OpenEuphoria

2. Re: Calling Func/Proc

• Posted by CChris <christian.cuvier at agriculture.gou?.fr> Jul 11, 2007
• 673 views

c.k.lester wrote:
> 
> I would really like to have a function that returns the name of the
> calling func/proc. To be used like as follows:
> 
> function query_db(sequence s)
>    if debug then
>       logfile( calling_proc() )
>    end if
>    ...
> end function
> 
> logfile() is my output function. In this case, I am using it for debugging
> purposes.
> 
> Instead of putting hundreds of lines of code (one for each function
> that would call query_db()), I'd rather just have the one line in query_db().
> That seems reasonable, doesn't it?
> 
> Is this possible, and how easy if so?

I would use this sort of stuff a lot for debugging. Not straightforward if you
want to avoid any impact at all on performance.

In il.e, you can see that routine names are kept in the stripped down symbol
table passed on to the backend. So there's a hope.

The SymTab index of the callee is computed when calling a routine, but not kept
(sub is a local var in the big switch routine in be_execute.c).

You could get a full image of the call stack by keeping a "sequence", which
doesn't exist in C, of indexes of called routines, and your routine would query
the symtab image for routine names.

It is relatively easy to make the mod, except for the details of dynamically
growing the sequence of called symtabs when the preallocated quota is about to
overflow.
I'd suggest limiting such recording to sections bracketed by a with debug and
without debug directives, without debug being the default. In this case, the only
impact on performance with debug off would be testing a flag to decide whether to
record subroutine symtab indexes or not, on every routine call.

CChris

4. Re: Calling Func/Proc

• Posted by CChris <christian.cuvier at agriculture.gou?.fr> Jul 11, 2007
• 659 views
• Last edited Jul 12, 2007

c.k.lester wrote:
> 
> CChris wrote:
> > 
> > c.k.lester wrote:
> > > I would really like to have a function that returns the name of the
> > > calling func/proc.
> > The SymTab index of the callee is computed when calling a routine, but not
> > kept
> > (sub is a local var in the big switch routine in be_execute.c).
> 
> It doesn't seem to me like you'd need to store anything extra. Doesn't the
> interpreter already know what routine called another routine?
> 
> proc A calls
>   +- proc B which calls
>        +- proc C which calls
>             +- proc D
> 
> Is there a list kept like this by the interpreter? If so, then when I'm in
> proc D and want to know the parent, I just return parent_proc( current_proc
> ).
> 
> > You could get a full image of the call stack by keeping a "sequence", which
> > doesn't exist in C, of indexes of called routines, and your routine would
> > query
> > the symtab image for routine names.
> 
> I would be surprised if the interpreter couldn't already provide this
> information, quickly and easily.

You are right in the sense that indeed there's a C variable (search for
expr_top) which records both the calling routine and the calling point, if I got
it right. Hence all "dynamic growth details" are already taken care of.
I must look at this more carefully, but then there would be no performance issue
in having a function return the last caller name/the whole list.

As I said, I'd use it too.
Any objections? There must be some around <smile/>.
CChris

6. Re: Calling Func/Proc

• Posted by CChris <christian.cuvier at agriculture.gou?.fr> Jul 12, 2007
• 665 views

c.k.lester wrote:
> 
> CChris wrote:
> 
> > I must look at this more carefully, but then there would be no performance
> > issue
> > in having a function return the last caller name/the whole list.
> > 
> > As I said, I'd use it too.
> > Any objections? There must be some around <smile/>.
> 
> + No performance hit
> + Enhancement is beneficial to programmers
> 
> There better be no objections! :P
> 
> Do it! Do it! Do it! :)

I had some eyestrain problems lately, keeping coding at a minimum this week. 

Things look better now. I'd like to plug once for all a probable DC leak in
win32lib "Patch #67" so that I can send Derek (and the testers, remember?)what I
can consider beta-stage code for him to merge into his code base, ultimately
leading to a new public release of the lib. The consequence of the leak is the
slowing down of all graphic rendering.

When this is done, I'll look into the 
<prospective doc entry>
s=function get_callers(i) 

Returns a sequence of pairs of strings if /i/ is greater than or equal to 0.
The pairs are of the form {/routine name,file name/} and represent (part of) the
current list of callers of the routine being executed.

The first pair in the returned list, if any, shows the routine that directly
called the code being executed. The second pair shows the routine that called the
latter, and so on. If the list is empty, it means that your routine was called
from code outside of any routine, or that your statement is outside of any
routine itself.

If /i/ is greater than zero, the /i/ first calling routines are returned, or
less if there are less. If /i/ is 0, the whole list is returned. A value less
than 0 will trigger a run time error.
</prospective doc entry>

This introduces a new opcode in the IL spec, and causes many places to need
small modificattions (just like find_from()). As a result, it would appear in
v3.2 at the earliest if it ever does.

Any thoughts?
CChris

8. Re: Calling Func/Proc

• Posted by c.k.lester <euphoric at ckle?ter.com> Jul 12, 2007
• 639 views

Robert Craig wrote:
> 
> I don't think we want to start 
> adding features that are not supported by the Translator, 
> unless there is a very good reason for it.

I agree.

> Is it that inconvenient to just add a parameter to the routine,
> or set a global variable with the name of the calling routine?

It is inconvenient and would violate the DRY principle.
(http://en.wikipedia.org/wiki/Don't_repeat_yourself)

I would have to add parameters to hundreds of routines.

> Can we see some examples of where this would be useful. C.K.?

I have a procedure called err() that gets passed an error message. err() is
called from hundreds of functions and procedures all spread throughout my
code. I usually call it like in this small example:

function rev_string(sequence x)
   if length(x) = 0 then
      err("rev_string(): Must be at least 1 character in length."
   else
      if length(x) > 17 then
         err("rev_string(): String too long."
      else
         x = reverse(x)
      end if
   end if
   return x
end function

A "fixed" version would look like this:

function rev_string(sequence x)
   if length(x) = 0 then
      err("Must be at least 1 character in length."
   else
      if length(x) > 17 then
         err("String too long."
      else
         x = reverse(x)
      end if
   end if
   return x
end function

This is a very short example. When dealing with a much larger func/proc,
there may be many more calls to err. Sometimes I change the name of a
func/proc. When that happens, I also have to adjust all the calls to err.

You suggest something like this:

function my_super_function()
   CURR_ROUTINE = "my_super_function"
   --or
   set_routine_name( "my_super_function" )
   ...
end function

but that violates DRY and, again, would have to be done for hundreds of
func/proc in my code. It is unnecessary overhead to something that should
be quite simple to resolve.

What about if I want to trace the call back a few func/procs? It makes
manual tracking even more complicated and time consuming.

Say that a calls b calls c which calls my_super_function... To track the
progress, you need something like:

function my_super_function()
   add_routine_name( "my_super_function" )
   -- do your stuff
   remove_routine_name( "my_super_function" )
end function

I'm open to suggestions, but the fact is, the most efficient way to
do this tracking of func/proc hierarchy is to let the interpreter/translated
program handle it.

The reason I need it is primarily for error handling/messaging and debugging.

10. Re: Calling Func/Proc

• Posted by Bernie Ryan <xotron at bl?efrog.com> Jul 13, 2007
• 671 views

c.k.lester wrote:
> 
> I have a procedure called err() that gets passed an error message. err() is
> called from hundreds of functions and procedures all spread throughout my
> code. I usually call it like in this small example:


CK: A suggestion !

-- MastError sequence will contain a list of all routines
-- up to the failure plus the error message of where it failed.
-- This is dynamic so it any code changes need only be made
-- in a single location.
sequence MastError MastError = {}

procedure err() printf(1,"\n%s: %s\n",{MastError[$-1],MastError[$]}) end procedure

function rev_string(sequence x) MastError &= {"rev_string"} if length(x) = 0 then MastError &= {"Must be at least 1 character in length."} err() err() else if length(x) > 17 then MastError &= {"String too long."} err() else x = reverse(x) end if end if return x end function

</eucode

Bernie

My files in archive: WMOTOR, XMOTOR, W32ENGIN, MIXEDLIB, EU_ENGIN, WIN32ERU, WIN32API

Can be downloaded here: http://www.rapideuphoria.com/cgi-bin/asearch.exu?dos=on&win=on&lnx=on&gen=on&keywords=bernie+ryan }}}

12. Re: Calling Func/Proc

• Posted by c.k.lester <euphoric at ckleste?.com> Jul 13, 2007
• 639 views

Also, Bernie, with your method, you would want to remove the last bit off
MastError each time you exited a routine. So you would have to do:

function rev_string(sequence x)
MastError &= {"rev_string"}
if length(x) = 0 then
  MastError &= {"Must be at least 1 character in length."} err()
  err()
else
  if length(x) > 17 then
    MastError &= {"String too long."} err()
  else
    x = reverse(x)
  end if
end if
MastError = MastError[1..$-2] -- <-- if only one message was issued from this
routine
return x
end function

I have database processing functions that could potentially have
several levels of messages (more than just one message per routine),
so returning [$-1] and [$] from MastError would not suffice. I use
it for messaging, also, not just error messages. Things like status
notifications, etc.

14. Re: Calling Func/Proc

• Posted by Matt Lewis <matthewwalkerlewis at gmail?com> Jul 13, 2007
• 669 views

c.k.lester wrote:
> 
> I didn't understand Rob's post regarding it's non-doability in regard to the
> Translator, but I've got a moment now to read for comprehension. :)

The issue is that, in the interpreter, you have a compiled C program that
maintains the state of your euphoria application, including its own
call stack, which would be fairly trivial to see where you came from
(obviously, since you get this in ex.err).

However, once you translate, you now have a compiled C program, and the
interpreter no longer maintains all that information.  The stack is the
native stack, and while you could, in theory, walk the stack (like a 
debugger would do), you'd lose a lot of information (well, in theory,
you could read debugging symbols like a debugger would).

I guess there's a case to be made that this is a debugging tool (like trace),
and that some differences to the interpreter and translator are acceptable.
We could, in theory, maintain a custom call stack in the translator by 
automatically calling some library routine to maintain this stuff for us,
at least within "with trace" code.

It's not a trivial thing to do, but might be possible, and be faithful to
the interpreter/translator compatibility goal.

Matt

16. Re: Calling Func/Proc

• Posted by c.k.lester <euphoric at cklester.?om> Jul 13, 2007
• 644 views

Robert Craig wrote:
> 
> c.k.lester wrote:
> > The reason I need it is primarily for error handling/messaging and
> > debugging.
> Have you considered using crash_routine()?

Well, I should have written, "error handling, messaging, and debugging."

That is, I gather up a bunch of messages to display on the page. They could
be standard messages like, "The operation was completed." The err() messages
are usually something like, "Could not find that record." Crash messages
aren't part of this vision, but you're right that I should redirect on a
crash. :)

I often write these things to a log, as well.

18. Re: Calling Func/Proc

• Posted by c.k.lester <euphoric at cklester.c?m> Jul 13, 2007
• 633 views

Bernie Ryan wrote:
> 
> I think you should take a new look at your code because I
> think you are confusing when you should be performing error
> checking and when you should be performing debugging.

Yeah, I wasn't clear when stating my case. Many times, more often than not,
the messages I'm looking for aren't crash messages. They are either status
messages ("The operation completed.") or other error messages ("Could not
open the database.").

In these cases, I'm not dealing with bugs.