New FrontEnd

[Per Östlund]

This ticket is tracking the progress on the the new frontend. Some documentation on the new frontend development is available on wiki:NewFrontEnd.

The new frontend development is almost complete. It is used by default by OMEdit (with the old frontend as fallback in case of problems) since version 1.14.0, and already delivers better performance than the old frontend in almost all cases. It will be the default one also for the command line compiler from version 1.17.0.

All 425 models of the MSL are handled successfully by the new front end, see ​report. There are a handful of models which are not handled correctly yet, but that is not due to the new frontend, see #5288.

The coverage with some other selected open-source library is reported in these diagrams. Note that until 14 Dec 2020 the libraries were tested with the old frontend:

This is a list of all tickets opened for the new frontend (New Instantiation module). Note that tickets with id > 4000 were opened specifically during the development of the new front end, while older ones refer to old issues with the old front end that are expected to be fixed by the new one.

Number of tickets: 57

Milestone: 1.19.0 (2 matches)

Ticket	Summary	Status	Owner	Type	Priority	Version
#2447	Let OM automatically load packages as specified in import statements	reopened	Per Östlund	enhancement	high	trunk
#6126	Error while compiling wind turbine model from PowerSystems library	assigned	Per Östlund	defect	blocker
Milestone: 2.0.0 (22 matches)
Ticket	Summary	Status	Owner	Type	Priority	Version
#2385	Missing variable declarations in instantiated model when using expandable connectors	reopened	Adrian Pop	defect	high	trunk
#2387	Operator overloading fails for matrices	assigned	Per Östlund	defect	high	trunk
#2484	Expandable connector resolution misses components	assigned	Per Östlund	defect	blocker	trunk
#3186	Partial binding of record sub-components in functions are lost	assigned	Per Östlund	defect	critical	trunk
#3221	Equation count of models containing expandable connectors	assigned	Mahder Alemseged Gebremedhin	defect	normal	trunk
#3977	Better diagnostics for non-balanced models	new	Lennart Ochel	enhancement	blocker
#4120	Out-of-bounds array elements undetected by front-end	assigned	Per Östlund	defect	blocker
#4442	Incorrect evaluation of inner/outer and record dependencies	assigned	Per Östlund	defect	high	v1.11.0
#4496	Complex 'sum' does not work	assigned	Per Östlund	defect	blocker
#4515	Automatic inference of nominal attributes for scaling	new	Per Östlund	discussion	high
#4960	Evaluate all parameters except some	new	Per Östlund	enhancement	high
#5061	[NF] Are public parameters allowed in functions by the Modelica spec?	new	Per Östlund	defect	high
#5240	NFEvalFunction.evaluateExternal is not yet implemented	reopened	Per Östlund	defect	blocker
#5597	Two examples of the new Modelica.Electrical.Batteries package do not work with the new frontend	new	Per Östlund	defect	blocker	v1.14.0-dev-nightly
#5605	Parameter with only start value not accepted	new	Per Östlund	defect	blocker
#5837	Unit checking is not performed on operator record equations	new	Per Östlund	defect	blocker
#5847	Constant is used without having been given a value	new	Per Östlund	defect	blocker	v1.16.0-dev
#5866	Update Save Total to NF	new	Per Östlund	defect	blocker
#6047	The NF incorrectly infers the variability of a function call in Modelica.Media	new	Per Östlund	defect	blocker
#6114	Unit checking scales with O(2^N) in case of large expressions involving +-*/ operators	assigned	Per Östlund	defect	normal	v1.16.0-dev
#6149	The NF accepts illegal type declarations?	new	Per Östlund	defect	blocker
#6383	Omc doesn't check correctly the constraining-clause	new	Per Östlund	defect	blocker	1.18.0-dev
Milestone: 2.1.0 (14 matches)
Ticket	Summary	Status	Owner	Type	Priority	Version
#1080	Constraining type on redeclarations (from MathCore)	new	krsta	defect	normal	trunk
#1842	Illegal redeclare should be rejected by the NF with appropriate error message	assigned	Per Östlund	defect	high
#2010	Prefix rules needs more work in new instantiation	new	Per Östlund	defect	high	trunk
#2153	Better analysis for function completion	new	Per Östlund	defect	high	trunk
#2276	detecting size mismatch for output arguments in functions	reopened	somebody	defect	normal	trunk
#3121	Missing bindings in flattened operator overloading example	assigned	Per Östlund	defect	low	trunk
#4843	NF doesn't evaluate size expression of function input parameters	new	Per Östlund	defect	normal	v1.13.0-dev-nightly
#4901	The NF does not check if redeclared functions are legal	new	Per Östlund	defect	high
#4905	Restriction on input connection is not enforced by OMC	assigned	Per Östlund	defect	high
#5178	Incorrect Warning - not balanced connector	reopened	Per Östlund	defect	normal
#5334	Check for wrong annotations	new	Per Östlund	enhancement	high
#5545	Improve diagnostics for wrong connectors	new	Per Östlund	enhancement	high
#5667	Arrays of records insufficiently type checked	new	Per Östlund	defect	high	v1.14.0-dev-nightly
#6163	Constrainedby clause not verified by the NF	new	Per Östlund	defect	high
Milestone: Future (6 matches)
Ticket	Summary	Status	Owner	Type	Priority	Version
#1747	Implement checks for all connection rules in chapter 9	new	Per Östlund	task	low
#2367	Record with vectors, zeros(), size()	accepted	Adrian Pop	defect	high	trunk
#4648	Add support for annotation for turning off replaceable check	new	Per Östlund	enhancement	normal	v1.13.0-dev-nightly
#4908	[NF] Issues with inner/outer with redeclare	new	Per Östlund	defect	high
#5579	New Frontend does not check for circular dependencies in parameter bindings	new	Per Östlund	defect	high
#5688	Use of removed conditional component not reported	new	Per Östlund	defect	normal
Milestone: None (13 matches)
Ticket	Summary	Status	Owner	Type	Priority	Version
#1939	Duplicate class detection not working when loading MSL	reopened	Per Östlund	defect	normal	trunk
#4526	Propagate displayUnits on simple equations	new	Mahder Alemseged Gebremedhin	enhancement	high
#4603	Provide support for start-value selection according to the Modelica Specification	assigned	Karim Adbdelhak	defect	high
#5474	[NF] Lookup of inner class fails	new	Per Östlund	defect	high
#5598	redeclaration of input as output with NF	assigned	Per Östlund	discussion	normal	v1.14.0-dev-nightly
#5730	Issue with operator record constructors in NF	new	Per Östlund	defect	high
#5852	getCommandLineOptions fails with newInst	new	Per Östlund	defect	normal	v1.16.0-dev
#6213	Lookup in wrong scope when redeclaring function field in a function	accepted	Per Östlund	defect	high	v1.17.0-dev
#6293	Function non-inputs are initialized in the wrong order	new	Per Östlund	defect	low	v1.17.0-dev
#6325	Support for ExternData	assigned	Per Östlund	defect	high	1.16.0
#6373	Electrical Quasistationary MutualInductor in standard library not working in OM	assigned	Per Östlund	defect	critical	1.16.2
#6390	Cannot extend from RealOutput	new	Per Östlund	defect	low	1.16.2
#6423	Optionally list all empty connection sets	new	Per Östlund	enhancement	high	1.16.2

[Francesco Casella]

Still some problems with component modifiers. Running the following test script

loadString("
package P
  model A
    parameter Real p;
  end A;
  model B
    parameter Real q = 2;
    A a1(p = q);
  end B;
  model C
    parameter Real q = 2;
    A a1(p = q);
    A a2(p = q);
  end C;
end P;
");
setCommandLineOptions("-d=newInst");
instantiateModel(P.B);
instantiateModel(P.C);

returns

true
true
"class B
  parameter Real q = 2;
  parameter Real a1.p = q;
end B;
"
""

so for some reason model C is not instantiated at all.

[Per Östlund]

Replying to casella:

Still some problems with component modifiers.

I am well aware of that, and have already fixed this particular issue locally a while ago. At the moment I'm not really interested in bug reports though, the new instantiation is not ready for public testing yet.

[Francesco Casella]

Replying to perost:

At the moment I'm not really interested in bug reports though, the new instantiation is not ready for public testing yet.

Q1: Can you give a reasonable estimate of when this will be possible?

Q2: Is the interface towards the other tasks of Phase 2 already stabilized? If not, when will it reasonably be?

[Per Östlund]

Replying to casella:

Q1: Can you give a reasonable estimate of when this will be possible?

No

Q2: Is the interface towards the other tasks of Phase 2 already stabilized? If not, when will it reasonably be?

Maybe

[Per Östlund]

Class modifiers have been implemented now, but needs more testing.

[Francesco Casella]

The test case reported in comment:4 now produces the expected result

[Francesco Casella]

However, instantiating P.S in the attached test case still fails

[Per Östlund]

Replying to casella:

However, instantiating P.S in the attached test case still fails

It works fine for me, are you using the latest nightly build?

[Francesco Casella]

Replying to perost:

Replying to casella:

However, instantiating P.S in the attached test case still fails

It works fine for me, are you using the latest nightly build?

I thought so, but I wasn't. With OpenModelica-v1.11.0-dev-288-g2ccad20-64bit, it works fine for me also :)

[Per Östlund]

Initial support for enumerations is now implemented.

[Francesco Casella]

Added link to wiki page

[Adrian Pop]

Small status update.

[Francesco Casella]

I just tried out ScalableTestSuite.Elementary.SimpleODE.Models.CascadedFirstOrder:

model CascadedFirstOrder
  parameter Integer N = 10 "Order of the system";
  parameter Modelica.SIunits.Time T = 1 "System delay";
  final parameter Modelica.SIunits.Time tau = T/N "Individual time constant";
  Real x[N] (each start = 0, each fixed = true);
equation
  tau*der(x[1]) = 1 - x[1];
  for i in 2:N loop
    tau*der(x[i]) = x[i-1] - x[i];
  end for;
end CascadedFirstOrder;

but it still doesn't work, I get

Error: Type mismatch for positional argument 1 in der(x=x[1]).
The argument has type: Real[10]
expected type: Real

This is a very basic model, and it doesn't use arrays in any fancy way. This error looks to me more like a bug than a missing feature.

Would you mind fixing this issue so we can start getting some results from testing the ScalableTestSuite with the new front-end?

Thanks

Francesco

[Per Östlund]

Replying to casella:

This is a very basic model, and it doesn't use arrays in any fancy way. This error looks to me more like a bug than a missing feature.

Subscripting is not implemented yet, so it is actually a missing feature. It's high up on my todo list, but other things currently has higher priority for me.

[Francesco Casella]

OK. I interpreted "Subscripting needs to be improved" as "there is already some subscripting implemented" :)

[Per Östlund]

Updated description for redeclares (better in general, but class extends/element redeclares currently broken) and imports.

[Per Östlund]

Replying to casella:

Would you mind fixing this issue so we can start getting some results from testing the ScalableTestSuite with the new front-end?

This model should start working in about five minutes, once Hudson pulls in my latest changes.

[Francesco Casella]

Very good!

[Per Östlund]

Move all tickets from NF - New Frontend to New Instantiation so we don't have two different categories for the same thing.

[Adrian Pop]

Overconstrained connection graph is now implemented via #4966.

[Francesco Casella]

Status update as of 24 Sep 2020.

The new front end development continues, the functionality of this module has now overcome the old front end, which is now starting to become obsolete.

The new front end is already the default choice in OMEdit since version 1.14.0, with automatic fallback to the old frontend in case of problems. The command line version of the compiler still uses the old frontend as default, and requires the --newInst flag to use the new one. The plan is to make the new front end the default choice also from the command line starting with version 1.17.0, due December 2020.

The fallback option will probably be removed from OMEdit at that point, because it will mostly be advising users to try the old front end in response to models which are in fact broken. Users that will still need to use the old frontend with their models will be able to do that by setting the option in the simulation settings window. However, please note that the old frontend has no longer been maintained since 2019.

[Francesco Casella]

Status update as of Dec 21 2020.

The new frontend was made the default on 14 Dec 2020. There are several regressions w.r.t. the old frontend on older libraries, because they contain invalid Modelica code. Those will not be addressed. Regressions in the latest versions of currently maintained libraries should be fixed in time for the 1.17.0 release.

[Francesco Casella]

This ​regression report shows the difference between the old and the new frontend.

There are 1000 improvementes in the library testsuite, due to the better capability of the new frontend to handle complex Modelica models, and also 1432 performance improvements, due to the much faster operation of the new front end compared to the old one.

There are 673 regressions, mostly in older libraries that contain illegal Modelica code that the new front end rejects because of its stricter policy of conformance to the Modelica specification.

[Francesco Casella]

The new front end becomes the default one with release 1.17.0, -d=newInst is no longer necessary to use it. All test reports now use it, except the oldInst branch.

The development continues, but the transition phase from the old front-end to the new one is now complete, regarding the compilation of simulation models.

The old front end is still used for bootstrapping and for most API functions. Regarding the former, there is currently no urgent need to supersede it. Regarding the latter, the API functions should gradually be switched to use the old one, for much faster performance. This has already been done in some critical cases, the work will continue there.

With release 1.17.0, we can close this ticket for good.