Graphical emacs doesn't work; spins on poll()/writev()/recvmsg()

[aseering]

Splitting this out from #187 so that it doesn't get lost.

Steps to reproduce:

• Install Xming on Windows
• Install emacs in WSL: sudo apt-get install emacs24
• Set export DISPLAY=localhost:0 in a WSL terminal
• Run emacs in the same WSL terminal

Expected:

• A graphical emacs window should appear, rendered by Xming

Actual:

• This appears:

And emacs hangs / is unresponsive. The X button in the top-right corner does not close it. Ctrl-C in the WSL terminal window doesn't close it either. It's necessary to get a shell prompt and do a kill -9 to get it to close.

[aseering]

strace emacs looks like a constant stream of:

...
poll([{fd=3, events=POLLIN|POLLOUT}], 1, 4294967295) = 1 ([{fd=3, revents=POLLOUT}])
writev(3, [{"+\0\1\0", 4}, {NULL, 0}, {"", 0}], 3) = 4
poll([{fd=3, events=POLLIN}], 1, 4294967295) = 1 ([{fd=3, revents=POLLIN}])
recvmsg(3, {msg_name(0)=NULL, msg_iov(1)=[{"\1\1]\n\0\0\0\0\22\0\200\0\0\0\0\0\1\0\0\0T\4\0\0\320\312\213\2\4\0\0\0", 4096}], msg_controllen=0, msg_flags=0}, 0) = 32
recvmsg(3, 0x7fffc79d9f30, 0)           = -1 EAGAIN (Resource temporarily unavailable)
recvmsg(3, 0x7fffc79d9f30, 0)           = -1 EAGAIN (Resource temporarily unavailable)
recvmsg(3, 0x7fffc79d9f40, 0)           = -1 EAGAIN (Resource temporarily unavailable)
poll([{fd=3, events=POLLIN|POLLOUT}], 1, 4294967295) = 1 ([{fd=3, revents=POLLOUT}])
writev(3, [{"+\0\1\0", 4}, {NULL, 0}, {"", 0}], 3) = 4
poll([{fd=3, events=POLLIN}], 1, 4294967295) = 1 ([{fd=3, revents=POLLIN}])
recvmsg(3, {msg_name(0)=NULL, msg_iov(1)=[{"\1\1^\n\0\0\0\0\22\0\200\0\0\0\0\0\1\0\0\0T\4\0\0\320\312\213\2\4\0\0\0", 4096}], msg_controllen=0, msg_flags=0}, 0) = 32
recvmsg(3, 0x7fffc79d9f00, 0)           = -1 EAGAIN (Resource temporarily unavailable)
recvmsg(3, 0x7fffc79d9f00, 0)           = -1 EAGAIN (Resource temporarily unavailable)
poll([{fd=3, events=POLLIN|POLLOUT}], 1, 4294967295) = 1 ([{fd=3, revents=POLLOUT}])
writev(3, [{"+\0\1\0", 4}, {NULL, 0}, {"", 0}], 3) = 4
poll([{fd=3, events=POLLIN}], 1, 4294967295) = 1 ([{fd=3, revents=POLLIN}])
recvmsg(3, {msg_name(0)=NULL, msg_iov(1)=[{"\1\1_\n\0\0\0\0\22\0\200\0\0\0\0\0\1\0\0\0T\4\0\0\320\312\213\2\4\0\0\0", 4096}], msg_controllen=0, msg_flags=0}, 0) = 32
recvmsg(3, 0x7fffc79d9f30, 0)           = -1 EAGAIN (Resource temporarily unavailable)
recvmsg(3, 0x7fffc79d9f30, 0)           = -1 EAGAIN (Resource temporarily unavailable)
recvmsg(3, 0x7fffc79d9f40, 0)           = -1 EAGAIN (Resource temporarily unavailable)
poll([{fd=3, events=POLLIN|POLLOUT}], 1, 4294967295) = 1 ([{fd=3, revents=POLLOUT}])
...

being dumped out about as quickly as I would expect the Windows Console to be able to render it.

[aseering]

Interestingly, if I open up a WSL terminal and do ssh -X real-ubuntu-1404-server, then run emacs remotely, it shows up correctly in Windows/Xming, and the application is responsive.

In this setup, strace ssh looks a lot like:

...
select(10, [3 4 7 8 9], [], NULL, NULL) = 1 (in [3])
clock_gettime(CLOCK_MONOTONIC, {97174, 557010000}) = 0
read(3, "\0\0\5\360\7<\7\270\374\301\r\215\215D\331\272\363\326o\203N\0173\233\355F?\31=\36\313|"..., 8192) = 1592
clock_gettime(CLOCK_MONOTONIC, {97174, 559067000}) = 0
clock_gettime(CLOCK_MONOTONIC, {97174, 559545000}) = 0
select(10, [3 4 7 8 9], [8], NULL, NULL) = 1 (out [8])
clock_gettime(CLOCK_MONOTONIC, {97174, 561989000}) = 0
write(8, "=\0\4\0\27\0\200\0J\2\334\1\20\0\16\0\221\4\6\0\216\0\200\0\27\0\200\0G\0\0\0"..., 1508) = 1508
clock_gettime(CLOCK_MONOTONIC, {97174, 563682000}) = 0
clock_gettime(CLOCK_MONOTONIC, {97174, 564052000}) = 0
select(10, [3 4 7 8 9], [], NULL, NULL) = 1 (in [8])
clock_gettime(CLOCK_MONOTONIC, {97174, 566094000}) = 0
read(8, "\1\0018\7\0\0\0\0\22\0\200\0\0\0\0\0\0\0\0\0l\4\0\0\260\314\213\2\4\0\0\0", 16384) = 32
clock_gettime(CLOCK_MONOTONIC, {97174, 568366000}) = 0
clock_gettime(CLOCK_MONOTONIC, {97174, 569510000}) = 0
select(10, [3 4 7 8 9], [3], NULL, NULL) = 1 (out [3])
clock_gettime(CLOCK_MONOTONIC, {97174, 572291000}) = 0
write(3, "\0\0\0000\300\331}3>\31\331_iQ\200\336l+\340xj\331\245\334\3468{\23\202,\327\357"..., 68) = 68
..

And strace emacs looks like this:

...
poll([{fd=4, events=POLLIN}, {fd=3, events=POLLIN}, {fd=5, events=POLLIN}], 3, 0) = 0 (Timeout)
pselect6(4, [3], [], NULL, {0, 342785043}, {NULL, 8}) = 0 (Timeout)
rt_sigprocmask(SIG_BLOCK, [WINCH IO], NULL, 8) = 0
rt_sigprocmask(SIG_UNBLOCK, [WINCH IO], NULL, 8) = 0
poll([{fd=3, events=POLLIN|POLLOUT}], 1, 4294967295) = 1 ([{fd=3, revents=POLLOUT}])
writev(3, [{";\0\5\0\32\0\200\0\0\0\0\0\n\0\0\0\7\0\16\0\221\6\5\0@\0\200\0\0\0\0\0"..., 124}, {NULL, 0}, {"", 0}], 3) = 124
recvmsg(3, 0x7ffd8ade9600, 0)           = -1 EAGAIN (Resource temporarily unavailable)
recvmsg(3, 0x7ffd8ade9610, 0)           = -1 EAGAIN (Resource temporarily unavailable)
recvmsg(3, 0x7ffd8ade8f40, 0)           = -1 EAGAIN (Resource temporarily unavailable)
poll([{fd=4, events=POLLIN}, {fd=3, events=POLLIN}, {fd=5, events=POLLIN}], 3, 0) = 0 (Timeout)
pselect6(4, [3], [], NULL, {0, 497207317}, {NULL, 8}) = 0 (Timeout)
rt_sigprocmask(SIG_BLOCK, [WINCH IO], NULL, 8) = 0
rt_sigprocmask(SIG_UNBLOCK, [WINCH IO], NULL, 8) = 0
poll([{fd=3, events=POLLIN|POLLOUT}], 1, 4294967295) = 1 ([{fd=3, revents=POLLOUT}])
writev(3, [{";\0\5\0C\0\200\0\0\0\0\0\n\0\0\0000\2\16\0\221\6\5\0@\0\200\0\0\0\0\0"..., 124}, {NULL, 0}, {"", 0}], 3) = 124
...

In this scenario, emacs is emitting syscalls considerably more slowly than it was when running locally. The server in question is slower. But I wonder whether perhaps some syscall is not causing its thread to de-schedule for as long as it would under Linux, resulting in some sort of starvation?

[aseering]

@xarthurx mentioned a somewhat similar problem in a comment on #187 , but with different symptoms. I wonder if this could be a race condition of some sort.

[aseering]

It's probably worth pointing out that command-line emacs works just fine. You can verify this by either un-setting the DISPLAY environment variable and running emacs, or running emacs -nw.

[redviper]

This has, oddly, been fixed in emacs-25 pretest. I am using graphical emacs atm.

[aseering]

Thanks!

I can confirm that graphical emacs works for me with the "emacs-snapshot" package in the "ppa:ubuntu-elisp" PPA, which currently reports as "GNU Emacs 25.1.50.2".

[lateau]

@aseering @redviper Do you guys have any problem with maximize emacs window? This is my problem emacs-snapshot with Xming. It doesn't refresh current frame fit to maximized window width & height.

$ emacs -fs
$ emacs -mm
$ emacs -fh -fw

These doesn't work for me. Any ideas?

• Surface book
• Build 14379
• Xming
• emacs-snapshot on ppa:ubuntu-elisp

[aseering]

Oh, good point -- yes, I do have that problem now, with the new emacs version.

[lateau]

I tried to run emacs with MobaXTerm. No luck on maximize window, but I think MobaXTerm is better than Xming on display images and more clear fonts.

Some of commands that related to widgets such like GTK does not work, so I guess emacs with some other GUI library would possibly work if it supported.

(menu-bar-mode 0) ;; no effect
(tool-bar-mode 0) ;; no effect too

[lateau]

Frame paramters are not modifiable.

(frame-parameter (selected-frame) 'width) ;; Returns 80
(set-frame-parameter (selected-frame) 'width 300)
(frame-parameter (selected-frame) 'width) ;; Returns 80

;; Whole list of frame parameters
(frame-parameters)

[lateau]

@aseering Would you try this? It seems like emacs's --geomerty option works. Menubar and toolbar does not appear but the frame extended properly.

$ emacs -g 185x53

• Build 14379
• emacs-snapshot(25.1.50.2) on ppa:ubunt-elisp
• MobaXTerm Home Edition

Here's a strace log(constantly loop):

poll([{fd=7, events=POLLIN}, {fd=6, events=POLLIN}], 2, 0) = 0 (Timeout)
clock_gettime(CLOCK_REALTIME, {1467462336, 718280100}) = 0
clock_gettime(CLOCK_REALTIME, {1467462336, 720600500}) = 0
clock_gettime(CLOCK_REALTIME, {1467462336, 723383700}) = 0
clock_gettime(CLOCK_REALTIME, {1467462336, 724630900}) = 0
rt_sigprocmask(SIG_BLOCK, [WINCH IO], NULL, 8) = 0
rt_sigprocmask(SIG_UNBLOCK, [WINCH IO], NULL, 8) = 0
poll([{fd=6, events=POLLIN|POLLOUT}], 1, 4294967295) = 1 ([{fd=6, revents=POLLOUT}])
writev(6, [{";\0\5\0i\0\200\0\0\0\0\0\10\0\34\1\10\0\21\0\212\6\5\0004\0\200\0\0\0\0\0"..., 144}, {NULL, 0}, {"", 0}], 3) = 144
recvmsg(6, 0x7fffd2800ae0, 0)           = -1 EAGAIN (Resource temporarily unavailable)
recvmsg(6, 0x7fffd2800910, 0)           = -1 EAGAIN (Resource temporarily unavailable)
poll([{fd=7, events=POLLIN}, {fd=6, events=POLLIN}], 2, 0) = 0 (Timeout)
clock_gettime(CLOCK_REALTIME, {1467462336, 750162800}) = 0
pselect6(8, [6 7], [], NULL, {0, 460140400}, {NULL, 8}) = 0 (Timeout)
recvmsg(6, 0x7fffd2800400, 0)           = -1 EAGAIN (Resource temporarily unavailable)
poll([{fd=7, events=POLLIN}, {fd=6, events=POLLIN}], 2, 0) = 0 (Timeout)
clock_gettime(CLOCK_REALTIME, {1467462337, 219505000}) = 0
clock_gettime(CLOCK_REALTIME, {1467462337, 221883400}) = 0
clock_gettime(CLOCK_REALTIME, {1467462337, 224421200}) = 0
clock_gettime(CLOCK_REALTIME, {1467462337, 226127100}) = 0
rt_sigprocmask(SIG_BLOCK, [WINCH IO], NULL, 8) = 0
rt_sigprocmask(SIG_UNBLOCK, [WINCH IO], NULL, 8) = 0
poll([{fd=6, events=POLLIN|POLLOUT}], 1, 4294967295) = 1 ([{fd=6, revents=POLLOUT}])
writev(6, [{";\0\5\0]\0\200\0\0\0\0\0\10\0\34\1\310\5\21\0\212\6\5\0004\0\200\0\0\0\0\0"..., 144}, {NULL, 0}, {"", 0}], 3) = 144
recvmsg(6, 0x7fffd2800ae0, 0)           = -1 EAGAIN (Resource temporarily unavailable)
recvmsg(6, 0x7fffd2800910, 0)           = -1 EAGAIN (Resource temporarily unavailable)

[aseering]

@lateau -- yes, -g lets me start with a larger window size.

I don't see a continuous loop of strace output. I see a lot of output whenever I interact with the window in any way -- mouse is over the window, cursor is blinking, etc -- but that's what I would expect given that emacs is interacting with the screen (and over TCP, even) whenever those things happen.

[lateau]

@aseering You're right. Launch emacs with -nbc -Q does not produce that continuous output. Thank you. Loop was my misunderstanding it's just cursor blinking event output.

[redviper]

You can fix the resizing issue by compiling a Lucid version instead of GTK. The problem seems to be something with the GTK resizing code. Get emacs from the git repo (I recommend the GitHub mirror btw, Savannah is slow), and build with ./configure --with-x-toolkit=lucid

You can get all the dependencies worked out before hand too with apt-get build-dep emacs24-lucid.

[masari]

xemacs works fine (although I never use it on real Linux)

[redviper]

Recently I find that the yank for x-selection is broken. Is this happening for others also or have I messed up my WSL configuration?

[lateau]

@redviper It works on my build but sometimes it causes freezing emacs.

[foopang]

I have problem with pasting text into emacs from Windows, while copying text from emacs to windows is just fine. Any workaround?

[fehrin]

Where do i find the emacs-25 pretest package mentioned in this issue?

[lateau]

@fehrin

sudo add-apt-repository ppa:ubuntu-elisp/ppa
sudo apt-get update

[VMatrix1900]

Seems it's fixed in Creator's update, follow the instructions on the link: https://www.reddit.com/r/bashonubuntuonwindows/comments/65dpjd/getting_xapplications_running_in_ubuntu_16041/

[therealkenc]

This circa 2016 issue is I think (?) fundamentally dupe #1609 (epoll wait doesn't). Even if the problem wasn't that issue specifically, there is a lot of select/poll water under the bridge since this issue was submitted. We'll call it fixedcreatorsupdate. If there are any outstanding issues with xemacs running on WSL that would not manifest in the same way when running remotely from Real Linux to Xming, VcXsrv, Windows Remote Desktop via xrdp (etc etc) do feel to break out a follow-up.