Emacs loops/segfaults with certain buffer contents in certain major modes

Question

Emacs loops/segfaults with certain buffer contents in certain major modes

nathanvy opened this issue a year ago · comments

Hi all,

edit: Original title was "Prism mode causes emacs to freeze/lock up when typing double-quote characters on ARM macbooks"

Title says it all. I am not sure how to debug this further, because emacs hangs entirely and the only way to recover is to force-quit the application. I suppose one could use gdb or another debugger but I am not an expert in that sort of debugging and wouldn't know where to start.

Steps to reproduce:

start emacs with blank config on M2 Macbook pro
Install Prism from MELPA (version "20230416.626" is what package.el pulled down for me)

(use-package prism
    :commands prism-mode
    :init
    (add-hook 'prog-mode-hook #'prism-mode))

Open a .lisp file and attempt to type the following: (#P"") which is a Common Lisp pathname literal. I have also seen this when editing HTML and attempting to type <html lang="en">

Adam Porter · Answer 1 · Sat May 13 2023 09:42:24 GMT+0800 (China Standard Time)

Hi Nathan,

Thanks for reporting this. I can confirm that there is a problem. It's quite strange, and I'm surprised it took this long to be found. Yet, it seems to only happen under certain circumstances, ones which I haven't encountered in my own usage, despite using prism every day.

Here's what I did to reproduce:

Using https://github.com/alphapapa/with-emacs.sh: Run with-emacs.sh -e emacs-28.2 -i prism.
C-x C-f /tmp/test.lisp RET
M-x prism-mode RET
Type (#P"").

After the first quotation mark (actually, after I press the key, but before it is displayed) Emacs goes into a 100%-CPU usage loop. If I press C-g, Emacs will show the quotation mark and go into the loop again. If I keep pressing C-g, I can get Emacs to respond to another keypress each time. But as long as prism-mode is active in the buffer, Emacs will go into the loop trying to fontify that text. (Another way to reproduce is to write that string into the buffer and then activate prism-mode.)

If, while Emacs is looping, I send the SIGUSR2 signal, which should cause Emacs to quit and show a backtrace, instead Emacs segfaults:

Fatal error 11: Segmentation fault
Backtrace:
/home/me/.guix-profile/bin/emacs-28.2[0x529b7e]
/home/me/.guix-profile/bin/emacs-28.2[0x4257ca]
/home/me/.guix-profile/bin/emacs-28.2[0x425c62]
/home/me/.guix-profile/bin/emacs-28.2[0x528208]
/home/me/.guix-profile/bin/emacs-28.2[0x528279]
/gnu/store/5h2w4qi9hk1qzzgi1w83220ydslinr4s-glibc-2.33/lib/libpthread.so.0(+0x11d80)[0x7fac20ab1d80]
/home/me/.guix-profile/bin/emacs-28.2[0x4d6f9c]
/home/me/.guix-profile/bin/emacs-28.2[0x4d6fc3]
/home/me/.guix-profile/bin/emacs-28.2[0x4d6fc3]
/home/me/.guix-profile/bin/emacs-28.2[0x4d6fc3]
/home/me/.guix-profile/bin/emacs-28.2[0x4d6fc3]
/home/me/.guix-profile/bin/emacs-28.2[0x4d6fc3]
/home/me/.guix-profile/bin/emacs-28.2[0x4d6fc3]
/home/me/.guix-profile/bin/emacs-28.2[0x4d6fc3]
/home/me/.guix-profile/bin/emacs-28.2[0x4d6fc3]
/home/me/.guix-profile/bin/emacs-28.2[0x4d6fc3]
/home/me/.guix-profile/bin/emacs-28.2[0x4d6fc3]
/home/me/.guix-profile/bin/emacs-28.2[0x4d6fc3]
/home/me/.guix-profile/bin/emacs-28.2[0x4d6fc3]
/home/me/.guix-profile/bin/emacs-28.2[0x4d6fc3]
/home/me/.guix-profile/bin/emacs-28.2[0x4d6fc3]
/home/me/.guix-profile/bin/emacs-28.2[0x4d6fc3]
/home/me/.guix-profile/bin/emacs-28.2[0x4d6fc3]
/home/me/.guix-profile/bin/emacs-28.2[0x4d6fc3]
/home/me/.guix-profile/bin/emacs-28.2[0x4d6fc3]
/home/me/.guix-profile/bin/emacs-28.2[0x4d6fc3]
/home/me/.guix-profile/bin/emacs-28.2[0x4d6fc3]
/home/me/.guix-profile/bin/emacs-28.2[0x4d6fc3]
/home/me/.guix-profile/bin/emacs-28.2[0x4d6fc3]
/home/me/.guix-profile/bin/emacs-28.2[0x4d6fc3]
/home/me/.guix-profile/bin/emacs-28.2[0x4d6fc3]
/home/me/.guix-profile/bin/emacs-28.2[0x4d6fc3]
/home/me/.guix-profile/bin/emacs-28.2[0x4d6fc3]
/home/me/.guix-profile/bin/emacs-28.2[0x4d6fc3]
/home/me/.guix-profile/bin/emacs-28.2[0x4d6fc3]
/home/me/.guix-profile/bin/emacs-28.2[0x4d6fc3]
/home/me/.guix-profile/bin/emacs-28.2[0x4d6fc3]
/home/me/.guix-profile/bin/emacs-28.2[0x4d6fc3]
/home/me/.guix-profile/bin/emacs-28.2[0x4d6fc3]
/home/me/.guix-profile/bin/emacs-28.2[0x4d6fc3]
/home/me/.guix-profile/bin/emacs-28.2[0x4d6fc3]
...
/home/me/.bin/with-emacs: line 215: 130799 Segmentation fault      "$emacs" "${emacs_args[@]}"

I can confirm that the same thing happens in html-mode with the sample text you provided. Note that prism is not intended for use with HTML (see #16). But, of course, Emacs should not freeze or crash with it activated. However, prism-whitespace-mode does not cause a problem with that HTML fragment.

And, of course, typing the (#P"") text in emacs-lisp-mode (which is what I mostly use prism-mode for) causes no problems.

I tried using font-lock-studio to reveal the problem, but Emacs froze and crashed with it also.

So there may be a bug in Emacs here (because AFAIK Elisp should never be able to cause Emacs to segfault), but maybe also a bug in prism that's triggering it. I'm guessing it's related to the syntax tables used by the major modes, since it happens in lisp-mode but not in emacs-lisp-mode.

And, of course, I'm not on an M2 Mac, so that appears to be unrelated to the problem.

For reference, what version of Emacs are you using, and where did the build come from?

As you said, I suppose one would need to use gdb to debug this. I'm also not an expert with it, but I do recall that, if one makes an Emacs bug report, the developers will advise the reporter how to use gdb to get the necessary information. Would you be willing to pursue that? If not, I'll try to get to it eventually, but since I don't experience this problem myself, I probably won't have time to dig deeper anytime soon.

nathanvy · Answer 2 · Sat May 13 2023 15:06:35 GMT+0800 (China Standard Time)

Hi Nathan,

Thanks for reporting this. I can confirm that there is a problem. It's quite strange, and I'm surprised it took this long to be found. Yet, it seems to only happen under certain circumstances, ones which I haven't encountered in my own usage, despite using prism every day.

You're most welcome, and thank you in return for Prism.

I can confirm that the same thing happens in html-mode with the sample text you provided. Note that prism is not intended for use with HTML (see #16). But, of course, Emacs should not freeze or crash with it activated. However, prism-whitespace-mode does not cause a problem with that HTML fragment.

Thanks, I wasn't aware of prism-whitespace-mode!

For reference, what version of Emacs are you using, and where did the build come from?

I've reproduced this on the emacs-mac fork which is currently on 28.2, as well as the 28.2 build from emacsformacosx.com which purports to be a straight port with minimal/no changes.

As you said, I suppose one would need to use gdb to debug this. I'm also not an expert with it, but I do recall that, if one makes an Emacs bug report, the developers will advise the reporter how to use gdb to get the necessary information. Would you be willing to pursue that? If not, I'll try to get to it eventually, but since I don't experience this problem myself, I probably won't have time to dig deeper anytime soon.

Sure, I'll pursue that this week and report back here with any findings.

Adam Porter · Answer 3 · Sun May 14 2023 12:08:37 GMT+0800 (China Standard Time)

Thank you!

Adam Porter · Answer 4 · Thu May 18 2023 20:31:06 GMT+0800 (China Standard Time)

Hi Nathan,

I hope I have fixed it. Please try this commit: ba53105 It seems to work for me with the following test code in Elisp mode, and your test code in Lisp mode, and the HTML snippet:

(P"")
(P "")
(P P)
(P"foo bar")
(P "foo bar")
(P "foo")
(P """""")

Please test it and let me know if it works for you. (Obviously, be sure to save any work in progress first, or use a separate Emacs session, in case it does still cause Emacs to crash.)

Thanks.

nathanvy · Answer 5 · Fri May 19 2023 06:09:14 GMT+0800 (China Standard Time)

Thanks for this. I'll take a look as soon as able, life's been very hectic lately so I haven't yet been able to put together an emacs bug report to upstream.

Adam Porter · Answer 6 · Fri May 19 2023 14:22:09 GMT+0800 (China Standard Time)

Nevermind, I found that that change causes another bug. I'll continue working on it.

Adam Porter · Answer 7 · Fri May 19 2023 14:42:31 GMT+0800 (China Standard Time)

Ok, this commit seems to fix the problems noted here without causing any new problems: c2d52ed Please test it carefully and let me know what you find. Thanks.

Adam Porter · Answer 8 · Fri May 19 2023 18:10:53 GMT+0800 (China Standard Time)

It seems to be working well, so I merged it to master and released v0.3 with this fix. @nathanvy Please do let me know how it works for you. If there are any more problems, I'll release a fix in v0.3.1.

nathanvy · Answer 9 · Fri May 19 2023 23:56:38 GMT+0800 (China Standard Time)

Thanks again. I noticed in your commit you mentioned prism isn't intended for HTML. Should I be using prism-whitespace-mode instead for that?

Adam Porter · Answer 10 · Sat May 20 2023 00:14:06 GMT+0800 (China Standard Time)

You can try it, but it's not intended for HTML either. See #16

nathanvy · Answer 11 · Sat May 20 2023 00:15:56 GMT+0800 (China Standard Time)

Okay thanks. I basically have been using prism for literally anything, haha.

Adam Porter · Answer 12 · Sat May 20 2023 00:18:39 GMT+0800 (China Standard Time)

I'm glad whereever it works. :) As I described in #16, the Emacs functions for SGML-type languages aren't quite sufficient to do what prism is intended to do, but Emacs 29 will probably make that easier.

Adam Porter · Answer 13 · Sat May 20 2023 00:19:03 GMT+0800 (China Standard Time)

@nathanvy BTW, did you confirm that this issue is fixed by v0.3?

nathanvy · Answer 14 · Sat May 20 2023 00:26:57 GMT+0800 (China Standard Time)

I've just tested now, while Peppa Pig was parenting my kids.

Edit: Just so we're on the same page, I upgraded to latest using straight-pull-all rather than pull down what's on MELPA. I navigate into ~/.emacs.d/straight/repos/prism.el, then git log --format="%H" -n 1 returns the commit 8d4b7726632eac8850b2509239ad0283a80afc93 which appears to be correct.

In any case, this no longer freezes emacs:

(#P"")

But this still does, immediately upon typing the doublequote.

<html lang="

I'm willing to just cease using Prism for HTML and friends, as a long-term workaround, as it seems the lisp side of things is resolved.

Adam Porter · Answer 15 · Sat May 20 2023 01:15:53 GMT+0800 (China Standard Time)

@nathanvy Thanks for finding that. I seem to have fixed it here: b1513e1 Please let me know if that helps. (It's the tip of the wip/18 branch, which you can install with Straight.)

nathanvy · Answer 16 · Sat May 20 2023 01:38:32 GMT+0800 (China Standard Time)

Thanks I'll test it this evening

nathanvy · Answer 17 · Sun May 21 2023 15:26:12 GMT+0800 (China Standard Time)

@nathanvy Thanks for finding that. I seem to have fixed it here: b1513e1 Please let me know if that helps. (It's the tip of the wip/18 branch, which you can install with Straight.)

Sorry for the delay; that change resolves the issue on my end, no longer able to reproduce!

Jason Zhen · Answer 18 · Tue May 23 2023 08:28:10 GMT+0800 (China Standard Time)

Hello I've also experienced this issue, I tried using the wip/18 branch but it still hangs here. I've was able to narrow it down to this minimum example to reproduce.

// I'm a comment
foo = bar

The issue seems to be apostrophes in comments, I've disabled prism-comments for now as a workaround. Just FYI to anyone else stumbling across this

fkr-0 · Answer 19 · Tue May 23 2023 22:35:58 GMT+0800 (China Standard Time)

[67181.425287] emacs[155037]: segfault at 560f914e5840 ip 00007f8abd68121a sp 00007ffdac05a7f0 error 7 in emacs-zmq.so[7f8abd600000+1a8000] likely on CPU 1 (core 0, socket 0)
[67181.425306] Code: 48 8b 03 48 8d 35 53 dd 0d 00 48 8b 68 10 48 89 c7 ff 50 58 48 8b 3b 48 89 c6 ff d5 48 8b 15 dd 2a 33 00 48 8d 35 db 53 0d 00 <48> 89 02 48 8b 03 48 8b 68 10 48 89 c7 ff 50 58 48 8b 3b 48 89 c6

just wanted to note that I also experienced this issue and can confirm that in my case
a) there were apostrophes in comments
b) turning off prism-comments works

Adam Porter · Answer 20 · Tue May 23 2023 23:30:27 GMT+0800 (China Standard Time)

@JasZhe Thanks for reporting that.

@fkr-0 Thanks. It's puzzling why Emacs is segfaulting, but when this bug is fixed, that shouldn't happen.

Adam Porter · Answer 21 · Tue May 23 2023 23:35:00 GMT+0800 (China Standard Time)

@JasZhe @fkr-0 I just pushed another commit to the WIP branch:

https://github.com/alphapapa/prism.el/tree/wip/18

It seems to fix that problem. Please test it and let me know if it does for you. Thanks.

Adam Porter · Answer 22 · Tue May 23 2023 23:48:17 GMT+0800 (China Standard Time)

I went ahead and pushed v0.3.1 that seems to fix this problem. Please let me know your experience. Thanks.

Simon Pugnet · Answer 23 · Tue May 23 2023 23:59:00 GMT+0800 (China Standard Time)

I was having this same issue this morning when trying to enable prism-mode in a buffer containing my Emacs init.el and this update has fixed it, thanks!

Adam Porter · Answer 24 · Wed May 24 2023 01:38:07 GMT+0800 (China Standard Time)

@polaris64 Thanks for letting me know.

soerlemans · Answer 25 · Wed May 24 2023 01:52:59 GMT+0800 (China Standard Time)

Just updated as well.
It stopped hanging 👍.

Jason Zhen · Answer 26 · Wed May 24 2023 02:51:59 GMT+0800 (China Standard Time)

Updated to the latest master, re-enabled comments, it no longer freezes emacs, but I do get the infinite loop error message:

Error during redisplay: (jit-lock-function 1) signaled (error "prism: Infinite loop detected in ‘prism-match’ (buffer:#<buffer test_prism.go> point:5).  Please report this bug")

using the same example from before in test_prism.go:

// I'm a comment
foo = bar

Thank you for the quick turnaround on this!

Adam Porter · Answer 27 · Wed May 24 2023 16:38:34 GMT+0800 (China Standard Time)

@JasZhe Hm, thanks, I'll look at that.

Adam Porter · Answer 28 · Wed May 24 2023 19:36:30 GMT+0800 (China Standard Time)

@JasZhe I just pushed v0.3.2 which should fix that Go example. Please let me know if it works for you.

And to all who have chimed in on this issue: Thanks. I didn't know that this package was being used for so many different major modes and languages. And I didn't know that some of those major modes don't work the same way internally with regard to syntax tables, so I've had to make some minor adjustments to compensate for that.

In the future, Emacs 29 will likely make this kind of code simpler to write by using tree-sitter. But for now, it's complicated.

Jason Zhen · Answer 29 · Thu May 25 2023 05:56:35 GMT+0800 (China Standard Time)

🎉 working on my end now thanks!

Adam Porter · Answer 30 · Thu May 25 2023 05:59:53 GMT+0800 (China Standard Time)

@JasZhe Great, thank you!

I didn't think many people were using prism-comments, but I'm glad to see it "in the wild." :)

Jason Zhen · Answer 31 · Thu May 25 2023 06:17:32 GMT+0800 (China Standard Time)

I'm a simple man, I just like seeing more pretty colors in my code :)

Adam Porter · Answer 32 · Thu May 25 2023 06:19:39 GMT+0800 (China Standard Time)

We've gotta keep the fun in computing!

BTW, feel free to suggest any screenshots you think might look good in the readme: #19 Your theme seems to fit prism well.