[bug#62848] [PATCH] environment: Add --remote option and emacsclient-eshell backend.

[Maxim Cournoyer]

Hi Antero,

Antero Mejr <antero@mailbox.org> writes:

> * guix/scripts/environment.scm (launch-environment/eshell): New procedure.
> (guix-environment*): Add logic for remote backend switching.
> (%options): Add --remote and --list-remote-backends options.
> (show-environment-options-help): Add help text for new options.
> * guix/profiles.scm (load-profile)[getenv-proc, setenv-proc, unsetenv-proc]:
> New optional keyword arguments.
> (purify-environment)[unsetenv-proc]: New argument.
> * doc/guix.texi(Invoking guix shell): Document --remote and
> --list-remote-backends options.

This looks interesting!

[...]

> +(define* (launch-environment/eshell args command profile manifest
> +                                    #:key pure? (white-list '()))

Maybe use the modern allow-list / block-list terminology.

> +  "Create an new eshell buffer with an environment containing PROFILE,
> +with the search paths specified by MANIFEST.  When PURE?, pre-existing
> +environment variables are cleared before setting the new ones, except those
> +matching the regexps in WHITE-LIST."
> +
> +  (define (escape in)
> +    (string-append "\"" (string-replace-substring in "\"" "\\\"") "\""))
> +
> +  (define* (ec cmd #:optional (check? #f))

                                 ^
Shouldn't we always check for errors?  When is it useful to let them
through?

> +    (let* ((cmd (string-append "emacsclient " args " -e " (escape cmd)))

Instead of having to escape stuff, it'd be better to avoid using a shell
for the invocation by opening the pipe with (open-pipe* OPEN_READ prog
arg ...).  There's an example if you grep for
'with-input-pipe-to-string' in the update-guix-package.scm file.  This
should make the rest of the code more readable.

> +           (port (open-input-pipe cmd))
> +           (str (read-line port))
> +           (code (status:exit-val (close-pipe port))))
> +      (if (and check? (or (not (eqv? code 0)) (eof-object? str)))
                             ^
                             (zero? code)

> +          (leave
> +           (G_ "Emacs server connection failed. Is the server running?~%")))
> +      str))
> +
> +  (let ((buf (ec "(buffer-name (eshell t))" #t)))
> +    (define (ec-buf cmd)
> +      (ec (string-append "(with-current-buffer " buf " " cmd ")")))
> +
> +    (load-profile
> +     profile manifest #:pure? pure? #:white-list-regexps white-list
> +     #:setenv-proc (lambda (var val)
> +                     (ec-buf
> +                      (if (string=? var "PATH")
> +                          ;; TODO: TRAMP support?
> +                          (string-append "(eshell-set-path " (escape val) 
> ")")
> +                          (string-append "(setenv " (escape var) " "
> +                                         (escape val) ")"))))
> +     #:unsetenv-proc (lambda (var)
> +                       (ec-buf
> +                        (string-append "(setenv " (escape var) ")"))))
> +
> +    (match command
> +      ((program . args)
> +       (ec-buf
> +        (string-append
> +         "(eshell-command "
> +         (escape (string-append program " " (string-join args)))")"))))))
> +
>  (define* (launch-environment/container #:key command bash user user-mappings
>                                         profile manifest link-profile? 
> network?
>                                         map-cwd? emulate-fhs? nesting?
> @@ -1081,7 +1140,10 @@ (define (guix-environment* opts)
>                             '("/bin/sh")
>                             (list %default-shell))))
>           (mappings   (pick-all opts 'file-system-mapping))
> -         (white-list (pick-all opts 'inherit-regexp)))
> +         (white-list (pick-all opts 'inherit-regexp))
> +         (remote (match (string-split (assoc-ref opts 'remote) #\=)
> +                   ((x) (cons x ""))

Why is this needed?

> +                   ((x . y) (cons x (string-join y))))))

If using open-pipe* as suggested above, all the arguments could be kept
as a list.

>
>      (define store-needed?
>        ;; Whether connecting to the daemon is needed.
> @@ -1119,6 +1181,10 @@ (define-syntax-rule (with-store/maybe store exp ...)
>        (when (pair? symlinks)
>          (leave (G_ "'--symlink' cannot be used without '--container~%'"))))
>
> +    (when (and remote (not (member (car remote) '("emacsclient-eshell"))))
> +      (leave
> +       (G_ "Invalid remote backend, see --list-remote-backends for 
> options.~%'")))

This code and the --list-remote-backends associated procedure could
share a same %supported-backends list containing currently just
'("emacsclient-eshell").

>      (with-store/maybe store
>        (with-status-verbosity (assoc-ref opts 'verbosity)
>          (define manifest-from-opts
> @@ -1172,15 +1238,23 @@ (define manifest
>
>                  (mwhen (assoc-ref opts 'check?)
>                    (return
> -                   (if container?
> +                   (if (or container? remote)
>                         (warning (G_ "'--check' is unnecessary \
> -when using '--container'; doing nothing~%"))
> +when using '--container' or '--remote'; doing nothing~%"))
>                         (validate-child-shell-environment profile manifest))))
>
>                  (cond
>                   ((assoc-ref opts 'search-paths)
>                    (show-search-paths profile manifest #:pure? pure?)
>                    (return #t))
> +                 (remote
> +                  (match (car remote)
> +                    ("emacsclient-eshell"
> +                     (return
> +                      (launch-environment/eshell (cdr remote)
> +                                                 command profile manifest
> +                                                 #:white-list white-list
> +                                                 #:pure? pure?)))))
>                   (container?
>                    (let ((bash-binary
>                           (if bootstrap?

It'd be nice to have a functional test for it; some inspiration could be
taken from tests/build-emacs-utils.scm, which skips the test if emacs is
not available.

Thanks for this contribution!  Could you please send a v2 taking into
account the above comments?

-- 
Thanks,
Maxim