Re: feature/android b91e8ada70e 3/5: Fix auto-revert-mode on Android

[Po Lu]

Michael Albinus <michael.albinus@gmx.de> writes:

> Po Lu via Mailing list for Emacs changes <emacs-diffs@gnu.org> writes:
>
> Hi,
>
>>     Fix auto-revert-mode on Android
>>
>>     * src/inotify.c (Finotify_add_watch): Handle asset files.
>> ---
>>  src/inotify.c | 19 ++++++++++++++++++-
>>  1 file changed, 18 insertions(+), 1 deletion(-)
>>
>> diff --git a/src/inotify.c b/src/inotify.c
>> index 7562ffb1701..b2a48884efa 100644
>> --- a/src/inotify.c
>> +++ b/src/inotify.c
>> @@ -419,6 +419,7 @@ IN_ONESHOT  */)
>>    int wd = -1;
>>    uint32_t imask = aspect_to_inotifymask (aspect);
>>    uint32_t mask = imask | IN_MASK_ADD | IN_EXCL_UNLINK;
>> +  char *name;
>>
>>    CHECK_STRING (filename);
>>
>> @@ -432,7 +433,23 @@ IN_ONESHOT  */)
>>      }
>>
>>    encoded_file_name = ENCODE_FILE (filename);
>> -  wd = inotify_add_watch (inotifyfd, SSDATA (encoded_file_name), mask);
>> +  name = SSDATA (encoded_file_name);
>> +
>> +#if defined HAVE_ANDROID && !defined ANDROID_STUBIFY
>> +  /* If FILENAME actually lies in a special directory, return now
>> +     instead of letting inotify fail.  These directories cannot
>> +     receive file notifications as they are read only.  */
>> +
>> +  if (strcmp (name, "/assets")
>> +      || strcmp (name, "/assets/")
>> +      || strcmp (name, "/content")
>> +      || strcmp (name, "/content/")
>> +      || strncmp (name, "/assets/", sizeof "/assets")
>> +      || strncmp (name, "/content", sizeof "/content"))
>> +    return Qnil;
>> +#endif
>> +
>> +  wd = inotify_add_watch (inotifyfd, name, mask);
>>    if (wd < 0)
>>      report_file_notify_error ("Could not add watch for file", filename);
>
> I have two problems with this patch. First, nil is not defined as return
> value of *-add-watch functions. If we want to allow this, we must extend
> the interface.
>
> Second, do we really want to have such a hard coded list in inotify.c?
> What about other directories on other systems, like "/proc"? Shouldn't
> we add rather a common interface for excluding directories from being
> watched, say file-notify-excluded-directories (a list of strings)? This
> could be modified on Lisp level, no need to do it in the notification
> backends.

Shouldn't watching /proc work, in that no error is signalled, unlike
watching /content and /assets, which are wholly inventions of android.c
which do not exist in the VFS?

Anyway, what return value do you think would be appropriate here?

Thanks.