qucs 0.0.21-dev, Parameter Sweep, bad netlist

[GerryHickman]

In the schematic, create Parameter Sweep
In the GUI Properties dialog box, it looks like this

Simulation 'AC1', Sweep Parameter 'Ang', Type 'linear', Start '20', Stop '230', Step '21', Number '11'

BUT

in the netlist, it looks like this

.SW:SW1 Sim="AC1" Type="lin" Start="Ang" Stop="20" Points="230" Points="11"

The 'Param' is missing, and Start is called 'Ang', and there are two "Points"
Looks like positional values are "off by one"

If I manually edit the netlist, and use command line to run the simulation, everything starts working; back in the GUI I can switch to the *.dpl file and see my graphs.

I'm fairly sure this used to work in qucs 0.0.19

[felix-salfelder]

On Mon, Dec 22, 2025 at 02:44:20PM -0800, GerryHickman wrote: The 'Param' is missing, and Start is called 'Ang', and there are two "Points" Looks like positional values are "off by one"

Thank you Gerry. I can confirm there is a bug, but not the exact same. In our test suite $ grep .SW tests/AC_SW_resonance.prj/resonance.sch <.SW SW1 1 440 200 0 51 0 0 "AC1" 1 "log" 1 "R_par" 1 "1 Ohm" 1 "20 Ohm" 1 "6" 1> $ grep .SW tests/ref/AC_SW_resonance.prj_resonance.sch.net (*) .SW:SW1 Sim="AC1" Type="log" Start="R_par" Stop="1 Ohm" Points="20 Ohm" Points="6" $ grep .SW tests/ref/AC_SW_resonance.prj_resonance.sch.vs (* .. *) \.SW #(.Sim(AC1),.Type(log),.Start(R_par),.Stop(1 Ohm),.Points(20 Ohm),.Points(6)) SW1 ( ); Curiously the qucsator netlist (*) looks right, and the Verilog schematic dump is still wrong. This needs some analysis.

I'm fairly sure this used to work in qucs 0.0.19

There has been a lot of work going on. The integration of the tests is quite recent..

[GerryHickman]

Curiously the qucsator netlist (*) looks right

You mean the one with the asterisk?
It doesn't look right to me, e.g. Start="R_Par"
It looks broken in the same way as my example above. It's even got two "Points"

[felix-salfelder]

On Tue, Dec 23, 2025 at 06:50:00AM -0800, GerryHickman wrote: It looks broken in the same way as my example above. It's even got two "Points"

Well, you are perfectly right. Not sure what I was thinking. In fact, both .net and .vs have the same problem. (sorry for the noise.)

[GerryHickman]

OK, good news that this is confirmed.

Which function calls do I need to look at to see where it's going wrong?

[felix-salfelder]

On Tue, Dec 23, 2025 at 07:47:35AM -0800, GerryHickman wrote: OK, good news that this is confirmed. Which function calls do I need to look at to see where it's going wrong?

print_args (see lang_verilog.cc) calls "param_name" and "param_value". I suspect that these are not implemented correctly in the sweep "component". qucs/components/param_sweep.{cpp,h}?

[GerryHickman]

Some background on how it works, and where it goes wrong:

Debug: "main.cpp : main : calling constructor QucsApp(files)"
Debug: "qucs.cpp : QucsApp(files) constructor Enter"
Debug: "qucs.cpp : QucsApp(files) calling Module::registerModules()"
Debug: "module.cpp : Module::registerModules : Enter"
Debug: "module.cpp : Module::registerModules : About to register components using macros"
Debug: "schematic.cpp : Schematic : Enter constructor"
Debug: "qucs.cpp : gotoPage : doc was not found, trying to load"
Debug: "schematic_file.cpp : loadDocument : Enter"
Info DocName "/data/build/qucs/tests/AC_SW_resonance.prj/resonance.sch"
Info schematic_file_read.cpp : readLegacy
Debug: "schematic_file_read.cpp : loading properties"
Debug: "schematic_file_read.cpp : loading components"
Debug: "schematic_file.cpp : loadComponents : calling getComponentFromName"
Debug: "components/component.cpp : calling module getComponent()" ".SW"
Debug: "components/param_sweep.cpp : Param_Sweep::Param_Sweep() constructor : Enter"
Debug: "components/component.cpp : got c with param_count : " 6
Debug: "components/component.cpp : calling p->loadComponent"
Debug: "components/component.cpp : loadComponent : from text : _s" "<.SW SW1 1 440 200 0 51 0 0 \"AC1\" 1 \"log\" 1 \"R_par\" 1 \"1 Ohm\" 1 \"20 Ohm\" 1 \"6\" 1>"
Debug: "components/component.cpp : loadComponent : Model : " ".SW"
Debug: "components/component.cpp : loadComponent : quotes count" 12 // param_count * 2
Debug: "components/component.cpp : loadComponent : tmp set to" 13   // quotes count plus one
Debug: "components/component.cpp : loadComponent : Prop (default) :" "Sim" "" "simulation to perform parameter sweep on"
Debug: "components/component.cpp : loadComponent : setting prop value : n : " "AC1"
	set other props
Debug: "components/component.cpp : calling c->recreate(0)"
Debug: "components/param_sweep.cpp : Param_Sweep::recreate(Schematic*) : Enter" // BREAKS HERE
Debug: "schematic_file.cpp : loadComponents : calling simpleInsertComponent(c)" // 'c' is now corrupt

qucs registers the components with default property values, the schematic is then read and it's components are loaded; default property values are overwritten with values from the schematic. At this point, the data is correct, but then it calls 'recreate' on the component. The param_sweep component has a 'recreate' override, and this corrupts the property names by overwriting them with hard coded (wrong) names. The fix I tried was just to simply comment out this block because the names are already correct.

qucs/components/param_sweep.cpp
void Param_Sweep::recreate(Schematic*)
remove the 'else' block, or comment out all within

After this change, the netlist becomes valid and the parameter sweep starts working.

Two other things to note:

The test is misleading, it checks the netlist in the 'ref' directory, but the netlist is wrong, so the test passes!
tests/ref/AC_SW_resonance.prj_resonance.sch.net

When you run
qucs -n -i tests/AC_SW_resonance.prj/resonance.sch -o ~/resonance.net
it actually does everything twice, once for the -i and once for the -n
it basically loads the schematic into a "hidden" GUI, and then later it loads it again to get the netlist, not sure if that's needed.

[felix-salfelder]

On Tue, Dec 30, 2025 at 11:40:25AM -0800, GerryHickman wrote: [..] but then it calls 'recreate' on the component. The param_sweep component has a 'recreate' override, and this corrupts the property names by overwriting them with hard coded (wrong) names.

Any idea what recreate is supposed to do? Perhaps I never fully understood. It's plausible that removing it will fix this one.

The test is misleading, it checks the netlist in the 'ref' directory, but the netlist is wrong, so the test passes! tests/ref/AC_SW_resonance.prj_resonance.sch.net

The tests have been added in a rush. Think of the refs as "this is what the code produces". I have checked critical and relevant bits, but not all of it. If you patch the code, also patch the refs. This is needed for review, and to exclude side effects. I will try to see if the ref change looks right, and then I'll try to build and reproduce it. Thanks felix

[GerryHickman]

Any idea what recreate is supposed to do?

No, maybe it was important in the past, would be good to know the purpose of it.

[gucio321]

same issue here.
I'm experiencing same problem mentionede in the OP. Also the simulation does not run.
Happens on master but not on 0.0.20.
any chance this will be fixed anytime soon?

[felix-salfelder]

This one would require a patch. See CONTRIBUTE.

thanks

[gucio321]

The problem is, that this list

  Props.push_back(qucs::Property("Sim", "", true,
                QObject::tr("simulation to perform parameter sweep on")));
  Props.push_back(qucs::Property("Type", "lin", true,
                QObject::tr("sweep type")+" [lin, log, list, const]"));
  Props.push_back(qucs::Property("Param", "R1", true,
                QObject::tr("parameter to sweep")));
  Props.push_back(qucs::Property("Start", "5 Ohm", true,
                QObject::tr("start value for sweep")));
  Props.push_back(qucs::Property("Stop", "50 Ohm", true,
                QObject::tr("stop value for sweep")));
  Props.push_back(qucs::Property("Points", "20", true,
                QObject::tr("number of simulation steps")));

qucs/qucs/components/param_sweep.cpp

Line 43 in d4c4640

Props.push_back(qucs::Property("Sim", "", true,

which is then reconstructed in

qucs/qucs/mouseactions.cpp

Line 1191 in d4c4640

auto c = std::dynamic_pointer_cast<Component>(e);

somewhere in between those lines gets corrupted (for some reason) so that it looks as foollows:

Debug: Comp->Props : "Sim" "DC1" true "simulation to perform parameter sweep on"
Debug: Comp->Props : "Type" "lin" true "sweep type [lin, log, list, const]"
Debug: Comp->Props : "Start" "x" true "parameter to sweep"
Debug: Comp->Props : "Stop" "0 Ohm" true "start value for sweep"
Debug: Comp->Props : "Points" "100 Ohm" true "stop value for sweep"
Debug: Comp->Props : "Points" "101" true "number of simulation steps"

Hope this helps someone.
I'd really like to fix this problem but my C++ backend is not strong enough to understand the situation good enough.
I'd be really greatful if someone could fix this. I'm a student and am forced to use this programm during my classes. I really don't want to switch to windows and use an exe from sourceforge...

[felix-salfelder]

Debug: Comp->Props : "Start" "x" true "parameter to sweep"

looks corrupt to me. perhaps you can add an assertion that checks if prop[2].name is actually "Param", as it should?

.. especially to track down which mouseaction has screwed it up.

you can also "netlist" a qucs schematic without mouse actions, from command line. if this is buggy, too, it should be more straightforward to find the problem (and write a test that shows the problem).

[gucio321]

Thank you for your response.

I've switched to the qucs-s which seems to be less bugged than this project and also seems to be quite compatible.