Re: libsecret and dlopen/dlclose thread cleanup, dlopen freeze
- From: Emmanuele Bassi <ebassi gmail com>
- To: Gábor Csárdi <csardi gabor+gtk-list gmail com>
- Cc: ML-gtk <gtk-list gnome org>
- Subject: Re: libsecret and dlopen/dlclose thread cleanup, dlopen freeze
- Date: Sun, 5 Feb 2017 15:46:25 +0000
Not that I know of. Libsecret is a DBus wrapper around the shared
secret services provided by environments like GNOME (gnome-keyring)
and KDE (Kwallet); this usually implies that you're targeting either
of these environments, using their own DBus API (GDBus, for GNOME;
QtDBus for KDE).
Additionally, the python-dbus bindings used by that Python module use
the deprecated dbus-glib API, so I would not recommend using them
either.
Ciao,
Emmanuele.
On 5 February 2017 at 15:19, Gábor Csárdi
<csardi gabor+gtk-list gmail com> wrote:
Thanks much! This again saved me a lot of time and effort. OK, so I'll see
what I can do with libdbus then.
My last question. Do you know if anyone has even written a libsecret client
on top of (lib)dbus only? The secretstorage Python module
(https://github.com/mitya57/secretstorage) seems to do that, but it uses the
Python dbus client, and I'll need to write everything in C.
Thanks!
Gabor
On Sun, Feb 5, 2017 at 1:30 PM, Emmanuele Bassi <ebassi gmail com> wrote:
Hi;
On 5 February 2017 at 12:51, Gábor Csárdi
<csardi gabor+gtk-list gmail com> wrote:
Thanks! R is single threaded and synchronous (focus is computation, not
I/O), so I don't need threaded code at all.
Okay; just remember this in case you need to do asynchronous work.
Btw. I could also just have a dbus client using TCP/IP, right?
No. The support for unauthenticated TCP connections is meant for
private connections, which means you would not be able to access the
session bus where the secrets manager lives.
You'd need a special system configuration to allow TCP connections to
the session bus, but it's *strongly* discouraged to do so, unless you
want third parties to listen in to everything that gets sent through
the bus.
Ciao,
Emmanuele.
Gabor
On Sun, Feb 5, 2017 at 11:36 AM, Emmanuele Bassi <ebassi gmail com>
wrote:
Yes. Libdbus is a fairly low level API and has known limitations for
threaded code. It's the reference implementation of the DBus
specification,
and as such it should only be used if nothing else is available.
Ciao,
Emmanuele.
On Sun, 5 Feb 2017 at 10:51, Gábor Csárdi
<csardi gabor+gtk-list gmail com> wrote:
Thanks much, this saves me from a lot of trouble!
I'll look at libdbus. That's probably a lot more work, I'll need to
implement a small libsecret client, right?
Thanks,
Gabor
On Sun, Feb 5, 2017 at 10:36 AM, Emmanuele Bassi <ebassi gmail com>
wrote:
Hi;
You cannot ever unload libgobject - and any other gobject based
libraries - with dlclose() after dlopen()-ing them. The type system
cannot
be reset on library unload, and thus types that were registered once
will
still be alive the next time you load the library. Additionally, GLib
and
GObject and GIO have threads and additional state created for things
like
GDBus, which libsecret uses internally to talk to the secrets
service.
In short: if you want to have a plugin that talks to the password and
secret storage through the Secret Service DBus API, you will have to
use a
library that allows unloading and reloading; likely only libdbus.
Alternatively, do not allow disabling the plugin if you want to use
libsecret.
Ciao,
Emmanuele.
On Sun, 5 Feb 2017 at 10:27, Gábor Csárdi
<csardi gabor+gtk-list gmail com> wrote:
Hi all,
I have a plugin that uses libsecret, and thus glib, on Linux. (It is
a
package for R.)
It is loaded with dlopen(), and it (ideally) can be unloaded with
dlclose(). I use the synchronous libsecret functions, which AFAICT
run a
temporary glib event loop for each libsecret query. E.g. this is
what I am
doing:
GError *err = NULL;
gchar *password = secret_password_lookup_sync (
keyring_secret_service_schema(),
/* cancellable = */ NULL,
&err,
"service", cservice,
"username", cusername,
NULL);
if (err) {
if (password) secret_password_free(password);
keyring_secret_service_handle_status("get", TRUE, err);
}
if (!password) {
error("keyring item not found");
return R_NilValue;
} else {
secret_password_free(password);
/* ... */
return result;
}
When the lib is unloaded I call secret_service_disconnect(), I
checked
that this is called. I do link to libgobject-2.0, I even call
g_type_ensure
(G_TYPE_OBJECT) to make sure that the constructors in libgobject
run.
However, it seems that the worker threads for glib (?) are never
cleaned up. Even if I dlclose() the lib, they are there. What am I
missing
here? Do I have to eliminate the threads manually? If yes, how can I
do
that?
An even bigger problem is, that if I dlopen() the same plugin again,
and call a libsecret function, then it just freezes with this:
(process:18663): GLib-GObject-WARNING **: cannot register existing
type
'SecretService'
(process:18663): GLib-GObject-CRITICAL **:
g_type_add_interface_static:
assertion 'G_TYPE_IS_INSTANTIATABLE (instance_type)' failed
(process:18663): GLib-GObject-CRITICAL **:
g_type_add_interface_static:
assertion 'G_TYPE_IS_INSTANTIATABLE (instance_type)' failed
(process:18663): GLib-CRITICAL **: g_once_init_leave: assertion
'result
!= 0' failed
(process:18663): GLib-GIO-CRITICAL **:
g_async_initable_new_valist_async: assertion
'G_TYPE_IS_ASYNC_INITABLE
(object_type)' failed
Which seems to suggest that I need to unregister the SecretService
type
at dlclose(). Is that right? How can I do that?
I don't have too much experience with glib, and I am just probably
doing something wrong, or missing something. Any help is
appreciated. I can
also put together a reproducible example if needed.
Thanks, Best,
Gabor
_______________________________________________
gtk-list mailing list
gtk-list gnome org
https://mail.gnome.org/mailman/listinfo/gtk-list
--
https://www.bassi.io
[@] ebassi [@gmail.com]
--
https://www.bassi.io
[@] ebassi [@gmail.com]
--
https://www.bassi.io
[@] ebassi [@gmail.com]
--
https://www.bassi.io
[@] ebassi [@gmail.com]
[
Date Prev][
Date Next] [
Thread Prev][
Thread Next]
[
Thread Index]
[
Date Index]
[
Author Index]