Intermittent errors about luatrace HOT 4 CLOSED

Any update on this?

from luatrace.

Sorry, last time I looked at this I was pushed for time and couldn't work anything out because the stack traces in the two are exactly the same. Turns out that the problem is that both traces trace off the top of the trace stack (I'm not quite sure how this happens because when the trace starts there's an attempt to build the full current stack), but a the failing trace records a line with non-zero time when were "above" the top of the stack. The quick and dirty solution is to just ignore time that occurs in a stack frame we don't know about.
I'll have a look at the initial stack build and see if there's anything obvious, but I'm not hopeful.

from luatrace.

Nope, that wasn't it at all.
trace-out-failing.txt includes a line that records time on a line that isn't in the current function's set of lines. I don't know why that might be.
When this happens, profile guesses that it's LuaJIT forgetting to tell us about a return (this used to happen, I don't know if it still does) and adds an extra return before continuing.
In this case, it's the wrong thing to do. The extra return means that we jump off the top of the stack when we should be returning to test_json.lua.
So now, when we find time attributed to a line that's not in the current function, we look and see if the line is in the caller of the current function. If it is, then we guess the missing return as before. If it's not, we give up, complain, but don't add the extra return.
Which, in this case, prevents the stack getting out-of-whack by one frame.

from luatrace.

Well, with your latest changes I can't reproduce the issue, so it seems to be fixed :)

Thanks a lot!

from luatrace.