How to replace modules that have been loaded into a Slang session before with updated code?

[brussig-tud]

Hi,

I'm having trouble finding out how – or if at all – it's possible to replace a module once it has been compiled to IR inside a session. My use case is essentially something like hot reload of shaders: users should be able to change files on disk, and only a file that changed should have to undergo the from-scratch translation to IR. Or certain shader code should be online-editable within the application and get recompiled. I'm aware of course that my program is then responsible for re-composing and re-linking if that module was used somewhere.

Currently, if I just load the module again via ISession::loadModule, Slang reuses the cached IR translation result for that module. Same for ISession::loadModuleFromSourceString if I use identical values for moduleName and path.

Of course I could just use a new moduleName whenever I want to recompile a module, but that will start eating up RAM (and rather quickly at that), and I don't see a way to "unload" modules. I feel like I'm missing something obvious. I keep my references to the affected slang::IModules around of course, if that helps.

[H7perus]

There is currently no way to unload a module afaik. The only way to get the newest version of a shader/module is to recreate the session.
At least that was the answer when it came up before.

Imo there should be a way to invalidate and reload a module, since it saves a lot of time in hot reloading single shaders.

[brussig-tud]

That's a shame. It would be nice to get some sort "official" statement why module reloads aren't a thing though, or if it just wasn't on the agenda yet. Do you happen to still have a link to the discussion where this came up before?

[H7perus]

https://discord.com/channels/1303735196696445038/1387145052605911090/1387940864642256977
I saw it in other places too, but I don't recall exactly right now. I might come back to it, if I come across such info again.

[brussig-tud]

Thanks. Unfortunately, no explanation is offered beyond "nope". Oh well, maybe I'll open an issue later.