[bug#57345] Add gnupg to propagated-inputs of trezor-agent

This is my first patch submission ever.
Please tell me if there is anything I did wrong.

The patch is motivated by failure of `trezor-gpg init` command.
After I applied it to the system, I was able to initialize my Trezor-based GPG identity.

Hi Marek,

thanks for your patch.  This looks like a good fix of the problem.

Propagation is generally used sparingly, when there is no other option.
We try to avoid it when possible, because it can lead to unexpected
conflicts, e.g. when a user has a particular version of gnupg installed
in their profile and then installs trezor-agent with a more recent
version of Guix it would abort with a conflict error as it cannot
install two variants of gnupg into the same Guix profile.

I wonder if we can avoid propagation here.  We could, for example, look
up gnupg in PATH (so that a user’s installation of gnupg is preferred)
and fall back to an explicit gnupg from the inputs (not
propagated-inputs).  This would require minor patching of trezor-agent.

What do you think?

Nothing I do seems to work at all.

Last night I thought of a fundamental question:
Are `inputs` accessible in the guix store at runtime?

If they are, that means I am too dumb to correctly parse the documentation. My initial patch proves, that `gnupg` has to be available at runtime. At this point I am incapable of doing anything more. I got physically sick of trying to understand and use all the procedures.

On Mon, Sep 05, 2022 at 05:49:55AM +0000, guix-patches--- via wrote:
> Nothing I do seems to work at all.
> 
> Last night I thought of a fundamental question:
> Are `inputs` accessible in the guix store at runtime?
> 
> If they are, that means I am too dumb to correctly parse the documentation. My initial patch proves, that `gnupg` has to be available at runtime. At this point I am incapable of doing anything more. I got physically sick of trying to understand and use all the procedures.

After some poking around on IRC and in trezor-agent and
python-trezor-agent, it seems that there are a number of calls to the
gpgconf binary and the actual path to gpg is probably set in
python-trezor-agent in libagent/gpg/keyring.py. With all of this it's
probably best to just wrap the files in the bin output of trezor agent
so that we don't miss any files.

(ins)efraim@3900XT /tmp$ tree /gnu/store/46br8illcfv93ryh28s2haz1s59n584v-trezor-agent-0.14.4
/gnu/store/46br8illcfv93ryh28s2haz1s59n584v-trezor-agent-0.14.4
├── bin
│   ├── __pycache__
│   │   └── trezor_agent.cpython-39.pyc
│   ├── trezor-agent
│   ├── trezor_agent.py
│   ├── trezor-gpg
│   ├── trezor-gpg-agent
│   └── trezor-signify
└── lib
    └── python3.9
        └── site-packages
            └── trezor_agent-0.11.0-py3.9.egg-info
                ├── dependency_links.txt
                ├── entry_points.txt
                ├── PKG-INFO
                ├── requires.txt
                ├── SOURCES.txt
                └── top_level.txt

trezor_agent.py is an actual python script, so I'd use wrap-script, but
the other extensionless files I'd use wrap-program, which IIRC will Do
The Right Thing™ and not double-wrap those files, which are already
wrappers.

I haven't tested the code out, but something like this should work
inside trezor-agent.

(add-before 'check 'wrap-more
  ;; We want to make sure we're before 'check but after 'wrap
  (lambda* (#:key inputs outputs #:allow-other-keys)
    (wrap-script (string-append #$output "/bin/trezor_agent.py")
      `(("PATH" ":" prefix (,(dirname (search-input-file inputs
      "gpg"))))))
    (for-each
      (lambda (file)
        (wrap-program file
          `(("PATH" ":" prefix (,(dirname (search-input-file inputs
          "gpg")))))))
      (find-files (string-append #$output "/bin") "^trezor-"))))

I have encountered the same problem as in this email:
https://lists.gnu.org/archive/html/help-guix/2019-06/msg00073.html

I do not understand how to apply the proposed fix.

Let us not have perfect be the enemy of good, and merge the `propagated-inputs` solution. Once an `inputs` solution is found, it would be used instead.

A potential conflict with other `gnupg` instances is less harmful than the currently happening guaranteed loss of major functionality.

What do you think?

On 05-09-2022 13:26, guix-patches--- via wrote:
> I do not understand how to apply the proposed fix.

Copy-paste Efraim's phase to the %standard-phases (I assume before 
fixup-agent-py).

Greetings,
Maxime

On 06-09-2022 15:06, guix-patches--- via wrote:
> Let us not have perfect be the enemy of good, and merge the `propagated-inputs` solution. Once an `inputs` solution is found, it would be used instead.
>
> A potential conflict with other `gnupg` instances is less harmful than the currently happening guaranteed loss of major functionality.
>
> What do you think?

Efraim has posted a potential 'inputs' solution.

Greetings,
Maxime.

On 05-09-2022 13:26, guix-patches--- via wrote:
> I have encountered the same problem as in this email:
> https://lists.gnu.org/archive/html/help-guix/2019-06/msg00073.html
>
> I do not understand how to apply the proposed fix.
>
IIRC, python-build-system uses G-exps nowadays, so to use #$, all you 
need to do is replace

(arguments `(#:phases (modify-phases ...)))

by

(arguments (list #:phases #~(modify-phases ...))),

no need for separate inputs.

On 05-09-2022 07:49, guix-patches--- via wrote:
> Nothing I do seems to work at all.
>
> Last night I thought of a fundamental question:
> Are `inputs` accessible in the guix store at runtime?
If a reference (i.e., a /gnu/store/the-input-... string) ends up in the 
resulting store item of trezor-agent, then yes.
> If they are, that means I am too dumb to correctly parse the documentation.
As far as I'm aware it is not documented that the GC is based on 
reference-scanning.
> My initial patch proves, that `gnupg` has to be available at runtime. At this point I am incapable of doing anything more. I got physically sick of trying to understand and use all the procedures.

Look where 'gnupg' is used in trezor-agent (git grep -F can be useful) 
and replace it with substitute* + search-input-file. Alternatively, 
wrap-program, as proposed in Efraim's reply.

Greetings,
Maxime.

I give up.

So far, the cleanest workaround is to explicitly declare `gnupg` installation.
I decided to use that going forward.

```
(inputs (list gnupg guile-2.0 python-3 python-trezor python-trezor-agent))

(add-before 'check 'wrap-more
  (lambda* (#:key inputs outputs #:allow-other-keys)
    (wrap-script (string-append #$output "/bin/trezor_agent.py")
      '("PATH" ":" prefix '("/gnu/store/3bpq5knfvzhxhqfwzqm9br917nz7r0yp-gnupg-2.2.32/bin")))))
```

This is as far as I was able to go while maintaining any sense of understanding and control over the code.
All attempts to progress end with this failure:

```
Backtrace [...]

ERROR:
  1. &wrap-error:
      program: "/gnu/store/rb3wzd5pi899mlvv4wj35afc4z8ys60m-trezor-agent-0.14.4/bin/trezor_agent.py"
      type: no-interpreter-found
```

Clearly, the wrapping documentation assumes something and fails to mention it - and now a programming noob like me is totally lost.

My conclusions:
1. Current state of `trezor-agent` - broken critical functionality by default.
2. Presence of `gnupg` in profile fixes the above.
3. `propagated-inputs` also fixes 1., but is not acceptable.
4. In theory, it should be possible to expose a PATH to `gnupg` in `inputs`; but after days of working towards it, I am still not capable of reaching this solution.

the problem is that there's no #! shebang line at the head of trezor_agent.py

but i'm not sure that file needs to be patched. the user-facing entry point is trezor-agent, not the .py file.