[PATCH v2] Re: Bug in 9.5.3 org--file-default-apps

[Ihor Radchenko]

Max Nikulin <manikulin@gmail.com> writes:

> On 22/05/2022 11:10, Ihor Radchenko wrote:
>> Max Nikulin writes:
>> 
>>> The source of the problem is that Emacs-27 was released with the
>>> following bug:
>>>
>>> https://debbugs.gnu.org/cgi/bugreport.cgi?bug=40247
>>> mailcap-mime-data erased when parsing mime parts
>>>
>>> So `mailcap-parse-mailcaps' called from `mailcap-mime-info' erases
>>> predefined associations in Emacs-27.
>> 
>> If I understand correctly, this extra complication does not affect most
>> of the systems. I am not sure if we need to work around it.
>
> I would say that view-mode is quite reasonable default to open a 
> "text/plain" file and this bug broke it.

I agree. However, I am already a bit lost with all the complications.
This particular one does not appear to be directly relevant to the
initial problem with a file with no extension. I'd prefer if this
Emacs-27 issue were reported in a separate thread.

>> Also, I am attaching a patch to address the original issue. We can just
>> use file command when available. WDYT?
>
> Ihor, have you manged to reproduce the original issue? Are links with 
> explicit .txt suffix [[file:file.txt]] affected by the same problem? My 
> environments sometimes behave in a way unexpected to you and I have not 
> setup any tool to quickly launch transient virtual machines with no fear 
> to "broke" current state, so I have not tried to debug the reported 
> issue in its original form.
>
> I may be excessively suspicious.

Yes, I have managed to reproduce the original issue. Kind of.

We have the following in org-open-file:

(when (eq cmd 'mailcap)
      (require 'mailcap)
      (mailcap-parse-mailcaps)
      (let* ((mime-type (mailcap-extension-to-mime (or ext "")))
             (command (mailcap-mime-info mime-type)))
        (if (stringp command)
            (setq cmd command)
          (setq cmd 'emacs))))

with EXT being bound to file extension.
When file does not have an extension (the original issue),
(mailcap-extension-to-mime) returns nil.
Then, (mailcap-mime-info nil) returns the same result with
(mailcap-mime-info "text/plain"). It is hard-coded inside
mailcap-mime-info.

So, with the current default value of org-file-apps-gnu, files with no
extension always use mime handler assigned to "text/plain" mimetype.
In a way, it is not wrong - mailcap spec only considers file extension
and has undefined behavior for files with no extension.
However, it does not make sense from the user perspective. Hence, I am
suggesting this patch. =file= command (when available) is more powerful
and looks at the file contents, not just into its extension.

>> +      (let* ((mime-type (if (executable-find "file")
>> +                            (shell-command-to-string
>> +                             (format "%s --brief --mime-type %s"
>> +                                     (executable-find "file")
>> +                                     file))
>
> I hate elisp API related to executing of external processes because it 
> encourages proliferation of unsafe code. What if the linked file name 
> has some peculiarities and characters interpreted by shell?
>
>     See [[file:/tmp/`touch /tmp/hacked`/test][here]]

Thanks for the heads up! Updated version of the patch is attached.

> I can not say that I fully understand `org-open-file' code, so I am 
> unsure if remote file name can appear here, e.g. /ssh:user@host:testfie 
> or a file form an archive due to a relative link [[file:testfile]] from 
> a remote .org file. When remote files are not an issue, it is safer to 
> use functions that takes command arguments as a list of string, not the 
> command as a ready to execute string. Unfortunately there is no helper 
> returning a string and accepting a command as a list.

org-file-apps-gnu defaults to

((remote . emacs)
 (system . mailcap)
 (t . mailcap))

All the "remote" files (including /ssh) will be opened in emacs. We
should have no issue in this area.

>> +                          (mailcap-extension-to-mime (or ext ""))))
>>           (command (mailcap-mime-info mime-type)))
>>      (if (stringp command)
>>          (setq cmd command)
>
> P.S. `org-open-file' already has some problems with handling of some 
> file names:
>
> Maxim Nikulin to emacs-orgmode. greedy substitution in org-open-file. 
> Wed, 20 Jan 2021 23:08:35 +0700. 
> https://list.orgmode.org/ru9ki4$t5e$1@ciao.gmane.io

Agree. Though it is not directly related. Maybe you can bump that thread
and mark is as a bug report to be listed on
https://updates.orgmode.org/?

Best,
Ihor

>From 4aeff613c07d9025c5fc1f0b3851870a42d36869 Mon Sep 17 00:00:00 2001
Message-Id: 
<4aeff613c07d9025c5fc1f0b3851870a42d36869.1653538199.git.yantar92@gmail.com>
From: Ihor Radchenko <yantar92@gmail.com>
Date: Sun, 22 May 2022 12:04:35 +0800
Subject: [PATCH v2] org-open-file: Use file command to determine mime type,
 when available

* lisp/org.el (org-open-file): Prefer file command to determine file
type instead of relying on `mailcap-extension-to-mime'.  Only fallback
to the latter when file command is not available on the system.

Fixes 
https://list.orgmode.org/874k1n2hpv.fsf@localhost/T/#mcc10df4ea7937360671e6dea8061153dee518807
---
 lisp/org.el | 7 ++++++-
 1 file changed, 6 insertions(+), 1 deletion(-)

diff --git a/lisp/org.el b/lisp/org.el
index d7da8efc4..45a179a8a 100644
--- a/lisp/org.el
+++ b/lisp/org.el
@@ -7975,7 +7975,12 @@ (defun org-open-file (path &optional in-emacs line 
search)
     (when (eq cmd 'mailcap)
       (require 'mailcap)
       (mailcap-parse-mailcaps)
-      (let* ((mime-type (mailcap-extension-to-mime (or ext "")))
+      (let* ((mime-type (if (executable-find "file")
+                            (shell-command-to-string
+                             (format "%s --brief --mime-type %s"
+                                     (executable-find "file")
+                                     (shell-quote-argument 
(convert-standard-filename file))))
+                          (mailcap-extension-to-mime (or ext ""))))
             (command (mailcap-mime-info mime-type)))
        (if (stringp command)
            (setq cmd command)
-- 
2.35.1