[llvm-branch-commits] [lldb] release/22.x: [lldb] Backport FrameProviders fixes & improvements (PR #180071)

[Med Ismail Bennani via llvm-branch-commits]

medismailben wrote:

> Yes, those commits are part of the new `ScriptedFrameProvider` feature which was added in lldb 22. See #161870 for the "primary commit" of this feature.
> 
> * Crash & deadlock fix ([c373d76](https://github.com/llvm/llvm-project/commit/c373d7632ac1fe8374b7f2b9536d7e746ed24786)) - should clearly be backported
> * Usability improvement ([943782b](https://github.com/llvm/llvm-project/commit/943782be5aec5db854065145dd73a618c3a775be)) - without this `target frame-provider register` only takes effect the next time the debugee stops. Not a nice experience without this patch. I would also consider this a bug fix.
> * Chaining multiple frame providers ([17b01bb](https://github.com/llvm/llvm-project/commit/17b01bbc67c484068e174d9e9a80089102bea0e8) + test fix [cd70e2d](https://github.com/llvm/llvm-project/commit/cd70e2d8367493785f1591045db2151540b14b91)) - this was functionality was agreed upon as part of the original discussion in [[lldb] Introduce ScriptedFrameProvider for real threads #161870](https://github.com/llvm/llvm-project/pull/161870), but only landed after the branch cut. I guess we could live without it, but I consider it rather low risk
> * Add support for ScriptedFrame to provide values/variables ([10f2611](https://github.com/llvm/llvm-project/commit/10f2611c2173783efae8aebc32d1515013271b64) + preparatory commits [58f623c](https://github.com/llvm/llvm-project/commit/58f623c504d14743e465f988848f9c16dc220a3b) [8122d0e](https://github.com/llvm/llvm-project/commit/8122d0e4bc8b536e0bb6bb5fae97e8216986ee28)) - this adds additional capabilities to `ScriptedFrameProvider`, first discussed in https://discourse.llvm.org/t/synthetic-frame-variables/89585 11 days ago (i.e. after the branch cut). One could argue whether this is a new feature and shouldn't be backported. That being said, not backporting those might be annoying due to rebase conflicts

+1, this feature has already landed partially in lldb 22 but we found a bug late in the release cycle where if a user had a SBFrame and try running an expression, that would invalidate the SBFrame reference. This is very intrusive because it's very common thing to do, specifically in IDEs ... so yeah, we definitely want to pick this up in the upcoming release.

Thanks.

https://github.com/llvm/llvm-project/pull/180071