The offending commit is 01ed22489 ("A check to avoid multiple entries for the same node into the .save list", 2021-10-20).
I do not yet understand why.
In addition it is necessary to add model->BJTintCollResistGiven = TRUE;in bjtsetup.c to avoid the crash:
Good point, perhaps we have a similar problem in other devices too.
E.g. the serial resistance in the diode is also set to a certain value and later we relay in diotemp.c to DIOresistGiven. Here we will not fabricate a crash but the temperature dependency for the resistor is not handled for the pathological case of non given resistor and given temp coeff's.
I will check other devices like vdmos and vbic too.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Adding model->BJTintCollResistGiven = TRUE; in bjtsetup.c unfortunately will lead to several problems in our paranoia suite, 100W.cir woill fail completely.
With current pre-master-45 it is indeed not required for running the diffpair.cir successfully with TF, TH, and sens!
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
A fix to re-allow running diffpair.cir is now on git branch pre-master-45.
The 'make check' for the .TF still will deliver an output showing the deviations of the newly resulting output in comparison to the (old) diffpair.out. Fixinf this will require some more activity, as the code in the test/sensitivity is 20 years old, and we indeed in the meantime have some changes to the output. And make check does only do a simple text comparison.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
34 does not work, issue introduced around version 30?
ngspice-32 is o.k., with some warnings.
ngspice-33 - 35 segfault.
Since ngspice-36 the simulator shows the behaviour as described above.
After removing the .tf analyses, .sens is simulated.
Last edit: Holger Vogt 2025-07-23
The offending commit is
01ed22489 ("A check to avoid multiple entries for the same node into the .save list", 2021-10-20).I do not yet understand why.
In addition it is necessary to add
model->BJTintCollResistGiven = TRUE;in bjtsetup.c to avoid the crash:Good point, perhaps we have a similar problem in other devices too.
E.g. the serial resistance in the diode is also set to a certain value and later we relay in diotemp.c to DIOresistGiven. Here we will not fabricate a crash but the temperature dependency for the resistor is not handled for the pathological case of non given resistor and given temp coeff's.
I will check other devices like vdmos and vbic too.
Adding
model->BJTintCollResistGiven = TRUE;in bjtsetup.c unfortunately will lead to several problems in our paranoia suite, 100W.cir woill fail completely.With current pre-master-45 it is indeed not required for running the diffpair.cir successfully with TF, TH, and sens!
A fix to re-allow running diffpair.cir is now on git branch pre-master-45.
The 'make check' for the .TF still will deliver an output showing the deviations of the newly resulting output in comparison to the (old) diffpair.out. Fixinf this will require some more activity, as the code in the test/sensitivity is 20 years old, and we indeed in the meantime have some changes to the output. And make check does only do a simple text comparison.