Lv2 UI - Testing #7201
Lv2 UI - Testing #7201
Conversation
|
This now runs successfully in the CI, and it supports UI there for Linux and Macos, but not Windows yet. @DomClark , as discussed, I would appreciate if you would submit a vcpkg PR (and then add suil to our vcpkg.json in this PR or tell me to). In the end, the CI for Windows should not show but |
|
Dexed.
|
I assume it's related to this PR: #7247 |
The LV2 implementation in LMMS uses linked models, so mono plugins are actually 2 linked plugin instances (L and R) behind the scenes. Linked models need to be kept in sync for parameter updates which is already a complicated task, but adding a UI where updates can come not just from LMMS but also the plugin itself makes the linked model design untenable in my opinion. The issue with dexed sounds like exactly the type of synchronization problem I feared. I would strongly recommend to @JohannesLorenz to replace the linked model design with the L/R routing design from #7247 once it's merged. I've been using it successfully for both CLAP and VST plugins and it works great. Making the switch will almost certainly be less work than trying to fix the bugs with linked models. Not only is the L/R routing approach dramatically simpler and less buggy than trying to use linked models, it is also very flexible and intuitive from a user standpoint and uses roughly half the resources in the common situation where you only want a mono-to-stereo mix from your mono plugin. |
Has this already been fixed? It sound like a waste of resources, a source for tons of bugs and a complete misunderstanding of how these things work. Sorry, if I am a bit blunt here but keeping it like this would take the implementation in a completely wrong direction. I am pretty sure that no other DAW implements it like this. If a plugin reports as mono then simply copy the resulting mono output buffer of the single instance into the left and right channel buffers of the stereo track.
100% agree! |
|
I pushed a few commits to help with AppImage findings over on Discord's For now, I copy the (Until #7252 is merged, I can't test this on ARM64, so I'm just hoping this fixes the AppImages) Initial testing on MacOS is OK but we don't use a launcher script for MacOS, Perhaps 
|
|
When I reload a project the lv2 gui doesn't update. The settings seem to stick, just the gui is showing it's init setting. |
@zonkmachine Can you please elaborate? I did:
=> All params are as before. Did I understand "reload" wrong? |
With nekobi I do: I save and reload and I get: The sound is the same as when I saved but the settings show a default nekobi. |
|
@zonkmachine Thanks. I fixed it. |
This took care of the instruments but the effects see the same issue and this is not fixed by 3bcf80b. |
Actually, this commit should have fixed it for both... Which effect do you mean? Sometimes, producing input for an effect (i.e. pressing a key) might force a lazy UI to update. Not sure if this helps here. |
Ping Pong Pan There were more. Actually, the only Lv2 effect with gui that I've come across that works in this regard is Calf Analyzer (not an effect per se but the stored parameters are displayed correctly when reloading a project).
Nope. |
|
Oh, I get it. For instruments, first, the instrument is loaded, which includes loading the savefile. Only after that, the UI is initiated - at which time the mentioned fix sends all port values (knobs, sliders etc) to the UI. Thinking about a fix... Confirming: Ardour sends port updates of all ports to UI after loading a preset ( |
|
New issue. The setBfree organ plugin doesn't save the settings at all. |
Does this work in Ardour or another DAW |
Yes. Works fine in Ardour. |
|
setbfree gui explained here: https://setbfree.org/lv2/lv2_setbfree_up.html |
I think some use "Preset" and some use "State". Multiple of them contain "organ" in the description. Do you mean any specific one? |
No. I'm very unspecific here. |
|
I cannot reproduce your Dexed issues: I turned all 8 "EG" level in "1" to 0, and when loading, they were still 0 and I think it sounded correct. How do I reproduce? With X42 whirl (from the b_free plugins), I can reproduce the issue. |
|
77f2490 should now finally fix all loading issues. However, saving does not seem to work (as you reported) - even on the same effects that did not load: Ping Pong Pan and 3 Band EQ. Investigating further... |
|
The setBfree organ, Dexed and Zyn all are unable to save anything. This requires the "State" extension, since these do not have ports. I think we should (independently) ban all plugins that use state - until this is implemented. Otherwise, users will get annoyed because their GUI values are not being saved. |
Fixed in dbe71a5. From now, all know load/save issues should be fixed! |
Right. How is it with Ardour? If Ardour does not tell you what C4/C5 is, you could compare it with help of a different instrument. |
Right, right. Did a quick test in Ardour and it's the same issue there. I'll look into this some more. |
|
I fixed really a lot of issues. One remaining issue though is that the opening LMMS effect window sometimes is a bit too small to contain the UI. This will hopefully be fixed soon. Aside from this, if anyone finds any issues with the current version, please post it. Also, keep in mind the "When reporting an issue" section from above, and keep in mind that the PR does not cover automation of UI controls, presets and LMMS file browsers inside of the UI. |
|
I can't believe I did this again... Oh, lawd! |
include/Lv2Basics.h
Outdated
| template<class T> | ||
| struct LilvPtrDeleter | ||
| { | ||
| void operator()(T* n) { lilv_free(static_cast<void*>(n)); } | ||
| }; | ||
|
|
There was a problem hiding this comment.
I don't think this needs to be a class template. If the operator takes void*, it should still work with std::unique_ptr.
There was a problem hiding this comment.
A void* would suffice, but then, in statements like
auto bundlePath = AutoLilvPtr<char>(lilv_file_uri_parse(bundleUri, NULL));we would lose the information that this is a char pointer. So, to improve type safety, I think a template makes sense.
There was a problem hiding this comment.
No, you can have a std::unique_ptr<T> that stores T* but uses a deleter which takes void*. This works because T* implicitly converts to void*.
Here's a working demo: https://godbolt.org/z/Msfrc5jfa
There was a problem hiding this comment.
True, but if we get an explicit cast for free, I prefer that.
|
Update: I have multiple times asked whether the LMMS-UI should be displayed even if a Plugin-UI is displayed, and the answer was rather yes, because it is required for automation. While this might be doable via the "UI touch" function, many plugins do not support "touch" at all (e.g. zyn does not), so for those, there is no other choice. This leads to the following decision:
In other words, the LMMS UI is displayed in any case, though sometimes, an expander to hide it might be useful. |
http://lv2plug.in/ns/extensions/ui Co-authored-by: Tres Finocchiaro <[email protected]>
NOTE: This commit MUST GO BEFORE the lv2 UI commit
This fixes `EffectRackView` to have a permanent size hint instead of resizing the widget once in `InstrumentTrackWindow`. The size hint tells the `InstrumentTrackWindow` to not increase with a growing number of effects in the `EffectRackView`.
|
Some of the LV2's don't fit well in their windows. I don't remember this from earlier. This is how Dexed shows up, with a lot of margin on top and to the left and cut off on the bottom/right. I also can't change the size of the window manually. |
|
Don't you think it would be better if lv2 plugins had their own separate window like vst plugins? Don't you think it would be better to have a "vestige" type instrument but for lv2 plugins? I think we would have two advantages over the current way of displaying lv2 instruments and that is that it would be better visually and more consistent with the size of the other instruments, and that each plugin has a different size, they cannot be resized and this can be very annoying on small screens. The other advantage would be better performance since on low-resource computers the graphical interfaces of the plugins being open can consume more resources; however, from a "vestige" type instrument window but with the option (a "Show/Hide GUI" button) to show the graphical interface to make some precise changes, you can control some parameters of the plugin without opening its graphical interface. (Google Translate) |
This is a know issue (still open in my checkbox list at the OP): "Sometimes, plugins UIs do not have optimal size.". This waits for L/R routing because after that, the way we put the plugin into the window changes.
Dexed does not allow resizing. However, as soon as we have fixed the size, resizing should not be required anymore. |
|
@Gabrielxd195 Thanks for your input.
I don't find it necessary to be exactly like VST.
This could be a point, but most Lv2 instruments already have a larger than the default size (before lv2-ui). We did this to use the space better, especially on large monitors. E.g. try Helm, putting all these knobs into the small default-sized window with only 1 vertical scrollbar would be insane to navigate.
We had a discussion in Discord lately, and the result was that I wanted to try out a TabWidget, like we currently have it for the Instrument Track Window already (where you can switch between Plugin, Envelopes, MIDI etc.). If the UI is hidden at the start, it can be loaded at the moment where the user switches the TAB. Ultimatively, it is up to the users what they prefer. But whatever we do, it is a UI thing, so we can still change it easily... |
|
In any case there must not be another step to open the ui just because it's external. The way vestige works is horribly unintuitive and lv2 needs to move away from that approach. Plugin handling must be transparent and the user should get the same experience regardless of the plugin format. |
It might not be the scope of this pull request but in general I agree with what @zynskeywolf wrote. Ideally all external plugins should feel like entities of their own and the parameter handling should be transparent to the users (also with regards to plugin type, i.e. VST, LV2, CLAP, etc.). Many other DAWs just present you with a list of the available plugin parameters. If you select a parameter from that list you can then define the automation curve for it, etc. I once did some investigation with regards to some problems with plugin parameters and if I remember correctly, what I found was that LMMS seems to automate the movement and changes of the GUI elements and not of the parameters themselves. If that is really the case then this would need to be changed first before better and more uniform parameter handling could be implemented. If I understand correctly then this is also the reason why some extra parameter widgets need to be shown for the parameters in the first place. If the automations worked directly on the plugin parameters then the would not be needed. |
I agree here, though it is indeed a bit overhead to load the plugin UI. My idea is to have a tab widget to switch between both UIs. If the user previously selected the plugin UI, the next plugin will also let the TabWidget show the plugin UI. If they previously used the LMMS UI (the parameters), they get this when they open a plugin the next time. |
This is solely for early testing yet. From the code review perspective, there are still a few things to be done.
When reporting an issue, please try to answer the following questions:
Open issues: