> > This problem depends on machine very much.
> > 
> > My one PC with Xeon E3-1281 v3 CPU has a high probability of failure,
> > however, another machine with Core i7-6700K CPU does not.
> > 
> > I looked into this problem, and found OpenClipboard() for
> > CF_UNICODETEXT fails. It seems that OpenClipboard() just
> > after CloseClipboard() has high probability of failure.
> > 
> > You can see the following test case immediately stops with error.
> > Even with Core i7-6700K CPU machine above, the test case fails.

Additional information:
In the machine with Xeon E3-1281 v3 CPU, the test case fails
in a several times of the loop, while it fails after 1000 to
3000 times of the loop in the machine with Core i7-6700K CPU.

The probability of failure depends much on machine.

> > #include <windows.h>
> > #include <stdio.h>
> > 
> > int main()
> > {
> > 	for (;;) {
> > 		if (!OpenClipboard(0)) {
> > 			printf("Open error.\n");
> > 			break;
> > 		}
> > 		if (!EmptyClipboard()) {
> > 			printf("Empty error.\n");
> > 			break;
> > 		}
> > 		if (!CloseClipboard()) {
> > 			printf("Cloes error.\n");
> > 			break;
> > 		}
> > 	}
> > 	return 0;
> > }
> > 
> > I also found the patch attached solves the issue.
> > 
> > I would appreciate any suggestion.
> 
> Would be neet if this patch wasn't needed, but, oh well.  My only comment on the 
> patch would be to possibly call the new wrapper function MyOpenClipboard, or 
> CygOpenClipboard, to help keep the name paired with CloseClipboard.

Thanks for the comment. IIUC, the camel case name should be
used only for Win32 APIs related. However, this is the wrapper
for Win32 API. So, it might be acceptable. In other places,
some wrapper functions are named as camel case.

Corinna, WDYT?

> I suppose it might happen in the future that a faster processor will require the 
> loop count to be made larger.  Once again, oh well.

Maybe. In my most cases, OpenClipboard() successes after one
or two times of retrying. So, I think max_retry = 10 is enough
for the time being.

> My 2c (adjusted for inflation),

-- 
Takashi Yano <takashi.yano@nifty.ne.jp>