Abolishing tail call levels in 5.2 (Was Re: debug.getlocal losing information)

[GrayFace <sergroj@...>]

On 08.07.2010 4:17, Roberto Ierusalimschy wrote:
> > What are the drawbacks of old behavior from the practical point of
> > view? IMO, the more tail call is different from normal call, the
> > more confusion it makes.
> >      
> This has been extensively discussed in the past. (Of course in the
> reverse; list member complained about the old behavior and we defended
> it.)  Seach the Lua list for "tail calls" (or tail calls debug).
>    
In posts I found the main point against reporting tail calls was about 
stack machines. If a stack machine relies on tail calls, it would 
generate lots of tail calls that will pollute the traceback.
I think it can be dealt with by a simple debug.traceback modification. 
It can be made to report one "(15 tail calls): ?" line instead of 15 
"(tail call): ?" lines if there were 15 tail calls. This would look 
pretty, IMO.
The number of tail calls before a stack level can also be exposed to 
debug interface, so that it can skip the tail calls instead of iterating 
through them.

Here are the results of search - some posts on the issue I found most 
interesting:

Discussion of whether tail calls are good or not in a regular traceback:
http://lua-users.org/lists/lua-l/2005-08/msg00163.html

An argument about easier debugging with 5.1 behavior:
http://lua-users.org/lists/lua-l/2003-09/msg00330.html

Non-tailcallable closures proposal:
http://lua-users.org/lists/lua-l/2005-08/msg00159.html
(this one could solve the thread starter's problem with debug.getlocal use)

> In essence, tail calls *are* different from normal calls, no matter
> what the debug interface tells you. So, the question is not about how
> different tail calls are from normal calls, but how much the debug
> interface tries to hide these differences from you.
>    
It does the job well. In 5.1 only the debug interface shows that 
difference. After the change it would also affect 'error' function from 
the usual interface.
I can hardly imagine tail call levels being a source of confusion. I'm 
sure confusion would grow significantly after the removal of tail call 
levels. The same syntax of normal calls and tail calls suggests that 
they are mostly similar and that's the source of confusion.

--

___________________________________________
Best regards,
Sergey Rozhenkomailto:sergroj@mail.ru