Extra nulls in path string with 28.1 Linux package

[asmeurer]

With the emacs 28.1 package, I get this error when I start emacs

Warning (initialization): An error occurred while loading ‘/home/aaronmeurer/.emacs’:

Wrong type argument: filenamep, /home/aaronmeurer/anaconda3/share/info\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0/

This is coming from one of the packages in my init file, but the extra nulls indicate that this is likely an issue with the conda build. Somewhere, the extra null padding that is done for the binary path replacement messed up and created a string that emacs itself is interpreting as including nulls.

I didn't see this issue on macOS, only on Linux.

[allispaul]

I've been having the same issue. Has anyone found a solution?

[izahn]

If someone can give explicit reproduction steps for demonstrating this bug I'll look into it

[asmeurer]

Let me see if I can bisect my init file.

[asmeurer]

Hmm, now when I try to start emacs in a fresh environment, I just get

emacs: symbol lookup error: emacs: undefined symbol: malloc_set_state, version GLIBC_2.2.5

[izahn]

Ouch. We need to add tests that run the GUI. I've just recently learned how to do that:

• add xorg-x11-server-Xvfb to yum_requirements.txt as in https://github.com/conda-forge/rstudio-feedstock/blob/main/recipe/yum_requirements.txt
• use xvfb-run to test the GUI as in https://github.com/conda-forge/rstudio-feedstock/blob/47a828f47a12f14713429a28a7d022fa9d6b7fd9/recipe/meta.yaml#L151

[izahn]

Although in this case the error occurs with emacs -nw too, so that's not the problem.

Hmm, now when I try to start emacs in a fresh environment, I just get

emacs: symbol lookup error: emacs: undefined symbol: malloc_set_state, version GLIBC_2.2.5

This happens because malloc_set_state has been removed from glibc:

• The deprecated functions malloc_get_state and malloc_set_state have been
  moved from the core C library into libc_malloc_debug.so. Legacy applications
  that still use these functions will now need to preload libc_malloc_debug.so
  in their environment using the LD_PRELOAD environment variable.

So actually this is not because of a change in conda-forge packaging but because of a change in the host operating system. You can work-around with

export LD_PRELOAD=/usr/lib/libc_malloc_debug.so

(assuming you have /usr/lib/libc_malloc_debug.so on the host OS, adjust as needed). Can you do that and then continue trying to pin down the embedded nulls issue?

I think this is probably unrelated to the extra nulls issue here, so let's open a new issue for the the glibc problem.

[PythonNut]

The extra nulls first appear in configure-info-directory as far as I can tell. If you look at it with describe-variable you can see the nulls even in emacs -Q.

The error actually gets triggered in info.el (part of Emacs) which is probably loaded by many configs at some point or other. One temporary workaround to this problem is to add these two lines to the top of your config:

(setq configure-info-directory (string-replace "\0" "" configure-info-directory))
(setf (car Info-default-directory-list) configure-info-directory)

[izahn]

The extra nulls first appear in configure-info-directory as far as I can tell. If you look at it with describe-variable you can see the nulls even in emacs -Q.

The error actually gets triggered in info.el (part of Emacs) which is probably loaded by many configs at some point or other.

Thanks, I'll look into this when I get back from vacation next week.

[mhlr]

A temporary fix is
(setq Info-default-directory-list '("~/miniconda3/share/info/" "/usr/share/info/"))
in early-init.el

[zklaus]

Excellent find about the early-init.el, @mhlr! I didn't know about that. The line you posted is specific to your installation. A more general fix might be

(setq configure-info-directory (string-replace "\0" "" configure-info-directory))
(setf (car Info-default-directory-list) configure-info-directory)

I also did some more digging. In src/callproc.c, line 2065 we have Vconfigure_info_directory = build_string (PATH_INFO);, which is defined in src/lisp.h as

INLINE Lisp_Object
build_string (const char *str)
{
  return make_string (str, strlen (str));
}

so that's where things start to go wrong because make_string (src/alloc.c) proceeds to make a lisp string that is not null terminated but records the length in a struct field.

So perhaps the best way forward is to make patch the temporary fix mentioned above into a distribution file?

[izahn]

I think #64 should do it. I built it locally and uploaded to the izahn channel, you can install from there and test locally if you like.

[izahn]

There are currently two PRs being worked on to address this:

1. fix default info path #64 uses site-start.el to fix Info-default-directory-list after the fact
2. Add configure-info-dir patch #65 patches emacs to prevent the problem in the first place.

A third option suggested by https://www.gnu.org/software/emacs/manual/html_node/info/Emacs-Info-Variables.html is to set the INFOPATH environment variable, which should prevent Info-default-directory-list from being used at all.

[diekhans]

This still occurs on Emacs 29.2; any progress in finishing the PRs?

[asmeurer]

It seems the latest status on #65 was that it doesn't work. I don't know if anyone has looked at it recently.