No sound from aegisub

Esteemed maintainers (and contributors),

Attached is a patch which, as far as I can see, should work. But it does

not. I would like to submit it here for review. Please tell me what I am

doing wrong or what I am failing to do, so that I might clean up this patch

and submit a proper one.

1. The problem this patch is trying to solve:

Aegisub (guix install aegisub), as it currently is, does not give any sound

(at least on foreign distros). The terminal output reads as follows, when a

video file (with audio) is added into the program and an attempt is made at

playing it: (Video is added to aegisub by using 'Video > Open Video...')

```

ALSA lib conf.c:3683:(snd_config_hooks_call) Cannot open shared library

libasound_module_conf_pulse.so

(/gnu/store/zcjdb23gbhl0pcnvvm8rnlprkfl43cv5-alsa-lib-1.2.2/lib/alsa-lib/libasound_module_conf_pulse.so:

libasound_module_conf_pulse.so: cannot open shared object file: No such

file or directory)

ALSA lib pcm.c:2642:(snd_pcm_open_noupdate) Unknown PCM default

```

The complaint being that it can't find libasound_module_conf_pulse.so. This

.so comes from the guix package `alsa-plugins:pulseaudio`. The /gnu/store

directory for it is something like

/gnu/store/...-alsa-plugins-1.2.2-pulseaudio/lib/alsa-lib/libasound_module_conf_pulse.so.

However, the .so file is being looked for inside of the /gnu/store path to

alsa-lib

(/gnu/store/...-alsa-lib-1.2.2/lib/alsa-lib/libasound_module_conf_pulse.so.

2. The Natural Solution:

Since libasound_module_conf_pulse.so is supplied by

`alsa-plugins:pulseaudio`, and the program is looking for it inside the

path to alsa-lib (supplied by `alsa-lib`), union-build is a logical

solution. This is what the patch implements.

3. Problem:

Despite this patch being applied, it continues to look for the .so file

inside the path to alsa-lib and not the union-alsa. What is more, alsa-lib

is not even in the inputs. Only union-alsa is. And yet, it seems to be

trying to use alsa-lib's /gnu/store path, which, the way I see it, violates

guix's functional package management.

4. It's not guix, it's me:

I'm sure this is some mistake on my part. Hence, I submit the diff here for

review.

P. S: The diff has changed a bit because I got this idea of moving

out/lib/alsa-lib/* into out/lib/ just to see if that makes any difference.

Only to realize a little too late that it doesn't. But I send the diff with

those changes included because I didn't stash the previous (cleaner) one.

I forgot to mention, but the union-build works. If one `guix build`s the

union package, and cd into the /gnu/store/...-union-build, one will see

that things are there as it should be (lib/normal-alsa-*.so and

lib/alsa-lib/*pulse*.so).

It's just that aegisub seems to be looking from the

/gnu/store/normal-alsa-lib-dir instead of /gnu/store/union-alsa-package.

Another information:

I have other programs on my foreign system (installed via guix) that

produce sound well, without any issues. They include:

espeak and kdenlive.

Also, I did try making the union-build use copies instead of symlinks using

(union-build ... #:symlink copy-recursively) but that didn't work out

either.

Now, I'm going to take a look at fixing the pkg-config rules, and will be

updating this thread should that resolve this issue.

On Sat, Sep 26, 2020 at 12:21 PM Prafulla Giri <pratheblackdiamond@gmail.com>

wrote:

Hi Prafulla,

Le 26 septembre 2020 03:31:38 GMT-04:00, Prafulla Giri <pratheblackdiamond@gmail.com> a écrit :

I think the issue with the union-build is that it is completely ignored: alsa doesn't look in it's current directory for plugins, but in the directory where it's been configured (at build time) to look for.

On the Guix System, there is a service that creates a /etc/asound.conf, which references alsa-plugins:pulseaudio. Maybe you actually want to configure that? Maybe try to copy this to a new file .asoundrc:

pcm_type.pule {

lib "/home/foo/.guix-profile/lib/alsa-lib/libasound_module_pcm_pulse.so"

}

ctl_type.pulse {

lib"/home/foo/.guix-profile/lib/alsa-lib/libasound_module_ctl_pulse.so"

}

pcm.!default {

type pulse

}

ctl.!default {

type pulse

}

So, there were a few issues with my union. I've fixed them. Attached is the

diff as it stands right now.

Hello Mr. Lepiller,

Thank you for the tip. For the moment, I want to poke around this issue.

Perhaps I get absolutely tired of it (or find a good hack - and an ugly

patch). I did manage to learn a bit more about union-builds in general. So,

that's a silver lining.

Also, your work on guix-android seems so very cool! Please keep up the

great work!

On Sat, Sep 26, 2020 at 3:28 PM Julien Lepiller <julien@lepiller.eu> wrote:

Okay, I think I'm stuck now. lol

On Sat, Sep 26, 2020 at 3:42 PM Prafulla Giri <pratheblackdiamond@gmail.com>

wrote:

Hehe, thanks for your interest in guix-android. It's not really done yet, but I'm very motivated :). It always feels weird when someone on the internet calls you Mr ^^. No need to be so stiff, you can call me "Julien" (and I suppose it's the norm to call people by their first name in this kind of internet community).

Maybe try to run aegisub with strace -f, you might find out something. You may have to take a look at alsa's sources to find out how it looks for its plugins exactly (maybe try to grep for PLUGIN or PLUGINDIR)?

Le 26 septembre 2020 08:07:34 GMT-04:00, Prafulla Giri <pratheblackdiamond@gmail.com> a écrit :

Oh, I come from a culture that does not call seniors by their first names.

(:

Yes! guix-android looks super awesome! I'm glad that you are working on

that. You're awesome.

(Also, I just realized that you've also got guix-home-manager and so many

more cool repositories! I need to check those out, and soon!)

I did manage to resolve this issue. Kinda'. Here:

All I had to do was re-define 'alsa-lib' to point to the /gnu/store path of

alsa-plugins. The /gnu/store path had to be figured out during runtime.

Could you please check how this fares in a GuixSD system? If it goes well,

perhaps I should send a patch in for aegisub. But then again, I wonder if I

should make the changes in alsa-lib itself, but I am not very sure about

that.

On Sat, Sep 26, 2020 at 8:20 PM Julien Lepiller <julien@lepiller.eu> wrote:

Thanks! On the guix system, aegisub already works properly (at least it did when I last used it), because the alsa config allows it to find pulseaudio.

There are three issues with your proposed change:

First using derivation-outputs is a bad idea, and it probably only works because alsa-plugin doesn't depend on your modified alsa-lib (you'd get an infinite loop I think). Since alsa-plugin doesn't depend on this modified alsa-lib, you don't need such a convoluted work around. When defining the conhigure flags, you can simply use (assoc-ref %build-inputs "alsa-plugin") when "alsa-plugin" is defined as an input to the package.

Second, setting plugindir at build-time is just asking for the same kind of issues you are trying to fix, only for finding another plugin that might be needed.

Third, the way you modify the arguments is… unconventional :) try using substitute-keyword-arguments instead. There are plenty of examples in the repo.

I remember having dealt with this issue before, and pushed a patch some 3 years ago: http://issues.guix.gnu.org/31239

Could you look at what happened in the meantime? I think it was reverted in favor of the asound.conf file we have in the guix system now. You might learn something if you find the reason why it was changed.

Re my other projects, framagit is going to close next year, so I'll be migrating to my own hosting, currently at git.lepiller.eu.

Ok for the cultural difference. It's probably too soon to call me a "senior" though, I'm still pretty young: only 3 years older than you :p

Le 26 septembre 2020 16:46:50 GMT-04:00, Prafulla Giri <pratheblackdiamond@gmail.com> a écrit :

First and foremost: wow. A post-doc at just three years older than me.

*cries in high-school drop-out* Super awesome!

I have cleaned up my package definition as per your review:

I will look into the matter then. Might take me a while, but I'll try.

It's sad that framagit is closing. I think they were doing a great job with

peertube and all...

On Sun, Sep 27, 2020 at 3:00 AM Julien Lepiller <julien@lepiller.eu> wrote:

Le 27 septembre 2020 01:56:31 GMT-04:00, Prafulla Giri <pratheblackdiamond@gmail.com> a écrit :

Well, it's framasoft, they started so many services that they can't handle the load anymore. So they've started closing them little by little last year. Some of them will stay, and others will disappear. They never wanted to centralize the web around them, so it's understandable. Instead, they propose other alternatives, they've started the chaton initiative (where other structures propose the same kind of services, at a smaller scale), etc.

They're still developping peertube, don't worry about that. But they're not going to host a public instance. Remember they're about 10 volunteers and one employee :)

I see...

Hmm.... I didn't know they were just 11 people.

Anyways, here is something that an ALSA developer told me... I couldn't

make much of it. But Perhaps you could?

The current alsa-lib code uses the RTLD_DI_ORIGIN glibc extension to

Perhaps you'd like to see our conversation over at GitHub:

You're far more qualified to do this than I am.

On Sun, Sep 27, 2020 at 4:55 PM Julien Lepiller <julien@lepiller.eu> wrote:

Ok, I tgink that means alsa-lib looks for plugins in its rpach, fails to do so, tgen fallls back to its hard-coded path, and fails again. So if we set the rpath correctly (which we cannot do in a package because it would introduce a cyclic dependency), it can load its plugin.

Could you try to set LD_LIBRARY_PATH like so:

LD_LIBRARY_PATH=<alsa-plugin's pulseaudio output>/lib aegisub

Le 28 septembre 2020 10:43:02 GMT-04:00, Prafulla Giri <pratheblackdiamond@gmail.com> a écrit :

Is this a duplicate of #40832, "alsa-lib cannot find its plugins"?

The patch I submitted in that bug ticket does work. I haven't found the

energy to perform a final review of the C code.

retitle 43628 No sound from aegisub