Menu
▾
▴
#4635 Tcl_(Find|Create)HashEntry stub entries can never be called
From tcl.h we can see that Tcl_(Find|Create)HashEntry
is a macro, but it is in the stub table as well. The macro
overrides the definition in tclDecls.h:
=========================================
#undef Tcl_FindHashEntry
#define Tcl_FindHashEntry(tablePtr, key) ...
#undef Tcl_CreateHashEntry
#define Tcl_CreateHashEntry(tablePtr, key, newPtr) ...
=========================================
So, the stub table entries are useless. The only effects
it has is:
- Need tricks like #undef, to make tclHash.c and
tclInitStubs.c compile.
- Tcl_(Find|Create)HashEntry could be static in tclHash.c,
but only for the stub table they are made EXTERN
Solution: remove those two functions from the stub
table, and we can remove all remaining trickery.
Another benifit: Making Tcl_(Find|Create)HashEntry
static means that the dynamic loader has two less
symbols which need to be resolved.
Just leave it alone, Jan. You're not the area maintainer.
OK, Miguel?
Why priority 9? If it's only because you want
this to be on the top of your list, OK!
Besides that, it's only about the removal
of dead code, so I don't see the need
for prio 9.
The stub table entries were added:
2000-07-21 Eric Melski <ericm@ajubasolutions.com>
* generic/tclStubInit.c:
* generic/tclObj.c:
* generic/tclInt.h:
* generic/tclHash.c:
* generic/tclDecls.h:
* generic/tcl.h:
* generic/tcl.decls:
* doc/Hash.3: Reapplied patch from Paul Duffin to extend hash tables
to allow custom key types, such as Tcl_Obj *'s, and others.
To reclaim that patch,
cvs -D 2000-07-21 -D 2000-07-22
in tcl/generic
Aim of the patch was to move toward a
revised hash table implementation. The change
was a binary incompat, though, so it was disabled
with #ifdef TCL_PRESERVE_BINARY_COMPATIBILITY.
Sometime in Tcl 8.5 development, the new maintainers
concluded that preserving binary compat was one of
the things Tcl 8.* development was commited to, so
dragging around code never to be enabled was no
longer worth doing.
So now we still have stub table entries that support
only a configuration we removed in the last minor
release. Seems like they can go away.
Probably best we make the change on in 8.6 development
and delay or forgo any changes to tcl.decls in 8.5.*
There is another option: We could keep the stub
table entries, but make the signature match the
doc again. That would be OK to me too.
fix signatures
Here is a better patch, which does the same as the
previous one, but restores the two functions having
the same signature and behavior as the macros.
Still, the stub entries are useless, but at least there
is an option now to remove the macros and still
having everything work as before.
Backported to 8.5: Now the Tcl_(Find|Create)HashEntry functions
do exactly the same as the Tcl_(Find|Create)HashEntry macro's, so
at least the confusion addressed in this issue is gone.