Add parsing of piecewise linear curves. #1149

endJunction · 2016-04-19T11:04:37Z

Adds a _curves map from names to MathLib::PiecewiseLinearInterpolations to project data when

<curves>
  <curve>
    <name>curveA</name>
    <coords>0 1 1.5</coords>
    <values>10 11 12.3</values>
  </curve>
</curves>

is present in the project file.

I currently use this feature to create time-dependent boundary conditions (but this is too early for PR); and @Yonghui56 started work on Richard's flow, where the curves are used for parameters.

chleh · 2016-04-19T11:15:02Z

Applications/ApplicationsLib/ProjectData.cpp

+	}
+
+	return std::unique_ptr<MathLib::PiecewiseLinearInterpolation>{
+	    new MathLib::PiecewiseLinearInterpolation{coords, values}};


You could move the vectors.

The PiecewiseLinearInterpolation ctors need to be modified. There are three possibilities:

Duplicate code and replace const& parameters with && and std::move in initializer list.

Use universal reference, i.e. template <typename T> PiecewiseLI(T&& coords, T&& values) : _coords(std::forward<T>(coords)), ....

Pass-by-value and std::move in initializer list.

I suggest to take the third option, because the vector's move operation is cheap and it covers the two cases of passed l-values and r-values w/o code duplication and templates.

cf. to "Effective modern c++" Item 41: "Consider pass by value for copyable parameters that are cheap to move and always copied."

Fourth possibility is to implement ctor taking r-values only. Currently there are any calls with l-values.

cf. to "Effective modern c++" Item 41: "Consider pass by value for copyable parameters that are cheap to move and always copied."

But we do not always copy the vectors. In fact, they are only copied in the unit tests, aren't they?

My opinion:

IMHO the best option

Problematic. Maybe then different calls of std::vector's ctor other than copy/move could be generated implicitly, e.g., vector(std::initializer_list), though I'm not sure about it. Anyway in this case the avoided code duplication is neglectable. Furthermore we don't do such things in any other place, do we?

There will always be a copy and we rely on the compilers optimizer to perform the copy elision.

There are have several other places where we've only implemented r-value reference arguments, so it wouldn't hurt (unless you can give an example that hurts 😄).

Anyway, that's only a minor issue. Don't waste time on it. If you can't decide on a specific solution, please just leave a TODO comment in the code, so we can find the spot later on. I'd vote for option 1 or 4, choose either one of them, if you like.

We always copy the vectors in PiecewiseLinearInterpolation ctor. Before this PR copy is via copy, with this PR copy is via move if r-value is passed.

So if I get you right, "copy" is everything except storing a reference to an external vector as a data member of the class?
Still I'd like to know the opinion of the others about that "paradigm shift". Maybe I'm the only one who is concerned.

chleh · 2016-04-19T11:22:01Z

Only minor comments 👍

chleh · 2016-04-19T13:08:28Z

MathLib/InterpolationAlgorithms/PiecewiseLinearInterpolation.cpp

-                                                           bool supp_pnts_sorted) :
-	_supp_pnts(supporting_points), _values_at_supp_pnts(values_at_supp_pnts)
+PiecewiseLinearInterpolation::PiecewiseLinearInterpolation(
+    std::vector<double> supporting_points,


I don't like that solution, because it contradicts our current paradigm. Probably the code works: Move constructor called twice. Still, we should be extremely sure that we know by heart what we are doing here, and that it works as a general approach. It probably does, if Scott Meyers suggests it. But passing by value/ref/ref&& is already tricky enough, and this technique adds another facet. Will we remember in half a year, why we pass by value here? Can we teach others, why that's good? Will those taught be confused and then pass everything by value?

Or a different point: Do we know by heart whether void f(std::vector<int>&& v) {} can be called with an l-value reference? And why it is like that?–You see, it's complicated 😄.

wenqing · 2016-04-19T16:06:28Z

Just one comment for future development: I think we need a class for curves (or functions), which can provide the value and also possibly the tangential of the function at a given variable.

endJunction · 2016-04-20T12:32:15Z

After discussion with Tom, we move to r-values constructor and drop the const& one.

TomFischer · 2016-04-21T05:53:41Z

@chleh Are your minor comments answered? If yes, from my point of view you could set the "rebase and merge" flag.

Reasoning: The vector's move operation is cheap and it covers the two cases of passed l-values and r-values w/o code duplication and templates. According to "Effective modern c++" Item 41: "Consider pass by value for copyable parameters that are cheap to move and always copied."

This way an expensive copy must be performed by the caller and is not hidden in the ctor.

The constructor performed two things simultaniously, which is not good for a ctor: Constructing the object and calculating the values at given points. If needed such code should be moved to a standalone algorithm.

ogsbot · 2020-06-19T12:26:58Z

OpenGeoSys development has been moved to GitLab.

See this pull request on GitLab.

chleh reviewed Apr 19, 2016
View reviewed changes

endJunction force-pushed the PiecewiseLinearCurves branch from 1773862 to fa4886e Compare April 20, 2016 12:59

chleh added the please rebase label Apr 21, 2016

endJunction force-pushed the PiecewiseLinearCurves branch from fa4886e to 0cfc89b Compare April 21, 2016 09:44

endJunction added 6 commits April 21, 2016 12:55

[AppL] Add parsing of piecewise linear curves.

4dc6a1e

Pass r-values to PiecewiseLinearInterpolation ctor.

23b7c6e

[MaL] Pass vectors as r-values to PiecewiseLinInt.

479958f

This way an expensive copy must be performed by the caller and is not hidden in the ctor.

[MaL] Remove unused PiecewiseLinInterp ctor.

7661c0c

The constructor performed two things simultaniously, which is not good for a ctor: Constructing the object and calculating the values at given points. If needed such code should be moved to a standalone algorithm.

[MaL] Make PiecewiseLinearInterpolation final.

965c0b0

endJunction force-pushed the PiecewiseLinearCurves branch from 0cfc89b to 965c0b0 Compare April 21, 2016 10:55

endJunction merged commit 591b2be into ufz:master Apr 21, 2016

endJunction deleted the PiecewiseLinearCurves branch April 21, 2016 11:31

endJunction removed the please rebase label Apr 21, 2016

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Add parsing of piecewise linear curves. #1149

Add parsing of piecewise linear curves. #1149

endJunction commented Apr 19, 2016 •

edited

Loading

chleh Apr 19, 2016

endJunction Apr 19, 2016 •

edited

Loading

endJunction Apr 19, 2016

chleh Apr 19, 2016

endJunction Apr 19, 2016

chleh Apr 19, 2016

chleh commented Apr 19, 2016

chleh Apr 19, 2016 •

edited

Loading

wenqing commented Apr 19, 2016

endJunction commented Apr 20, 2016

TomFischer commented Apr 21, 2016

ogsbot commented Jun 19, 2020

Add parsing of piecewise linear curves. #1149

Add parsing of piecewise linear curves. #1149

Conversation

endJunction commented Apr 19, 2016 • edited Loading

chleh Apr 19, 2016

Choose a reason for hiding this comment

endJunction Apr 19, 2016 • edited Loading

Choose a reason for hiding this comment

endJunction Apr 19, 2016

Choose a reason for hiding this comment

chleh Apr 19, 2016

Choose a reason for hiding this comment

endJunction Apr 19, 2016

Choose a reason for hiding this comment

chleh Apr 19, 2016

Choose a reason for hiding this comment

chleh commented Apr 19, 2016

chleh Apr 19, 2016 • edited Loading

Choose a reason for hiding this comment

wenqing commented Apr 19, 2016

endJunction commented Apr 20, 2016

TomFischer commented Apr 21, 2016

ogsbot commented Jun 19, 2020

endJunction commented Apr 19, 2016 •

edited

Loading

endJunction Apr 19, 2016 •

edited

Loading

chleh Apr 19, 2016 •

edited

Loading