Re: [TCLCORE] Tcl_InitStubs() thread-unsafe

[Vasiljevic Z. <zv...@ar...>]

On 02.04.2010, at 20:58, Kevin Kenny wrote:

> I'm disinclined to worry all that much.
>
> Tcl_InitStubs, in the current design, only, ever, either changes
> tclStubsPtr from NULL to the actual Stubs table pointer, or changes
> it from the actual Stubs table pointer to the *same* actual Stubs
> table pointer.

If this is so, then a simple and helgrind-savvy change
(if you ever happen to run helgrind on the code you
will understand what I mean) is to set tclStubsPtr only
if it is NULL. If it is already set, one can simply avoid
the change.

Net-effect is simpler more understandable code and absence
of tons of helgrind (a valgrind tool for detecting race-conditions)
warnings about race-condotions at that place.

The "logic" of setting the tclStubsPtr is really "convoluted".
I do not consider myself super intelligent, but I'm not dumb
either. From (lengthly) reading the (current) code I was not
able to understand what it does. And I do have some 15 years
Tcl (core) experience.

Why Tcl_PackageReqEx() returns a new copy of the value to be
set in (potentially initialized) tclStubsPtr?? I still did not
get the clue why this done. And reading the Tcl_PackageReqEx
comments does not bring any light into it.

The most trivial and simple change would really be:

  tclStubsPtr = &tclStubs;

as this is effectively what is being done after you walk
all the calls in the chain.

Cheers
Zoran