[Tcl-bugs] [ tcl-Bugs-582522 ] aliases do not fire exec traces

[<no...@so...>]

Bugs item #582522, was opened at 2002-07-16 19:07
You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=110894&aid=582522&group_id=10894

Category: 43. Parsing and Eval
Group: 8.4b2
Status: Open
Resolution: Fixed
Priority: 3
Submitted By: Don Porter (dgp)
Assigned to: miguel sofer (msofer)
Summary: aliases do not fire exec traces

Initial Comment:
% proc print args {puts "printtrace $args"}
% proc foo {} {puts orig_foo}
% interp alias {} al_foo {} foo
al_foo
% trace add execution foo enter {print foo}
% trace add execution al_foo enter {print al_foo}
% foo
printtrace foo foo enter
orig_foo
% al_foo
printtrace al_foo al_foo enter
orig_foo
% 


The [print foo] trace is missing.  It should
have been fired by [al_foo].

The problem seems to be that execution
traces are performed by
TclEvalObjvInternal() and aliases get
evaluated by TclObjInvoke() which
bypasses that and executes the
command procedure directly.

In combination with Bug 582506, it
seems the correct solution may be
to tie execution traces into the
command procedure itself instead
of in TclEvalObjvInternal().



----------------------------------------------------------------------

>Comment By: miguel sofer (msofer)
Date: 2002-07-29 10:53

Message:
Logged In: YES 
user_id=148712

I'm pretty sure safety has been maintained: I only modified
AliasObjCmd which processes aliases and never sent the flag
TCL_INVOKE_HIDDEN to TclObjInvoke. 
TEOVI looks for commands using Tcl_GetCommandFromObj(), so
essentially using a cached value if (available && valid),
and looking up by name otherwise. So the issue is: could the
cmdName object have a cached reference to a hidden command?
I'll build extra safety and make the Alias structure keep a
private copy of the targetCmdName object, so that it never
keeps a foreign cached value.

BTW, traces on hidden commands are still not being fired ...

----------------------------------------------------------------------

Comment By: Don Porter (dgp)
Date: 2002-07-29 02:03

Message:
Logged In: YES 
user_id=80530


I haven't reviewed the fix, but be sure to
check it carefully that it preserves the
same semantics.  The difference between
"invoke" and "eval" is very important,
especially to be sure that aliases in
safe interps don't "leak safety".

----------------------------------------------------------------------

Comment By: miguel sofer (msofer)
Date: 2002-07-28 21:17

Message:
Logged In: YES 
user_id=148712

Fixed by sending aliases through TEOI. Leaving openas
reminder to check into the other suggestion. An alternative
would be to eliminate the independent import mechanism, and
just use it as a front-end for aliases which are now almost
as fast.

----------------------------------------------------------------------

You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=110894&aid=582522&group_id=10894