Viktor Malík
Viktor Malík
Turns out the same problem exists when dereferencing via array accesses, i.e. `args.trace_events[0]->flags`. I'll try to fix that, too, in this PR.
Fixed the array access on our side, too. I'll report/fix kernel upstream. Originally, I wanted to separate the dereference fix from the array access one b/c the latter is one...
> Why is it that we need to reset `is_btftype` on the type contained within the array, but for pointers only the outer type gets reset? It's actually the other...
> Is it possible to add a unit test for the FieldAnalyser change? Sure! I added a test (and confirmed that it failed prior to this PR).
While adding the test, I updated `data_source.c` for BTF and DWARF which changed one codegen test.
This slipped off my radar. I rebased onto the current master and will merge today afternoon if there are no objections.
> Looks like a legit unit test failure This is a tricky one. It doesn't reproduce on a local build (at least not on my setup), only in Nix. It's...
> In the meantime, as a workaround to unblock this PR, you can add `__attribute__((noinline))` to `func_3`, which force GCC to emit enough DWARF for LLDB to correctly locate the...
> Fixing this would technically be backwards incompatible. But surely nobody is relying on this buggy behavior, right? Right, we shouldn't maintain backwards compatibility of an obviously buggy behaviour.
> The BPF verifier requires that all defined subprograms are used by the main program. > > Our current approach involves appending all subprograms into `.text`, whether they are used...