Gajim - 2025-08-07

[Debian 12, Xfce4.18] '~$ /usr/bin/gajim' answers: "No translation path available. Gajim is already running. The primary instance will handle remote commands." What remote command is suitable to start the GUI of the program?

bullgard4: Gajim is already running. You need to use either that instance or close it before opening another one

Hello guys, just updated to 2.3.3, how can i reach the debug console now?

Shift + Ctrl + X (see hotkeys help dialog) or Preferences -> Advanced -> Debug Console

THX

hello

comparing ram consumption gajim vs pidgin

gajim 659mb pidgin 90mb is this normal?

tester123x: yes

lovetox pushed 1 commits to branch gajim/master imprv: OMEMOTrustManager: Add copy button and make some design improvements - https://dev.gajim.org/gajim/gajim/-/commit/e3c99ef07c4b31b601aec7d99666da12932f1e21

> gajim 659mb pidgin 90mb is this normal? Some haters are saying that Gajim is being developed in a scripting language. I don't know is this is true. I hope not. Some haters are spreading fake news, they say it's python, the script kids "programming" language.

RAM consumption probably has more to do with pidgin using an outdated version or GTK.

nonsense, gtk4 does not consume more RAM

It does. Example: https://forum.manjaro.org/t/pamac-gui-ram-usage/146639 Ram usage blowing up 4 times just between gtk3->gtk4. And pidgin is gtk2. Also, Dino consumes about as much RAM as Gajim, while not being Python. ✏

it's also written in some scripting junk

also comparing apple with oranges, pidgin almost a text only client, like some irc client

BUT, i definitly think there are some problems with memory managment in pygobject

it seems impossible to release memory once its allocated

but most people that write GTK applications dont care, because almost nobody writes long running GUI applications

> it's also written in some scripting junk No ↺

My bad, Dino actually consumes "only" 220 megabytes.

I just checked out of curiosity. Nautilus consumes 120 megabytes. It would seem to be of comparable if not greater gui complexity than chat clients. ✏

my Gajim consumes 215mb after ~5 hours

its a leak problem, it will consume more, the more chats you are joined, and the more you let it running

a file browser is typically never opened for a long time, so you will never notice leak problems

Hey how can i have image preview in group chats? i activated all the settings for preview but they have no effect on group chats

you click the download button and the image should show

no i want it to show by itself

not by clicking it each time

thats not possible in public chats

whtas that setting for then?

mine is using 350mb

better than discord imo

or matrix

de-facto, hm right there is a setting for that

I never closed my file manager dolphin just so I can open it instantaneously. I actually come to think of it, I think your argument that most GUI programs do not run for a long time is wrong. I think I run most programs I commonly use, basically 24-7. I just don't close them to not have to restart them. This includes emacs (54 megabytes, been running for days if not weeks), dolphin (75 megabytes, likewise)

maybe your file size limit is to low, its a setting in the same dialog

Funny thing that even includes freeCAD, inarguably an order of magnitude more complex software than any chat client or file manager, and it only consumes 44 megabytes. And it's also Python!

ok not sure how that helps us, you are free to look at the code and show me the 200 mb of things we load

So it looks like it's particularly pygobject that is to blame. Also gtk itself perhaps but not that much.

how can i make gajim to apply those settings to a groupchat? https://iili.io/Fs28tfa.png

> you are free to look at the code and show me the 200 mb of things we load I think that would be the lines starting with "from gi import"

uhhh

de-facto, does it never preview anything, or only sometimes not

never so far

i want to have auto preview of images because we have a flux image generation bot in another room

so send a image here

it outputs links to jpegs like above the image i linked e.g. https://iili.io/Fs28tfa.png

it shows the url, but i want ot see the image

without clicking on it each time

yes, send a image here

https://iili.io/Fs28tfa.png

works for me with the setting on

maybe check debug logs

but not like this https://iili.io/Fs28tfa.png

the bot will not output only the url, its contained in text

yes we dont preview links within text

thats bad :(

> > you are free to look at the code and show me the 200 mb of things we load > I think that would be the lines starting with "from gi import" how much ram does it use if theres no avatars? ↺

ok then that setting is useless to me and i will deactivate it

nautilus uses 302MiB of vram and 886MiB of ram

for me

is there a plugin that can fix this?

no

we work on supporting proper media sharing xeps

then your bot can use that to send images

it could also simply send a text message, and then a second message with the link

yeah its behind biboumi

lovetox, now you making me curious, what kind of proper media sharing xeps do you have in mind for e.g. a stable diffusion bot that directly is on xmpp?

does it matter which type of bot sends images+messages?

https://xmpp.org/extensions/xep-0447.html if you're curious

i see so that also includes http upload ✏

I’m profiling Gajim, and I’m surprised it spends 0.3s during startup compiling regexps on my laptop.

6056 0.009 0.000 0.323 0.000 /usr/lib/python3.13/re/__init__.py:330(_compile)

It’s compiling 6056 re expressions, with a mean time of 9ms per regexp.

de-facto, no, you can set any http link as source, you dont need to upload something somewhere

point is yo can attach many metadata, and a description

thats actually nice

Link Mauve, ok but how are you sure its Gajim?

it can be any library loaded with python, and im sure python standard libs itself compile regex

For sure, but Gajim does use Python’s standard libs.

Also many regexps are the exact same a lot of times.

It might make sense to cache at least some.

there is a single compile statement we dont cache

and its used once during connect

Hmm… Traceback (most recent call last): (?m)^ File "/usr/lib/python3.13/site-packages/gajim/gtk/application.py", line 372, in _handle_local_options self._startup() ~~~~~~~~~~~~~^^ File "/usr/lib/python3.13/site-packages/gajim/gtk/application.py", line 227, in _startup if not self._init_core(): ~~~~~~~~~~~~~~~^^ File "/usr/lib/python3.13/site-packages/gajim/common/application.py", line 113, in _init_core app.connections[account] = Client(account) ~~~~~~^^^^^^^^^ File "/usr/lib/python3.13/site-packages/gajim/common/client.py", line 107, in __init__ self._create_client() ~~~~~~~~~~~~~~~~~~~^^ File "/usr/lib/python3.13/site-packages/gajim/common/client.py", line 174, in _create_client self._client.set_lang(get_rfc5646_lang()) ^^^^^^^^^^^^^^^^^^^^^ AttributeError: 'Client' object has no attribute 'set_lang'

you need nbxmpp master

if you use gajim master

lovetox pushed 1 commits to branch gajim/master fix: Reactions: Correctly count when reacting to own messages - https://dev.gajim.org/gajim/gajim/-/commit/eefe12d02481abe6587759d875180d44e7a5ac42

ut of the 6056 regexps being compiled during startup, only 283 are unique.

So yeah, probably something to fix in Python’s modules.

or in some other lib we use

> BUT, i definitly think there are some problems with memory managment in pygobject Have you checked their issue tracker? They have many unsolved threads and even more sad un-reviewed PRs to fix memory leaks open for long times.

Even during runtime, you still compile the "re.compile('\ufdd0|\ufdd1|\ufdd2|\ufdd3|\ufdd4|\ufdd5|\ufdd6|\ufdd7|\ufdd8|\ufdd9|\ufdda|\ufddb|\ufddc|\ufddd|\ufdde|\ufddf|\ufde0|\ufde1|\ufde2|\ufde3|\ufde4|\ufde5|\ufde6|\ufde7|\ufde8|\ufde9|\ufdea|\ufdeb|\uf)" regex multiple times.

thats probably the emoji lib

https://pypi.org/project/emoji/

/usr/lib/python3.13/site-packages/packaging/specifiers.py is also a culprit of compiling the same regexps again and again.

lovetox, no, emoji doesn’t seem to do it wrong at least.

yeah also just checked, we also use only 2 methods from this lib

Unlike nbxmpp’s get_invalid_xml_regex().

uh yeah i didnt check nbxmpp

meh

its cached, the compile res is stored in a var

It still gets compiled multiple times, I don’t quite understand why.

once per Client() instance

so on a reconnect, and if you have multiple accs

could be better but we talk about single digit compiles

Yeah.

The packaging regexp though gets compiled a good hundreds of times during startup.

> how much ram does it use if theres no avatars? I think the impact of a couple of dozen 64px pictures would be negligible compared to 800mb ↺

lovetox pushed 1 commits to branch python-nbxmpp/master imprv: Better cache invalid xml regex - https://dev.gajim.org/gajim/python-nbxmpp/-/commit/f9a1cb90a7d3fea2a0b63f237e2077666d6df609

chud, couple dozen? alone in this chat > 400

Are you loading them all into RAM at once?

when we join a chat yes

Like, even when the participants list isn't open?..

or lets say when tey are requested

we load them on request, but if you open the participant list once, then all are requested

all the stanzas are conforming to a schema right?

> all the stanzas are conforming to a schema right? in theory yes ↺

is it possible to derive at each point in time of streaming a subset of valid tokens (e.g. substrings) that could be used to constrain an llm sampler?

tracemalloc doesn’t give me much of anything here.

i dont understand that sentence, and it will not help if you rephrase it :)

i was just wondering if it would be possible to have an llm directly talk xmpp, but have its output enforced to conform to the schema, so its always valid

the xml from xmpp looks beautifully structured, so llms should love that

Heh, 2 MiB of the memory are the emoji JSON, according to tracemalloc.

let get rid of emoji, non essential anyway

Gajim start for me with 170 mb, but with no chat open

i remember back in GTK3 days, it was in the best times 100 mb at start

but since then many features were added, many more libs pulled into

Nah, 2 MiB is nothing compared to 170 MiB.

It’s not worth it to remove a feature just to gain 2 MiB.

yes obviously :)

i dont think 200 mb is that much, 2 firefox tabs have that

but yeah the leaks are a problem, when you let gajim run for days, it will grow

> i was just wondering if it would be possible to have an llm directly talk xmpp, but have its output enforced to conform to the schema, so its always valid "When you have a hammer everything starts to look like a nail" ↺

im more concerned with performance then ram