From Newsgroup: comp.lang.tcl


Hello,

I am experimenting with a compiled package for Tcl 9 (9.0.1 to be more precise) and I get an error message:

interpreter uses an incompatible stubs mechanism
while executing
"load ftcl.dll Ftcl"
(file "runpkg.tcl" line 4)

The call that causes this is Tcl_InitStubs( interp, "9.0", 0)

I have no idea what may be wrong, so any suggestions are appreciated.
(Note: the same code with a different import library of course works fine with Tcl 8.6)

Regards,

Arjen
--- Synchronet 3.21a-Linux NewsLink 1.2

From Newsgroup: comp.lang.tcl

Am 16.11.25 um 15:00 schrieb Arjen:

interpreter uses an incompatible stubs mechanism

https://core.tcl-lang.org/tcl/tktview/6f22c7a1fc
--- Synchronet 3.21a-Linux NewsLink 1.2

From Newsgroup: comp.lang.tcl


Christian Gollwitzer <auriocus@gmx.de> posted:

Am 16.11.25 um 15:00 schrieb Arjen:

interpreter uses an incompatible stubs mechanism

https://core.tcl-lang.org/tcl/tktview/6f22c7a1fc

Thanks, though I am not sure that applies in my case, as the code of my extension is Fortran-only ;) and therefore has nothing to do with a stray tcl.h file. Perhaps somehow the run-time library?
--- Synchronet 3.21a-Linux NewsLink 1.2

From Newsgroup: comp.lang.tcl


Arjen <user153@newsgrouper.org.invalid> posted:

Hello,

I am experimenting with a compiled package for Tcl 9 (9.0.1 to be more precise)
and I get an error message:

interpreter uses an incompatible stubs mechanism
while executing
"load ftcl.dll Ftcl"
(file "runpkg.tcl" line 4)

The call that causes this is Tcl_InitStubs( interp, "9.0", 0)

I have no idea what may be wrong, so any suggestions are appreciated.
(Note: the same code with a different import library of course works fine with
Tcl 8.6)

Regards,

Arjen

I overlooked the interface for the actual Tcl_InitStubs() function (there is an extra argument there). But a C version is not behaving correctly either. I will have to investigate this further.
--- Synchronet 3.21a-Linux NewsLink 1.2

From Newsgroup: comp.lang.tcl

Am 17.11.2025 um 13:04 schrieb Arjen:

Arjen <user153@newsgrouper.org.invalid> posted:

Hello,

I am experimenting with a compiled package for Tcl 9 (9.0.1 to be more precise)
and I get an error message:

interpreter uses an incompatible stubs mechanism
while executing
"load ftcl.dll Ftcl"
(file "runpkg.tcl" line 4)

The call that causes this is Tcl_InitStubs( interp, "9.0", 0)

I have no idea what may be wrong, so any suggestions are appreciated.
(Note: the same code with a different import library of course works fine with
Tcl 8.6)

Regards,

Arjen

I overlooked the interface for the actual Tcl_InitStubs() function (there is an
extra argument there). But a C version is not behaving correctly either. I will
have to investigate this further.

Yes, ping magic Jan, probably living next to you ;-).
There was a lot magic for tcl8.7 which is obsolate now.
And double compile tcl8.6/9.0 may also happen without knowing it...
Great features with pitfalls...

Take care,
Harald
--- Synchronet 3.21a-Linux NewsLink 1.2

From Newsgroup: comp.lang.tcl


Harald Oehlmann <wortkarg3@yahoo.com> posted:

Am 17.11.2025 um 13:04 schrieb Arjen:

Arjen <user153@newsgrouper.org.invalid> posted:

Hello,

I am experimenting with a compiled package for Tcl 9 (9.0.1 to be more precise)
and I get an error message:

interpreter uses an incompatible stubs mechanism
while executing
"load ftcl.dll Ftcl"
(file "runpkg.tcl" line 4)

The call that causes this is Tcl_InitStubs( interp, "9.0", 0)

I have no idea what may be wrong, so any suggestions are appreciated.
(Note: the same code with a different import library of course works fine with
Tcl 8.6)

Regards,

Arjen

I overlooked the interface for the actual Tcl_InitStubs() function (there is an
extra argument there). But a C version is not behaving correctly either. I will
have to investigate this further.

Yes, ping magic Jan, probably living next to you ;-).
There was a lot magic for tcl8.7 which is obsolate now.
And double compile tcl8.6/9.0 may also happen without knowing it...
Great features with pitfalls...

Take care,
Harald

Good suggestion, though he lives on the other side of the country.
Still, it is the same timezone :).

Regards,

Arjen
--- Synchronet 3.21a-Linux NewsLink 1.2

From Newsgroup: comp.lang.tcl


Arjen <user153@newsgrouper.org.invalid> posted:

Hello,

I am experimenting with a compiled package for Tcl 9 (9.0.1 to be more precise)
and I get an error message:

interpreter uses an incompatible stubs mechanism
while executing
"load ftcl.dll Ftcl"
(file "runpkg.tcl" line 4)

The call that causes this is Tcl_InitStubs( interp, "9.0", 0)

I have no idea what may be wrong, so any suggestions are appreciated.
(Note: the same code with a different import library of course works fine with
Tcl 8.6)

Regards,

Arjen

Hm, I just built Tcl 9.1a1 from source and tried again. Now it worked (at least
the simple C version). I have to figure out what the magic number is, since
on the Fortran side I will have to pass the extra argument directly, but it
is a start.
--- Synchronet 3.21a-Linux NewsLink 1.2

From Newsgroup: comp.lang.tcl


Arjen <user153@newsgrouper.org.invalid> posted:

Hm, I just built Tcl 9.1a1 from source and tried again. Now it worked (at least
the simple C version). I have to figure out what the magic number is, since on the Fortran side I will have to pass the extra argument directly, but it is a start.

Partial success: I printed the value of TCL_STUB_MAGIC as seen by the
C extension and inserted that as the last argument in the Fortran version. Now the error message is that the extension is built for 8.6 (while I clearly
used Tcl 9.1a1) ... Anyway, some progress.
--- Synchronet 3.21a-Linux NewsLink 1.2

From Newsgroup: comp.lang.tcl

* Arjen <user153@newsgrouper.org.invalid>
| Arjen <user153@newsgrouper.org.invalid> posted:
| > Hm, I just built Tcl 9.1a1 from source and tried again. Now it worked (at least
| > the simple C version). I have to figure out what the magic number is, since | > on the Fortran side I will have to pass the extra argument directly, but it
| > is a start.

| Partial success: I printed the value of TCL_STUB_MAGIC as seen by the
| C extension and inserted that as the last argument in the Fortran version. Now
| the error message is that the extension is built for 8.6 (while I clearly
| used Tcl 9.1a1) ... Anyway, some progress.

Could that be a byte-ordering problem? I seem to remember that we had
to byte-swap values when intermixing FORTRAN and C (but this was 30
years ago, so bear with me :-)

R'
--- Synchronet 3.21a-Linux NewsLink 1.2

From Newsgroup: comp.lang.tcl


Ralf Fassel <ralfixx@gmx.de> posted:

Could that be a byte-ordering problem? I seem to remember that we had
to byte-swap values when intermixing FORTRAN and C (but this was 30
years ago, so bear with me :-)

R'

More probable: swap array indices - Fortran uses column-major and C uses row-major to address elements in a multidimensional array. And then of course the start of the indices at 1 or 0.

Right now, my understanding of the C code fails with the bit operations
in place ... ;).
--- Synchronet 3.21a-Linux NewsLink 1.2

From Newsgroup: comp.lang.tcl


Arjen <user153@newsgrouper.org.invalid> posted:

Ralf Fassel <ralfixx@gmx.de> posted:

Could that be a byte-ordering problem? I seem to remember that we had
to byte-swap values when intermixing FORTRAN and C (but this was 30
years ago, so bear with me :-)

R'

More probable: swap array indices - Fortran uses column-major and C uses row-major to address elements in a multidimensional array. And then of course the start of the indices at 1 or 0.

Right now, my understanding of the C code fails with the bit operations
in place ... ;).

Got it! The problem was of course in a different place than where I was looking.
In the C code the function Tcl_InitStubs() is called via a macro and that manipulates the "exact" argument. My Fortran code was simply assuming the correct
value for "exact" was a logical (or Boolean if you insist) value 0 or 1.
The correct value to pass incorporates the major and minor version of Tcl and ends up in my case to be 67840. I could have guessed.

Regards,

Arjen
--- Synchronet 3.21a-Linux NewsLink 1.2

From Newsgroup: comp.lang.tcl


Arjen <user153@newsgrouper.org.invalid> posted:

Got it! The problem was of course in a different place than where I was looking.
In the C code the function Tcl_InitStubs() is called via a macro and that manipulates the "exact" argument. My Fortran code was simply assuming the correct
value for "exact" was a logical (or Boolean if you insist) value 0 or 1.
The correct value to pass incorporates the major and minor version of Tcl and
ends up in my case to be 67840. I could have guessed.

Now I have to adjust my code to make the command actually work.
--- Synchronet 3.21a-Linux NewsLink 1.2

From Newsgroup: comp.lang.tcl

<slap on forehead> Ah, TCL 9. I was wondering, since TCL-8 just has
if (exact)
in the stubs code.

* Arjen <user153@newsgrouper.org.invalid>
| Got it! The problem was of course in a different place than where I was looking.
| In the C code the function Tcl_InitStubs() is called via a macro and that
| manipulates the "exact" argument.

TCL 9 massages the passed-in value in tcl.h and adds some high-order
bits - this is what you need to do when calling Tcl_InitStubs from 'outside'.

# define Tcl_InitStubs(interp, version, exact) \
(Tcl_InitStubs)(interp, version, \
(exact)|(TCL_MAJOR_VERSION<<8)|(TCL_MINOR_VERSION<<16), \
TCL_STUB_MAGIC)

(or a derivate of it)

| My Fortran code was simply assuming the correct value for "exact" was
| a logical (or Boolean if you insist) value 0 or 1. The correct value
| to pass incorporates the major and minor version of Tcl and ends up in
| my case to be 67840. I could have guessed.

Note that the final value of 'exact' needs to have the 1-bit set for
"exact match required" (code tests for exact&1) , where tcl-8 accepted
anything != 0.

R'
--- Synchronet 3.21a-Linux NewsLink 1.2

From Newsgroup: comp.lang.tcl


Arjen <user153@newsgrouper.org.invalid> posted:

Arjen <user153@newsgrouper.org.invalid> posted:

Got it! The problem was of course in a different place than where I was looking.
In the C code the function Tcl_InitStubs() is called via a macro and that manipulates the "exact" argument. My Fortran code was simply assuming the correct
value for "exact" was a logical (or Boolean if you insist) value 0 or 1. The correct value to pass incorporates the major and minor version of Tcl and
ends up in my case to be 67840. I could have guessed.

Now I have to adjust my code to make the command actually work.

Well, that required a bit of work, indeed, due to the stubs table being different from Tcl 8.6, but it now works. It is just a proof of concept, but what I find attractive about it, is that involves no intermediate layer in C. The interfacing is purely in Fortran. I will probably not be able to do everything in
Fortran, because of the use of C macros, but for now it is a nice proof of concept.

Regards,

Arjen
--- Synchronet 3.21a-Linux NewsLink 1.2