From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from dmta0001.nifty.com (mta-snd00002.nifty.com [106.153.226.34]) by sourceware.org (Postfix) with ESMTPS id D95643858C2D for ; Fri, 4 Aug 2023 08:52:13 +0000 (GMT) DMARC-Filter: OpenDMARC Filter v1.4.2 sourceware.org D95643858C2D Authentication-Results: sourceware.org; dmarc=fail (p=none dis=none) header.from=nifty.ne.jp Authentication-Results: sourceware.org; spf=fail smtp.mailfrom=nifty.ne.jp Received: from HP-Z230 by dmta0001.nifty.com with ESMTP id <20230804085211496.TDSC.102246.HP-Z230@nifty.com> for ; Fri, 4 Aug 2023 17:52:11 +0900 Date: Fri, 4 Aug 2023 17:52:11 +0900 From: Takashi Yano To: cygwin-developers@cygwin.com Subject: Re: Rare character glitch in pty? Message-Id: <20230804175211.d2f5aa05b62ca0695f209e14@nifty.ne.jp> In-Reply-To: References: <6f0bfdf8-a557-0540-0367-6846406f941d@towo.net> <44ea96ac-25c5-40b3-f8e1-2d6c02a4c8da@towo.net> <20230802073506.d9f9ad8be3c8a8b585c1c494@nifty.ne.jp> X-Mailer: Sylpheed 3.7.0 (GTK+ 2.24.30; i686-pc-mingw32) Mime-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit X-Spam-Status: No, score=-3.1 required=5.0 tests=BAYES_00,KAM_DMARC_STATUS,NICE_REPLY_A,RCVD_IN_DNSWL_NONE,SPF_HELO_PASS,SPF_PASS,TXREP autolearn=ham autolearn_force=no version=3.4.6 X-Spam-Checker-Version: SpamAssassin 3.4.6 (2021-04-09) on server2.sourceware.org List-Id: Hi Thomas, On Wed, 2 Aug 2023 08:32:39 +0200 Thomas Wolff wrote: > Am 02.08.2023 um 00:35 schrieb Takashi Yano: > > On Tue, 1 Aug 2023 23:05:18 +0200 > > Thomas Wolff wrote: > >> I had reported this to the the main cygwin list: > >> > >> Am 20.07.2023 um 08:33 schrieb Thomas Wolff via Cygwin: > >>> There was a report that on switching terminal windows with the mouse, > >>> focus event reports (as enabled with ESC[?1004h) would sometimes get > >>> mangled. > >>> This is further described in issue reports to mintty/wsltty > >>> (https://github.com/mintty/wsltty/issues/335) and also to tmux > >>> (https://github.com/tmux/tmux/issues/3601). > >>> > >>> I could reproduce the issue without either of tmux or WSL. However, it > >>> happens only once in a few hundred instances: > >>> The escape sequence is changed so that the initial ESC is replaced by > >>> one of the two subsequent characters, so instead of ESC[O, either of > >>> [[O or sometimes O[O would be sent to the foreground application. > >>> > >>> I traced the issue down to the pty that connects mintty and its client > >>> application, verified that the correct sequence is passed into write() > >>> at the pty and the mangled sequence appears at read() in the client > >>> process. In addition to a minimised test program that behaves like > >>> `cat -v`, I also also put the test output directly into the client > >>> process within the mintty code. > >>> This leaves the only conclusion that the issue must be happening > >>> somehow within pty handling, maybe in the context of some timing > >>> condition or Windows messages being passed to mintty. > >>> Anyhow, I was about to prepare testing with older mintty versions, in > >>> case I overlooked something, when the OP reported that the issue does > >>> not happen with cygwin 3.2.0. > >>> Is there any recent change that could possibly cause such weird > >>> behaviour? > >>> Thomas > >>> > >> I have now automated the test case to chase the issue and associate it > >> with a cygwin version. As unlikely as it may appear, results suggest > >> that there is a case within the pty handling that may change sequences > >> sent through it once in a while, more likely on a system under heavy load. > >> I took some of the latest cygwin snapshots (by the way, there are no > >> newer ones on the snapshots page): > >> I can reproduce the issue with > >>     3.3.5-341.x86_64.snap 2022-02-17 11:53 UTC > >> while it did not occur with > >>     3.4.0-341.x86_64.snap 2022-01-27 14:34 UTC > >> Weird enough, the failing newer one is a cygwin 3.3.5 although the older > >> one is a 3.4.0 version. > > Can you provide the automated test case? I would like to > > investigate what is happning. > > Manual test case: > 1. Run `mintty echokey1` with the attached test program that runs like > `cat -v` and detects the glitch > 2. Run a second test window, same setup to increase the hit rate > 3. Switch these 2 windows by clicking them in change a few hundred > times, until the test program outputs "FAIL" > > Automated test case: > 1. Use the tool TinyTask (https://www.tinytask.net/) to save the > clicking. Sorry, I didn't manage to replace this with a self-made click > simulator. > 2. Place the 2 test windows in a specific position (e.g. covering the > two screen halves in a top position) > 3. Use the Record function of the tool and click to switch windows a > dozen times, stop recording > 4. Save the sequence as a TinyTask script, or use mine (2w.rec > attached), better: load the script (Open), then create a standalone test > executable (button "exe") to create tinytask2w.exe (I can send that to > you if you like) > 5. Optionally, create some background load on your system > 6. Minimize other windows in case the test script does not hit the test > windows > 7. Run the test script minptytest (attached) to start the 2 windows and > run the TinyTask standalone test in a loop; the script detects a marker > file created by the test program; it also detect premature termination > of a window (e.g. by hitting ^D in it) Thanks for the test case. I could reproduce the issue using the automated test case. I looked into this problem, and found the cause. Previously, though readahead buffer handling in pty master was not fully thread-safe, accept_input() was called from peek_pipe() thread in select.cc. The mechanism of the problem is: 1) accept_input() which is called from peek_pipe() thread calls eat_readahead(-1) before reading readahead buffer. This allows writing to the readahead buffer from another (main) thread. 2) The main thread calls fhandler_pty_master::write() just after eat_readahead(-1) was called and before reading the readahead buffer by accept_input() called from peek_pipe() thread. This overwrites the readahead buffer. 3) The read result from readahead buffer which was overwritten is sent to the slave. This issue has existed this decade. I'll submit a patch for this issue shortly. -- Takashi Yano