bug#59855: Guix Shell with FHS Feedback: lib/tls/x86_64/x86_64?

bug#59855: Guix Shell with FHS Feedback: lib/tls/x86_64/x86_64?

[Marek Paśnikowski via Bug reports for GNU Guix]

I recently started attempting to execute an AppImage with
guix shell --container --emulate-fhs .
I found a workflow, where I set LD_LIBRARY_PATH=$LIBRARY_PATH
and trace the executable with strace -ffZ ./Program.AppImage .
Its concept is that I want to turn each failing call to a successful
one, until the program runs flawlessly.
I program as a hobby, and have no way of knowing if any of the
trace calls can be safely ignored.

I would like to encourage development of such an implementation,
where strace -ffZ output is reduced as much as possible by default.
Here is the head of the current trace I am stuck at:

> access("/etc/ld.so.preload", R_OK)      = -1 ENOENT (No such file
> or directory)
>
> openat(AT_FDCWD, "/gnu/store/c8ndz685xbqm29r9zwlwsbjqj4ylh4m4-
> profile/lib/tls/x86_64/x86_64/libdl.so.2", O_RDONLY|O_CLOEXEC) = 
> -1 ENOENT (No such file or directory)

The second line is the interesting one.
According to an ancient post on the Linux Questions forum[1],
$PATH/lib/tls is the directory for glibc with thread support (NPTL).
I do not know if the glibc in Guix has the thread support or not.
However, it is the next two subdirectories that baffle me.
x86_64/x86_64?  I have never seen anything like this.
I thought, that maybe having a tls->. link in the lib directory
would suppress the failure, but lib also does not have
an x86_64 directory.

I have spent a couple of hours chasing the libdl libry,
only to realise that it is supplied by glibc, and it is
the AppImage program trying to reach it in a different place.
I strongly believe that solving this mode of failure will save
countless hours of futile struggle for future users of Guix Shell.
For the same purpose, the first error can be permanently silenced
by creating an empty file at /etc/ld.so.preload .

bug#59855: Guix Shell with FHS Feedback: lib/tls/x86_64/x86_64?

[John Kehayias via Bug reports for GNU Guix]

On Tue, Dec 06, 2022 at 09:07 AM, Marek Paśnikowski wrote:

Hello!

(resending as it seems debbugs in Emacs doesn't automatically add the
bug number email in the To/CC list.)

> I recently started attempting to execute an AppImage with
> guix shell --container --emulate-fhs .

Great! You may want to see this recent mailing list thread and draft
blog post which also covers exactly this in an example:

specifically: https://lists.gnu.org/r/guix-devel/2022-12/msg00156.html

and: https://issues.guix.gnu.org/60112

> I found a workflow, where I set LD_LIBRARY_PATH=$LIBRARY_PATH
> and trace the executable with strace -ffZ ./Program.AppImage .

In the FHS container you shouldn't need to set LD_LIBRARY_PATH. There
was a bug in the first weeks after it was merged (where this would work
around it), but assuming you have updated somewhat recently this
shouldn't be needed in genera.

> Its concept is that I want to turn each failing call to a successful
> one, until the program runs flawlessly.

I think this way lies madness. There will always be a ton of failing
calls, by nature of how things work. In particular, if you haven't see
this blog post, which covers exactly this (library loading), it was
enlightening for me at least:

https://guix.gnu.org/blog/2021/taming-the-stat-storm-with-a-loader-cache/

> I program as a hobby, and have no way of knowing if any of the
> trace calls can be safely ignored.
>
> I would like to encourage development of such an implementation,
> where strace -ffZ output is reduced as much as possible by default.

But in general I agree, I also use strace to find out what is breaking.
Often you want to see the last set of calls, but not always.

> Here is the head of the current trace I am stuck at:
>
>> access("/etc/ld.so.preload", R_OK)      = -1 ENOENT (No such file
>> or directory)
>>
>> openat(AT_FDCWD, "/gnu/store/c8ndz685xbqm29r9zwlwsbjqj4ylh4m4-
>> profile/lib/tls/x86_64/x86_64/libdl.so.2", O_RDONLY|O_CLOEXEC) =
>> -1 ENOENT (No such file or directory)
>
> The second line is the interesting one.

Yes, the first one should be ignored as far as I know and is not
important.

The second one could be a failure, if that is where it stops looking for
libdl and errors out. So a more complete log, or at least the end where
an actual failure happens, would be helpful. As well as the exact
command you are running and the AppImage (if it is free software, of
course) would be good. Usually you'll see lots of these ENOENT errors
until it finds it in one of the search paths, or else an actual error
that the library could not be found and loaded.

> According to an ancient post on the Linux Questions forum[1],

This link didn't show for me.

> $PATH/lib/tls is the directory for glibc with thread support (NPTL).
> I do not know if the glibc in Guix has the thread support or not.
> However, it is the next two subdirectories that baffle me.
> x86_64/x86_64?  I have never seen anything like this.
> I thought, that maybe having a tls->. link in the lib directory
> would suppress the failure, but lib also does not have
> an x86_64 directory.
>
> I have spent a couple of hours chasing the libdl libry,
> only to realise that it is supplied by glibc, and it is
> the AppImage program trying to reach it in a different place.
> I strongly believe that solving this mode of failure will save
> countless hours of futile struggle for future users of Guix Shell.
> For the same purpose, the first error can be permanently silenced
> by creating an empty file at /etc/ld.so.preload .

I'm not certain you are actually failing at libdl. The FHS container
will show this library in /lib and in the ld cache, which should cover
just about everything trying to find it. More details would be helpful.

Hope this helps in the meantime!
John

bug#59855: Emacs Debbugs reply (was Re: bug#59855: Guix Shell with FHS Feedback: lib/tls/x86_64/x86_64?)

[Simon Tournier]

Hi John,

On Fri, 23 Dec 2022 at 05:39, John Kehayias via Bug reports for GNU Guix <bug-guix@gnu.org> wrote:

> (resending as it seems debbugs in Emacs doesn't automatically add the
> bug number email in the To/CC list.)

Well, I think it is about Gnus and not Debbugs.  There is bunch of
option to reply with Gnus. :-)

    9 candidates:
        gnus-summary-reply
        gnus-summary-wide-reply
        gnus-summary-very-wide-reply
        gnus-summary-reply-with-original
        gnus-summary-reply-broken-reply-to
        gnus-summary-wide-reply-with-original
        gnus-summary-reply-to-list-with-original
        gnus-summary-very-wide-reply-with-original
        gnus-summary-reply-broken-reply-to-with-original


Personally, I have this config (the only Gnus keybinding I use ;-))

  (define-key gnus-summary-mode-map "R" 'gnus-summary-wide-reply-with-original)
  (define-key gnus-article-mode-map "R" 'gnus-summary-wide-reply-with-original)

which is consistent with Emacs Notmuch key binding.


Hope that helps,
simon

bug#59855: Emacs Debbugs reply (was Re: bug#59855: Guix Shell with FHS Feedback: lib/tls/x86_64/x86_64?)

[John Kehayias via Bug reports for GNU Guix]

Hi simon,

On Thu, Jan 05, 2023 at 12:31 PM, Simon Tournier wrote:

> Hi John,
>
> On Fri, 23 Dec 2022 at 05:39, John Kehayias via Bug reports for GNU
> Guix <bug-guix@gnu.org> wrote:
>
>> (resending as it seems debbugs in Emacs doesn't automatically add
>> the
>> bug number email in the To/CC list.)
>
> Well, I think it is about Gnus and not Debbugs.  There is bunch of
> option to reply with Gnus. :-)
>
>     9 candidates:
>         gnus-summary-reply
>         gnus-summary-wide-reply
>         gnus-summary-very-wide-reply
>         gnus-summary-reply-with-original
>         gnus-summary-reply-broken-reply-to
>         gnus-summary-wide-reply-with-original
>         gnus-summary-reply-to-list-with-original
>         gnus-summary-very-wide-reply-with-original
>         gnus-summary-reply-broken-reply-to-with-original
>
>
> Personally, I have this config (the only Gnus keybinding I use ;-))
>
>   (define-key gnus-summary-mode-map "R" 'gnus-summary-wide-reply-with-original)
>   (define-key gnus-article-mode-map "R" 'gnus-summary-wide-reply-with-original)
>
> which is consistent with Emacs Notmuch key binding.
>
>
> Hope that helps,
> simon

Yes that helps, exactly what I would have expected, thanks! It is now also my only Gnus keybindings, matching my others (mu4e in my case).

Much appreciated,
John