solvespace/src/expr.cpp

936 lines
26 KiB
C++
Raw Normal View History

//-----------------------------------------------------------------------------
// The symbolic algebra system used to write our constraint equations;
// routines to build expressions in software or from a user-provided string,
// and to compute the partial derivatives that we'll use when write our
// Jacobian matrix.
//
// Copyright 2008-2013 Jonathan Westhues.
//-----------------------------------------------------------------------------
#include "solvespace.h"
ExprVector ExprVector::From(Expr *x, Expr *y, Expr *z) {
ExprVector r = { x, y, z};
return r;
}
ExprVector ExprVector::From(Vector vn) {
ExprVector ve;
ve.x = Expr::From(vn.x);
ve.y = Expr::From(vn.y);
ve.z = Expr::From(vn.z);
return ve;
}
ExprVector ExprVector::From(hParam x, hParam y, hParam z) {
ExprVector ve;
ve.x = Expr::From(x);
ve.y = Expr::From(y);
ve.z = Expr::From(z);
return ve;
}
ExprVector ExprVector::From(double x, double y, double z) {
ExprVector ve;
ve.x = Expr::From(x);
ve.y = Expr::From(y);
ve.z = Expr::From(z);
return ve;
}
ExprVector ExprVector::Minus(ExprVector b) const {
ExprVector r;
r.x = x->Minus(b.x);
r.y = y->Minus(b.y);
r.z = z->Minus(b.z);
return r;
}
ExprVector ExprVector::Plus(ExprVector b) const {
ExprVector r;
r.x = x->Plus(b.x);
r.y = y->Plus(b.y);
r.z = z->Plus(b.z);
return r;
}
Expr *ExprVector::Dot(ExprVector b) const {
Expr *r;
r = x->Times(b.x);
r = r->Plus(y->Times(b.y));
r = r->Plus(z->Times(b.z));
return r;
}
ExprVector ExprVector::Cross(ExprVector b) const {
ExprVector r;
r.x = (y->Times(b.z))->Minus(z->Times(b.y));
r.y = (z->Times(b.x))->Minus(x->Times(b.z));
r.z = (x->Times(b.y))->Minus(y->Times(b.x));
return r;
}
ExprVector ExprVector::ScaledBy(Expr *s) const {
ExprVector r;
r.x = x->Times(s);
r.y = y->Times(s);
r.z = z->Times(s);
return r;
}
ExprVector ExprVector::WithMagnitude(Expr *s) const {
Expr *m = Magnitude();
return ScaledBy(s->Div(m));
}
Expr *ExprVector::Magnitude() const {
Expr *r;
r = x->Square();
r = r->Plus(y->Square());
r = r->Plus(z->Square());
return r->Sqrt();
}
Vector ExprVector::Eval() const {
Vector r;
r.x = x->Eval();
r.y = y->Eval();
r.z = z->Eval();
return r;
}
ExprQuaternion ExprQuaternion::From(hParam w, hParam vx, hParam vy, hParam vz) {
ExprQuaternion q;
q.w = Expr::From(w);
q.vx = Expr::From(vx);
q.vy = Expr::From(vy);
q.vz = Expr::From(vz);
return q;
}
ExprQuaternion ExprQuaternion::From(Expr *w, Expr *vx, Expr *vy, Expr *vz)
{
ExprQuaternion q;
q.w = w;
q.vx = vx;
q.vy = vy;
q.vz = vz;
return q;
}
ExprQuaternion ExprQuaternion::From(Quaternion qn) {
ExprQuaternion qe;
qe.w = Expr::From(qn.w);
qe.vx = Expr::From(qn.vx);
qe.vy = Expr::From(qn.vy);
qe.vz = Expr::From(qn.vz);
return qe;
}
ExprVector ExprQuaternion::RotationU() const {
ExprVector u;
Expr *two = Expr::From(2);
u.x = w->Square();
u.x = (u.x)->Plus(vx->Square());
u.x = (u.x)->Minus(vy->Square());
u.x = (u.x)->Minus(vz->Square());
u.y = two->Times(w->Times(vz));
u.y = (u.y)->Plus(two->Times(vx->Times(vy)));
u.z = two->Times(vx->Times(vz));
u.z = (u.z)->Minus(two->Times(w->Times(vy)));
return u;
}
ExprVector ExprQuaternion::RotationV() const {
ExprVector v;
Expr *two = Expr::From(2);
v.x = two->Times(vx->Times(vy));
v.x = (v.x)->Minus(two->Times(w->Times(vz)));
v.y = w->Square();
v.y = (v.y)->Minus(vx->Square());
v.y = (v.y)->Plus(vy->Square());
v.y = (v.y)->Minus(vz->Square());
v.z = two->Times(w->Times(vx));
v.z = (v.z)->Plus(two->Times(vy->Times(vz)));
return v;
}
ExprVector ExprQuaternion::RotationN() const {
ExprVector n;
Expr *two = Expr::From(2);
n.x = two->Times( w->Times(vy));
n.x = (n.x)->Plus (two->Times(vx->Times(vz)));
n.y = two->Times(vy->Times(vz));
n.y = (n.y)->Minus(two->Times( w->Times(vx)));
n.z = w->Square();
n.z = (n.z)->Minus(vx->Square());
n.z = (n.z)->Minus(vy->Square());
n.z = (n.z)->Plus (vz->Square());
return n;
}
ExprVector ExprQuaternion::Rotate(ExprVector p) const {
// Express the point in the new basis
return (RotationU().ScaledBy(p.x)).Plus(
RotationV().ScaledBy(p.y)).Plus(
RotationN().ScaledBy(p.z));
}
ExprQuaternion ExprQuaternion::Times(ExprQuaternion b) const {
Expr *sa = w, *sb = b.w;
ExprVector va = { vx, vy, vz };
ExprVector vb = { b.vx, b.vy, b.vz };
ExprQuaternion r;
r.w = (sa->Times(sb))->Minus(va.Dot(vb));
ExprVector vr = vb.ScaledBy(sa).Plus(
va.ScaledBy(sb).Plus(
va.Cross(vb)));
r.vx = vr.x;
r.vy = vr.y;
r.vz = vr.z;
return r;
}
Expr *ExprQuaternion::Magnitude() const {
return ((w ->Square())->Plus(
(vx->Square())->Plus(
(vy->Square())->Plus(
(vz->Square())))))->Sqrt();
}
Expr *Expr::From(hParam p) {
Expr *r = AllocExpr();
Convert all enumerations to use `enum class`. Specifically, take the old code that looks like this: class Foo { enum { X = 1, Y = 2 }; int kind; } ... foo.kind = Foo::X; ... and convert it to this: class Foo { enum class Kind : uint32_t { X = 1, Y = 2 }; Kind kind; } ... foo.kind = Foo::Kind::X; (In some cases the enumeration would not be in the class namespace, such as when it is generally useful.) The benefits are as follows: * The type of the field gives a clear indication of intent, both to humans and tools (such as binding generators). * The compiler is able to automatically warn when a switch is not exhaustive; but this is currently suppressed by the default: ssassert(false, ...) idiom. * Integers and plain enums are weakly type checked: they implicitly convert into each other. This can hide bugs where type conversion is performed but not intended. Enum classes are strongly type checked. * Plain enums pollute parent namespaces; enum classes do not. Almost every defined enum we have already has a kind of ad-hoc namespacing via `NAMESPACE_`, which is now explicit. * Plain enums do not have a well-defined ABI size, which is important for bindings. Enum classes can have it, if specified. We specify the base type for all enums as uint32_t, which is a safe choice and allows us to not change the numeric values of any variants. This commit introduces absolutely no functional change to the code, just renaming and change of types. It handles almost all cases, except GraphicsWindow::pending.operation, which needs minor functional change.
2016-05-20 08:31:20 +00:00
r->op = Op::PARAM;
r->parh = p;
return r;
}
Expr *Expr::From(double v) {
// Statically allocate common constants.
// Note: this is only valid because AllocExpr() uses AllocTemporary(),
// and Expr* is never explicitly freed.
if(v == 0.0) {
static Expr zero(0.0);
return &zero;
}
if(v == 1.0) {
static Expr one(1.0);
return &one;
}
if(v == -1.0) {
static Expr mone(-1.0);
return &mone;
}
if(v == 0.5) {
static Expr half(0.5);
return ½
}
if(v == -0.5) {
static Expr mhalf(-0.5);
return &mhalf;
}
Expr *r = AllocExpr();
Convert all enumerations to use `enum class`. Specifically, take the old code that looks like this: class Foo { enum { X = 1, Y = 2 }; int kind; } ... foo.kind = Foo::X; ... and convert it to this: class Foo { enum class Kind : uint32_t { X = 1, Y = 2 }; Kind kind; } ... foo.kind = Foo::Kind::X; (In some cases the enumeration would not be in the class namespace, such as when it is generally useful.) The benefits are as follows: * The type of the field gives a clear indication of intent, both to humans and tools (such as binding generators). * The compiler is able to automatically warn when a switch is not exhaustive; but this is currently suppressed by the default: ssassert(false, ...) idiom. * Integers and plain enums are weakly type checked: they implicitly convert into each other. This can hide bugs where type conversion is performed but not intended. Enum classes are strongly type checked. * Plain enums pollute parent namespaces; enum classes do not. Almost every defined enum we have already has a kind of ad-hoc namespacing via `NAMESPACE_`, which is now explicit. * Plain enums do not have a well-defined ABI size, which is important for bindings. Enum classes can have it, if specified. We specify the base type for all enums as uint32_t, which is a safe choice and allows us to not change the numeric values of any variants. This commit introduces absolutely no functional change to the code, just renaming and change of types. It handles almost all cases, except GraphicsWindow::pending.operation, which needs minor functional change.
2016-05-20 08:31:20 +00:00
r->op = Op::CONSTANT;
r->v = v;
return r;
}
Convert all enumerations to use `enum class`. Specifically, take the old code that looks like this: class Foo { enum { X = 1, Y = 2 }; int kind; } ... foo.kind = Foo::X; ... and convert it to this: class Foo { enum class Kind : uint32_t { X = 1, Y = 2 }; Kind kind; } ... foo.kind = Foo::Kind::X; (In some cases the enumeration would not be in the class namespace, such as when it is generally useful.) The benefits are as follows: * The type of the field gives a clear indication of intent, both to humans and tools (such as binding generators). * The compiler is able to automatically warn when a switch is not exhaustive; but this is currently suppressed by the default: ssassert(false, ...) idiom. * Integers and plain enums are weakly type checked: they implicitly convert into each other. This can hide bugs where type conversion is performed but not intended. Enum classes are strongly type checked. * Plain enums pollute parent namespaces; enum classes do not. Almost every defined enum we have already has a kind of ad-hoc namespacing via `NAMESPACE_`, which is now explicit. * Plain enums do not have a well-defined ABI size, which is important for bindings. Enum classes can have it, if specified. We specify the base type for all enums as uint32_t, which is a safe choice and allows us to not change the numeric values of any variants. This commit introduces absolutely no functional change to the code, just renaming and change of types. It handles almost all cases, except GraphicsWindow::pending.operation, which needs minor functional change.
2016-05-20 08:31:20 +00:00
Expr *Expr::AnyOp(Op newOp, Expr *b) {
Expr *r = AllocExpr();
r->op = newOp;
r->a = this;
r->b = b;
return r;
}
int Expr::Children() const {
switch(op) {
Convert all enumerations to use `enum class`. Specifically, take the old code that looks like this: class Foo { enum { X = 1, Y = 2 }; int kind; } ... foo.kind = Foo::X; ... and convert it to this: class Foo { enum class Kind : uint32_t { X = 1, Y = 2 }; Kind kind; } ... foo.kind = Foo::Kind::X; (In some cases the enumeration would not be in the class namespace, such as when it is generally useful.) The benefits are as follows: * The type of the field gives a clear indication of intent, both to humans and tools (such as binding generators). * The compiler is able to automatically warn when a switch is not exhaustive; but this is currently suppressed by the default: ssassert(false, ...) idiom. * Integers and plain enums are weakly type checked: they implicitly convert into each other. This can hide bugs where type conversion is performed but not intended. Enum classes are strongly type checked. * Plain enums pollute parent namespaces; enum classes do not. Almost every defined enum we have already has a kind of ad-hoc namespacing via `NAMESPACE_`, which is now explicit. * Plain enums do not have a well-defined ABI size, which is important for bindings. Enum classes can have it, if specified. We specify the base type for all enums as uint32_t, which is a safe choice and allows us to not change the numeric values of any variants. This commit introduces absolutely no functional change to the code, just renaming and change of types. It handles almost all cases, except GraphicsWindow::pending.operation, which needs minor functional change.
2016-05-20 08:31:20 +00:00
case Op::PARAM:
case Op::PARAM_PTR:
case Op::CONSTANT:
case Op::VARIABLE:
return 0;
Convert all enumerations to use `enum class`. Specifically, take the old code that looks like this: class Foo { enum { X = 1, Y = 2 }; int kind; } ... foo.kind = Foo::X; ... and convert it to this: class Foo { enum class Kind : uint32_t { X = 1, Y = 2 }; Kind kind; } ... foo.kind = Foo::Kind::X; (In some cases the enumeration would not be in the class namespace, such as when it is generally useful.) The benefits are as follows: * The type of the field gives a clear indication of intent, both to humans and tools (such as binding generators). * The compiler is able to automatically warn when a switch is not exhaustive; but this is currently suppressed by the default: ssassert(false, ...) idiom. * Integers and plain enums are weakly type checked: they implicitly convert into each other. This can hide bugs where type conversion is performed but not intended. Enum classes are strongly type checked. * Plain enums pollute parent namespaces; enum classes do not. Almost every defined enum we have already has a kind of ad-hoc namespacing via `NAMESPACE_`, which is now explicit. * Plain enums do not have a well-defined ABI size, which is important for bindings. Enum classes can have it, if specified. We specify the base type for all enums as uint32_t, which is a safe choice and allows us to not change the numeric values of any variants. This commit introduces absolutely no functional change to the code, just renaming and change of types. It handles almost all cases, except GraphicsWindow::pending.operation, which needs minor functional change.
2016-05-20 08:31:20 +00:00
case Op::PLUS:
case Op::MINUS:
case Op::TIMES:
case Op::DIV:
return 2;
Convert all enumerations to use `enum class`. Specifically, take the old code that looks like this: class Foo { enum { X = 1, Y = 2 }; int kind; } ... foo.kind = Foo::X; ... and convert it to this: class Foo { enum class Kind : uint32_t { X = 1, Y = 2 }; Kind kind; } ... foo.kind = Foo::Kind::X; (In some cases the enumeration would not be in the class namespace, such as when it is generally useful.) The benefits are as follows: * The type of the field gives a clear indication of intent, both to humans and tools (such as binding generators). * The compiler is able to automatically warn when a switch is not exhaustive; but this is currently suppressed by the default: ssassert(false, ...) idiom. * Integers and plain enums are weakly type checked: they implicitly convert into each other. This can hide bugs where type conversion is performed but not intended. Enum classes are strongly type checked. * Plain enums pollute parent namespaces; enum classes do not. Almost every defined enum we have already has a kind of ad-hoc namespacing via `NAMESPACE_`, which is now explicit. * Plain enums do not have a well-defined ABI size, which is important for bindings. Enum classes can have it, if specified. We specify the base type for all enums as uint32_t, which is a safe choice and allows us to not change the numeric values of any variants. This commit introduces absolutely no functional change to the code, just renaming and change of types. It handles almost all cases, except GraphicsWindow::pending.operation, which needs minor functional change.
2016-05-20 08:31:20 +00:00
case Op::NEGATE:
case Op::SQRT:
case Op::SQUARE:
case Op::SIN:
case Op::COS:
case Op::ASIN:
case Op::ACOS:
return 1;
}
Enable exhaustive switch coverage warnings as an error, and use them. Specifically, this enables -Wswitch=error on GCC/Clang and its MSVC equivalent; the exact way it is handled varies slightly, but what they all have in common is that in a switch statement over an enumeration, any enumerand that is not explicitly (via case:) or implicitly (via default:) handled in the switch triggers an error. Moreover, we also change the switch statements in three ways: * Switch statements that ought to be extended every time a new enumerand is added (e.g. Entity::DrawOrGetDistance(), are changed to explicitly list every single enumerand, and not have a default: branch. Note that the assertions are kept because it is legal for a enumeration to have a value unlike any of its defined enumerands, and we can e.g. read garbage from a file, or an uninitialized variable. This requires some rearranging if a default: branch is undesired. * Switch statements that ought to only ever see a few select enumerands, are changed to always assert in the default: branch. * Switch statements that do something meaningful for a few enumerands, and ignore everything else, are changed to do nothing in a default: branch, under the assumption that changing them every time an enumerand is added or removed would just result in noise and catch no bugs. This commit also removes the {Request,Entity,Constraint}::UNKNOWN and Entity::DATUM_POINT enumerands, as those were just fancy names for zeroes. They mess up switch exhaustiveness checks and most of the time were not the best way to implement what they did anyway.
2016-05-25 06:55:50 +00:00
ssassert(false, "Unexpected operation");
}
int Expr::Nodes() const {
switch(Children()) {
case 0: return 1;
case 1: return 1 + a->Nodes();
case 2: return 1 + a->Nodes() + b->Nodes();
default: ssassert(false, "Unexpected children count");
}
}
Expr *Expr::DeepCopy() const {
Expr *n = AllocExpr();
*n = *this;
int c = n->Children();
if(c > 0) n->a = a->DeepCopy();
if(c > 1) n->b = b->DeepCopy();
return n;
}
Expr *Expr::DeepCopyWithParamsAsPointers(IdList<Param,hParam> *firstTry,
IdList<Param,hParam> *thenTry) const
{
Expr *n = AllocExpr();
Convert all enumerations to use `enum class`. Specifically, take the old code that looks like this: class Foo { enum { X = 1, Y = 2 }; int kind; } ... foo.kind = Foo::X; ... and convert it to this: class Foo { enum class Kind : uint32_t { X = 1, Y = 2 }; Kind kind; } ... foo.kind = Foo::Kind::X; (In some cases the enumeration would not be in the class namespace, such as when it is generally useful.) The benefits are as follows: * The type of the field gives a clear indication of intent, both to humans and tools (such as binding generators). * The compiler is able to automatically warn when a switch is not exhaustive; but this is currently suppressed by the default: ssassert(false, ...) idiom. * Integers and plain enums are weakly type checked: they implicitly convert into each other. This can hide bugs where type conversion is performed but not intended. Enum classes are strongly type checked. * Plain enums pollute parent namespaces; enum classes do not. Almost every defined enum we have already has a kind of ad-hoc namespacing via `NAMESPACE_`, which is now explicit. * Plain enums do not have a well-defined ABI size, which is important for bindings. Enum classes can have it, if specified. We specify the base type for all enums as uint32_t, which is a safe choice and allows us to not change the numeric values of any variants. This commit introduces absolutely no functional change to the code, just renaming and change of types. It handles almost all cases, except GraphicsWindow::pending.operation, which needs minor functional change.
2016-05-20 08:31:20 +00:00
if(op == Op::PARAM) {
// A param that is referenced by its hParam gets rewritten to go
// straight in to the parameter table with a pointer, or simply
// into a constant if it's already known.
Param *p = firstTry->FindByIdNoOops(parh);
if(!p) p = thenTry->FindById(parh);
if(p->known) {
Convert all enumerations to use `enum class`. Specifically, take the old code that looks like this: class Foo { enum { X = 1, Y = 2 }; int kind; } ... foo.kind = Foo::X; ... and convert it to this: class Foo { enum class Kind : uint32_t { X = 1, Y = 2 }; Kind kind; } ... foo.kind = Foo::Kind::X; (In some cases the enumeration would not be in the class namespace, such as when it is generally useful.) The benefits are as follows: * The type of the field gives a clear indication of intent, both to humans and tools (such as binding generators). * The compiler is able to automatically warn when a switch is not exhaustive; but this is currently suppressed by the default: ssassert(false, ...) idiom. * Integers and plain enums are weakly type checked: they implicitly convert into each other. This can hide bugs where type conversion is performed but not intended. Enum classes are strongly type checked. * Plain enums pollute parent namespaces; enum classes do not. Almost every defined enum we have already has a kind of ad-hoc namespacing via `NAMESPACE_`, which is now explicit. * Plain enums do not have a well-defined ABI size, which is important for bindings. Enum classes can have it, if specified. We specify the base type for all enums as uint32_t, which is a safe choice and allows us to not change the numeric values of any variants. This commit introduces absolutely no functional change to the code, just renaming and change of types. It handles almost all cases, except GraphicsWindow::pending.operation, which needs minor functional change.
2016-05-20 08:31:20 +00:00
n->op = Op::CONSTANT;
n->v = p->val;
} else {
Convert all enumerations to use `enum class`. Specifically, take the old code that looks like this: class Foo { enum { X = 1, Y = 2 }; int kind; } ... foo.kind = Foo::X; ... and convert it to this: class Foo { enum class Kind : uint32_t { X = 1, Y = 2 }; Kind kind; } ... foo.kind = Foo::Kind::X; (In some cases the enumeration would not be in the class namespace, such as when it is generally useful.) The benefits are as follows: * The type of the field gives a clear indication of intent, both to humans and tools (such as binding generators). * The compiler is able to automatically warn when a switch is not exhaustive; but this is currently suppressed by the default: ssassert(false, ...) idiom. * Integers and plain enums are weakly type checked: they implicitly convert into each other. This can hide bugs where type conversion is performed but not intended. Enum classes are strongly type checked. * Plain enums pollute parent namespaces; enum classes do not. Almost every defined enum we have already has a kind of ad-hoc namespacing via `NAMESPACE_`, which is now explicit. * Plain enums do not have a well-defined ABI size, which is important for bindings. Enum classes can have it, if specified. We specify the base type for all enums as uint32_t, which is a safe choice and allows us to not change the numeric values of any variants. This commit introduces absolutely no functional change to the code, just renaming and change of types. It handles almost all cases, except GraphicsWindow::pending.operation, which needs minor functional change.
2016-05-20 08:31:20 +00:00
n->op = Op::PARAM_PTR;
n->parp = p;
}
return n;
}
*n = *this;
int c = n->Children();
if(c > 0) n->a = a->DeepCopyWithParamsAsPointers(firstTry, thenTry);
if(c > 1) n->b = b->DeepCopyWithParamsAsPointers(firstTry, thenTry);
return n;
}
double Expr::Eval() const {
switch(op) {
Convert all enumerations to use `enum class`. Specifically, take the old code that looks like this: class Foo { enum { X = 1, Y = 2 }; int kind; } ... foo.kind = Foo::X; ... and convert it to this: class Foo { enum class Kind : uint32_t { X = 1, Y = 2 }; Kind kind; } ... foo.kind = Foo::Kind::X; (In some cases the enumeration would not be in the class namespace, such as when it is generally useful.) The benefits are as follows: * The type of the field gives a clear indication of intent, both to humans and tools (such as binding generators). * The compiler is able to automatically warn when a switch is not exhaustive; but this is currently suppressed by the default: ssassert(false, ...) idiom. * Integers and plain enums are weakly type checked: they implicitly convert into each other. This can hide bugs where type conversion is performed but not intended. Enum classes are strongly type checked. * Plain enums pollute parent namespaces; enum classes do not. Almost every defined enum we have already has a kind of ad-hoc namespacing via `NAMESPACE_`, which is now explicit. * Plain enums do not have a well-defined ABI size, which is important for bindings. Enum classes can have it, if specified. We specify the base type for all enums as uint32_t, which is a safe choice and allows us to not change the numeric values of any variants. This commit introduces absolutely no functional change to the code, just renaming and change of types. It handles almost all cases, except GraphicsWindow::pending.operation, which needs minor functional change.
2016-05-20 08:31:20 +00:00
case Op::PARAM: return SK.GetParam(parh)->val;
case Op::PARAM_PTR: return parp->val;
Convert all enumerations to use `enum class`. Specifically, take the old code that looks like this: class Foo { enum { X = 1, Y = 2 }; int kind; } ... foo.kind = Foo::X; ... and convert it to this: class Foo { enum class Kind : uint32_t { X = 1, Y = 2 }; Kind kind; } ... foo.kind = Foo::Kind::X; (In some cases the enumeration would not be in the class namespace, such as when it is generally useful.) The benefits are as follows: * The type of the field gives a clear indication of intent, both to humans and tools (such as binding generators). * The compiler is able to automatically warn when a switch is not exhaustive; but this is currently suppressed by the default: ssassert(false, ...) idiom. * Integers and plain enums are weakly type checked: they implicitly convert into each other. This can hide bugs where type conversion is performed but not intended. Enum classes are strongly type checked. * Plain enums pollute parent namespaces; enum classes do not. Almost every defined enum we have already has a kind of ad-hoc namespacing via `NAMESPACE_`, which is now explicit. * Plain enums do not have a well-defined ABI size, which is important for bindings. Enum classes can have it, if specified. We specify the base type for all enums as uint32_t, which is a safe choice and allows us to not change the numeric values of any variants. This commit introduces absolutely no functional change to the code, just renaming and change of types. It handles almost all cases, except GraphicsWindow::pending.operation, which needs minor functional change.
2016-05-20 08:31:20 +00:00
case Op::CONSTANT: return v;
case Op::VARIABLE: ssassert(false, "Not supported yet");
Convert all enumerations to use `enum class`. Specifically, take the old code that looks like this: class Foo { enum { X = 1, Y = 2 }; int kind; } ... foo.kind = Foo::X; ... and convert it to this: class Foo { enum class Kind : uint32_t { X = 1, Y = 2 }; Kind kind; } ... foo.kind = Foo::Kind::X; (In some cases the enumeration would not be in the class namespace, such as when it is generally useful.) The benefits are as follows: * The type of the field gives a clear indication of intent, both to humans and tools (such as binding generators). * The compiler is able to automatically warn when a switch is not exhaustive; but this is currently suppressed by the default: ssassert(false, ...) idiom. * Integers and plain enums are weakly type checked: they implicitly convert into each other. This can hide bugs where type conversion is performed but not intended. Enum classes are strongly type checked. * Plain enums pollute parent namespaces; enum classes do not. Almost every defined enum we have already has a kind of ad-hoc namespacing via `NAMESPACE_`, which is now explicit. * Plain enums do not have a well-defined ABI size, which is important for bindings. Enum classes can have it, if specified. We specify the base type for all enums as uint32_t, which is a safe choice and allows us to not change the numeric values of any variants. This commit introduces absolutely no functional change to the code, just renaming and change of types. It handles almost all cases, except GraphicsWindow::pending.operation, which needs minor functional change.
2016-05-20 08:31:20 +00:00
case Op::PLUS: return a->Eval() + b->Eval();
case Op::MINUS: return a->Eval() - b->Eval();
case Op::TIMES: return a->Eval() * b->Eval();
case Op::DIV: return a->Eval() / b->Eval();
Convert all enumerations to use `enum class`. Specifically, take the old code that looks like this: class Foo { enum { X = 1, Y = 2 }; int kind; } ... foo.kind = Foo::X; ... and convert it to this: class Foo { enum class Kind : uint32_t { X = 1, Y = 2 }; Kind kind; } ... foo.kind = Foo::Kind::X; (In some cases the enumeration would not be in the class namespace, such as when it is generally useful.) The benefits are as follows: * The type of the field gives a clear indication of intent, both to humans and tools (such as binding generators). * The compiler is able to automatically warn when a switch is not exhaustive; but this is currently suppressed by the default: ssassert(false, ...) idiom. * Integers and plain enums are weakly type checked: they implicitly convert into each other. This can hide bugs where type conversion is performed but not intended. Enum classes are strongly type checked. * Plain enums pollute parent namespaces; enum classes do not. Almost every defined enum we have already has a kind of ad-hoc namespacing via `NAMESPACE_`, which is now explicit. * Plain enums do not have a well-defined ABI size, which is important for bindings. Enum classes can have it, if specified. We specify the base type for all enums as uint32_t, which is a safe choice and allows us to not change the numeric values of any variants. This commit introduces absolutely no functional change to the code, just renaming and change of types. It handles almost all cases, except GraphicsWindow::pending.operation, which needs minor functional change.
2016-05-20 08:31:20 +00:00
case Op::NEGATE: return -(a->Eval());
case Op::SQRT: return sqrt(a->Eval());
case Op::SQUARE: { double r = a->Eval(); return r*r; }
case Op::SIN: return sin(a->Eval());
case Op::COS: return cos(a->Eval());
case Op::ACOS: return acos(a->Eval());
case Op::ASIN: return asin(a->Eval());
}
Enable exhaustive switch coverage warnings as an error, and use them. Specifically, this enables -Wswitch=error on GCC/Clang and its MSVC equivalent; the exact way it is handled varies slightly, but what they all have in common is that in a switch statement over an enumeration, any enumerand that is not explicitly (via case:) or implicitly (via default:) handled in the switch triggers an error. Moreover, we also change the switch statements in three ways: * Switch statements that ought to be extended every time a new enumerand is added (e.g. Entity::DrawOrGetDistance(), are changed to explicitly list every single enumerand, and not have a default: branch. Note that the assertions are kept because it is legal for a enumeration to have a value unlike any of its defined enumerands, and we can e.g. read garbage from a file, or an uninitialized variable. This requires some rearranging if a default: branch is undesired. * Switch statements that ought to only ever see a few select enumerands, are changed to always assert in the default: branch. * Switch statements that do something meaningful for a few enumerands, and ignore everything else, are changed to do nothing in a default: branch, under the assumption that changing them every time an enumerand is added or removed would just result in noise and catch no bugs. This commit also removes the {Request,Entity,Constraint}::UNKNOWN and Entity::DATUM_POINT enumerands, as those were just fancy names for zeroes. They mess up switch exhaustiveness checks and most of the time were not the best way to implement what they did anyway.
2016-05-25 06:55:50 +00:00
ssassert(false, "Unexpected operation");
}
Expr *Expr::PartialWrt(hParam p) const {
Expr *da, *db;
switch(op) {
case Op::PARAM_PTR: return From(p == parp->h ? 1 : 0);
case Op::PARAM: return From(p == parh ? 1 : 0);
Convert all enumerations to use `enum class`. Specifically, take the old code that looks like this: class Foo { enum { X = 1, Y = 2 }; int kind; } ... foo.kind = Foo::X; ... and convert it to this: class Foo { enum class Kind : uint32_t { X = 1, Y = 2 }; Kind kind; } ... foo.kind = Foo::Kind::X; (In some cases the enumeration would not be in the class namespace, such as when it is generally useful.) The benefits are as follows: * The type of the field gives a clear indication of intent, both to humans and tools (such as binding generators). * The compiler is able to automatically warn when a switch is not exhaustive; but this is currently suppressed by the default: ssassert(false, ...) idiom. * Integers and plain enums are weakly type checked: they implicitly convert into each other. This can hide bugs where type conversion is performed but not intended. Enum classes are strongly type checked. * Plain enums pollute parent namespaces; enum classes do not. Almost every defined enum we have already has a kind of ad-hoc namespacing via `NAMESPACE_`, which is now explicit. * Plain enums do not have a well-defined ABI size, which is important for bindings. Enum classes can have it, if specified. We specify the base type for all enums as uint32_t, which is a safe choice and allows us to not change the numeric values of any variants. This commit introduces absolutely no functional change to the code, just renaming and change of types. It handles almost all cases, except GraphicsWindow::pending.operation, which needs minor functional change.
2016-05-20 08:31:20 +00:00
case Op::CONSTANT: return From(0.0);
case Op::VARIABLE: ssassert(false, "Not supported yet");
Convert all enumerations to use `enum class`. Specifically, take the old code that looks like this: class Foo { enum { X = 1, Y = 2 }; int kind; } ... foo.kind = Foo::X; ... and convert it to this: class Foo { enum class Kind : uint32_t { X = 1, Y = 2 }; Kind kind; } ... foo.kind = Foo::Kind::X; (In some cases the enumeration would not be in the class namespace, such as when it is generally useful.) The benefits are as follows: * The type of the field gives a clear indication of intent, both to humans and tools (such as binding generators). * The compiler is able to automatically warn when a switch is not exhaustive; but this is currently suppressed by the default: ssassert(false, ...) idiom. * Integers and plain enums are weakly type checked: they implicitly convert into each other. This can hide bugs where type conversion is performed but not intended. Enum classes are strongly type checked. * Plain enums pollute parent namespaces; enum classes do not. Almost every defined enum we have already has a kind of ad-hoc namespacing via `NAMESPACE_`, which is now explicit. * Plain enums do not have a well-defined ABI size, which is important for bindings. Enum classes can have it, if specified. We specify the base type for all enums as uint32_t, which is a safe choice and allows us to not change the numeric values of any variants. This commit introduces absolutely no functional change to the code, just renaming and change of types. It handles almost all cases, except GraphicsWindow::pending.operation, which needs minor functional change.
2016-05-20 08:31:20 +00:00
case Op::PLUS: return (a->PartialWrt(p))->Plus(b->PartialWrt(p));
case Op::MINUS: return (a->PartialWrt(p))->Minus(b->PartialWrt(p));
Convert all enumerations to use `enum class`. Specifically, take the old code that looks like this: class Foo { enum { X = 1, Y = 2 }; int kind; } ... foo.kind = Foo::X; ... and convert it to this: class Foo { enum class Kind : uint32_t { X = 1, Y = 2 }; Kind kind; } ... foo.kind = Foo::Kind::X; (In some cases the enumeration would not be in the class namespace, such as when it is generally useful.) The benefits are as follows: * The type of the field gives a clear indication of intent, both to humans and tools (such as binding generators). * The compiler is able to automatically warn when a switch is not exhaustive; but this is currently suppressed by the default: ssassert(false, ...) idiom. * Integers and plain enums are weakly type checked: they implicitly convert into each other. This can hide bugs where type conversion is performed but not intended. Enum classes are strongly type checked. * Plain enums pollute parent namespaces; enum classes do not. Almost every defined enum we have already has a kind of ad-hoc namespacing via `NAMESPACE_`, which is now explicit. * Plain enums do not have a well-defined ABI size, which is important for bindings. Enum classes can have it, if specified. We specify the base type for all enums as uint32_t, which is a safe choice and allows us to not change the numeric values of any variants. This commit introduces absolutely no functional change to the code, just renaming and change of types. It handles almost all cases, except GraphicsWindow::pending.operation, which needs minor functional change.
2016-05-20 08:31:20 +00:00
case Op::TIMES:
da = a->PartialWrt(p);
db = b->PartialWrt(p);
return (a->Times(db))->Plus(b->Times(da));
Convert all enumerations to use `enum class`. Specifically, take the old code that looks like this: class Foo { enum { X = 1, Y = 2 }; int kind; } ... foo.kind = Foo::X; ... and convert it to this: class Foo { enum class Kind : uint32_t { X = 1, Y = 2 }; Kind kind; } ... foo.kind = Foo::Kind::X; (In some cases the enumeration would not be in the class namespace, such as when it is generally useful.) The benefits are as follows: * The type of the field gives a clear indication of intent, both to humans and tools (such as binding generators). * The compiler is able to automatically warn when a switch is not exhaustive; but this is currently suppressed by the default: ssassert(false, ...) idiom. * Integers and plain enums are weakly type checked: they implicitly convert into each other. This can hide bugs where type conversion is performed but not intended. Enum classes are strongly type checked. * Plain enums pollute parent namespaces; enum classes do not. Almost every defined enum we have already has a kind of ad-hoc namespacing via `NAMESPACE_`, which is now explicit. * Plain enums do not have a well-defined ABI size, which is important for bindings. Enum classes can have it, if specified. We specify the base type for all enums as uint32_t, which is a safe choice and allows us to not change the numeric values of any variants. This commit introduces absolutely no functional change to the code, just renaming and change of types. It handles almost all cases, except GraphicsWindow::pending.operation, which needs minor functional change.
2016-05-20 08:31:20 +00:00
case Op::DIV:
da = a->PartialWrt(p);
db = b->PartialWrt(p);
return ((da->Times(b))->Minus(a->Times(db)))->Div(b->Square());
Convert all enumerations to use `enum class`. Specifically, take the old code that looks like this: class Foo { enum { X = 1, Y = 2 }; int kind; } ... foo.kind = Foo::X; ... and convert it to this: class Foo { enum class Kind : uint32_t { X = 1, Y = 2 }; Kind kind; } ... foo.kind = Foo::Kind::X; (In some cases the enumeration would not be in the class namespace, such as when it is generally useful.) The benefits are as follows: * The type of the field gives a clear indication of intent, both to humans and tools (such as binding generators). * The compiler is able to automatically warn when a switch is not exhaustive; but this is currently suppressed by the default: ssassert(false, ...) idiom. * Integers and plain enums are weakly type checked: they implicitly convert into each other. This can hide bugs where type conversion is performed but not intended. Enum classes are strongly type checked. * Plain enums pollute parent namespaces; enum classes do not. Almost every defined enum we have already has a kind of ad-hoc namespacing via `NAMESPACE_`, which is now explicit. * Plain enums do not have a well-defined ABI size, which is important for bindings. Enum classes can have it, if specified. We specify the base type for all enums as uint32_t, which is a safe choice and allows us to not change the numeric values of any variants. This commit introduces absolutely no functional change to the code, just renaming and change of types. It handles almost all cases, except GraphicsWindow::pending.operation, which needs minor functional change.
2016-05-20 08:31:20 +00:00
case Op::SQRT:
return (From(0.5)->Div(a->Sqrt()))->Times(a->PartialWrt(p));
Convert all enumerations to use `enum class`. Specifically, take the old code that looks like this: class Foo { enum { X = 1, Y = 2 }; int kind; } ... foo.kind = Foo::X; ... and convert it to this: class Foo { enum class Kind : uint32_t { X = 1, Y = 2 }; Kind kind; } ... foo.kind = Foo::Kind::X; (In some cases the enumeration would not be in the class namespace, such as when it is generally useful.) The benefits are as follows: * The type of the field gives a clear indication of intent, both to humans and tools (such as binding generators). * The compiler is able to automatically warn when a switch is not exhaustive; but this is currently suppressed by the default: ssassert(false, ...) idiom. * Integers and plain enums are weakly type checked: they implicitly convert into each other. This can hide bugs where type conversion is performed but not intended. Enum classes are strongly type checked. * Plain enums pollute parent namespaces; enum classes do not. Almost every defined enum we have already has a kind of ad-hoc namespacing via `NAMESPACE_`, which is now explicit. * Plain enums do not have a well-defined ABI size, which is important for bindings. Enum classes can have it, if specified. We specify the base type for all enums as uint32_t, which is a safe choice and allows us to not change the numeric values of any variants. This commit introduces absolutely no functional change to the code, just renaming and change of types. It handles almost all cases, except GraphicsWindow::pending.operation, which needs minor functional change.
2016-05-20 08:31:20 +00:00
case Op::SQUARE:
return (From(2.0)->Times(a))->Times(a->PartialWrt(p));
Convert all enumerations to use `enum class`. Specifically, take the old code that looks like this: class Foo { enum { X = 1, Y = 2 }; int kind; } ... foo.kind = Foo::X; ... and convert it to this: class Foo { enum class Kind : uint32_t { X = 1, Y = 2 }; Kind kind; } ... foo.kind = Foo::Kind::X; (In some cases the enumeration would not be in the class namespace, such as when it is generally useful.) The benefits are as follows: * The type of the field gives a clear indication of intent, both to humans and tools (such as binding generators). * The compiler is able to automatically warn when a switch is not exhaustive; but this is currently suppressed by the default: ssassert(false, ...) idiom. * Integers and plain enums are weakly type checked: they implicitly convert into each other. This can hide bugs where type conversion is performed but not intended. Enum classes are strongly type checked. * Plain enums pollute parent namespaces; enum classes do not. Almost every defined enum we have already has a kind of ad-hoc namespacing via `NAMESPACE_`, which is now explicit. * Plain enums do not have a well-defined ABI size, which is important for bindings. Enum classes can have it, if specified. We specify the base type for all enums as uint32_t, which is a safe choice and allows us to not change the numeric values of any variants. This commit introduces absolutely no functional change to the code, just renaming and change of types. It handles almost all cases, except GraphicsWindow::pending.operation, which needs minor functional change.
2016-05-20 08:31:20 +00:00
case Op::NEGATE: return (a->PartialWrt(p))->Negate();
case Op::SIN: return (a->Cos())->Times(a->PartialWrt(p));
case Op::COS: return ((a->Sin())->Times(a->PartialWrt(p)))->Negate();
Convert all enumerations to use `enum class`. Specifically, take the old code that looks like this: class Foo { enum { X = 1, Y = 2 }; int kind; } ... foo.kind = Foo::X; ... and convert it to this: class Foo { enum class Kind : uint32_t { X = 1, Y = 2 }; Kind kind; } ... foo.kind = Foo::Kind::X; (In some cases the enumeration would not be in the class namespace, such as when it is generally useful.) The benefits are as follows: * The type of the field gives a clear indication of intent, both to humans and tools (such as binding generators). * The compiler is able to automatically warn when a switch is not exhaustive; but this is currently suppressed by the default: ssassert(false, ...) idiom. * Integers and plain enums are weakly type checked: they implicitly convert into each other. This can hide bugs where type conversion is performed but not intended. Enum classes are strongly type checked. * Plain enums pollute parent namespaces; enum classes do not. Almost every defined enum we have already has a kind of ad-hoc namespacing via `NAMESPACE_`, which is now explicit. * Plain enums do not have a well-defined ABI size, which is important for bindings. Enum classes can have it, if specified. We specify the base type for all enums as uint32_t, which is a safe choice and allows us to not change the numeric values of any variants. This commit introduces absolutely no functional change to the code, just renaming and change of types. It handles almost all cases, except GraphicsWindow::pending.operation, which needs minor functional change.
2016-05-20 08:31:20 +00:00
case Op::ASIN:
return (From(1)->Div((From(1)->Minus(a->Square()))->Sqrt()))
->Times(a->PartialWrt(p));
Convert all enumerations to use `enum class`. Specifically, take the old code that looks like this: class Foo { enum { X = 1, Y = 2 }; int kind; } ... foo.kind = Foo::X; ... and convert it to this: class Foo { enum class Kind : uint32_t { X = 1, Y = 2 }; Kind kind; } ... foo.kind = Foo::Kind::X; (In some cases the enumeration would not be in the class namespace, such as when it is generally useful.) The benefits are as follows: * The type of the field gives a clear indication of intent, both to humans and tools (such as binding generators). * The compiler is able to automatically warn when a switch is not exhaustive; but this is currently suppressed by the default: ssassert(false, ...) idiom. * Integers and plain enums are weakly type checked: they implicitly convert into each other. This can hide bugs where type conversion is performed but not intended. Enum classes are strongly type checked. * Plain enums pollute parent namespaces; enum classes do not. Almost every defined enum we have already has a kind of ad-hoc namespacing via `NAMESPACE_`, which is now explicit. * Plain enums do not have a well-defined ABI size, which is important for bindings. Enum classes can have it, if specified. We specify the base type for all enums as uint32_t, which is a safe choice and allows us to not change the numeric values of any variants. This commit introduces absolutely no functional change to the code, just renaming and change of types. It handles almost all cases, except GraphicsWindow::pending.operation, which needs minor functional change.
2016-05-20 08:31:20 +00:00
case Op::ACOS:
return (From(-1)->Div((From(1)->Minus(a->Square()))->Sqrt()))
->Times(a->PartialWrt(p));
}
Enable exhaustive switch coverage warnings as an error, and use them. Specifically, this enables -Wswitch=error on GCC/Clang and its MSVC equivalent; the exact way it is handled varies slightly, but what they all have in common is that in a switch statement over an enumeration, any enumerand that is not explicitly (via case:) or implicitly (via default:) handled in the switch triggers an error. Moreover, we also change the switch statements in three ways: * Switch statements that ought to be extended every time a new enumerand is added (e.g. Entity::DrawOrGetDistance(), are changed to explicitly list every single enumerand, and not have a default: branch. Note that the assertions are kept because it is legal for a enumeration to have a value unlike any of its defined enumerands, and we can e.g. read garbage from a file, or an uninitialized variable. This requires some rearranging if a default: branch is undesired. * Switch statements that ought to only ever see a few select enumerands, are changed to always assert in the default: branch. * Switch statements that do something meaningful for a few enumerands, and ignore everything else, are changed to do nothing in a default: branch, under the assumption that changing them every time an enumerand is added or removed would just result in noise and catch no bugs. This commit also removes the {Request,Entity,Constraint}::UNKNOWN and Entity::DATUM_POINT enumerands, as those were just fancy names for zeroes. They mess up switch exhaustiveness checks and most of the time were not the best way to implement what they did anyway.
2016-05-25 06:55:50 +00:00
ssassert(false, "Unexpected operation");
}
void Expr::ParamsUsedList(std::vector<hParam> *list) const {
2017-05-12 02:39:07 +00:00
if(op == Op::PARAM || op == Op::PARAM_PTR) {
// leaf: just add ourselves if we aren't already there
2017-05-12 02:39:07 +00:00
hParam param = (op == Op::PARAM) ? parh : parp->h;
if(list->end() != std::find_if(list->begin(), list->end(),
[=](const hParam &p) { return p.v == param.v; })) {
// We found ourselves in the list already, early out.
return;
2017-05-12 02:39:07 +00:00
}
list->push_back(param);
2017-05-12 02:39:07 +00:00
return;
}
int c = Children();
2017-05-12 02:39:07 +00:00
if(c >= 1) {
a->ParamsUsedList(list);
if(c >= 2) b->ParamsUsedList(list);
}
}
bool Expr::DependsOn(hParam p) const {
if(op == Op::PARAM) return (parh == p);
if(op == Op::PARAM_PTR) return (parp->h == p);
int c = Children();
if(c == 1) return a->DependsOn(p);
if(c == 2) return a->DependsOn(p) || b->DependsOn(p);
return false;
}
bool Expr::Tol(double a, double b) {
return fabs(a - b) < 0.001;
}
bool Expr::IsZeroConst() const {
return op == Op::CONSTANT && EXACT(v == 0.0);
}
Expr *Expr::FoldConstants() {
Expr *n = AllocExpr();
*n = *this;
int c = Children();
if(c >= 1) n->a = a->FoldConstants();
if(c >= 2) n->b = b->FoldConstants();
switch(op) {
Convert all enumerations to use `enum class`. Specifically, take the old code that looks like this: class Foo { enum { X = 1, Y = 2 }; int kind; } ... foo.kind = Foo::X; ... and convert it to this: class Foo { enum class Kind : uint32_t { X = 1, Y = 2 }; Kind kind; } ... foo.kind = Foo::Kind::X; (In some cases the enumeration would not be in the class namespace, such as when it is generally useful.) The benefits are as follows: * The type of the field gives a clear indication of intent, both to humans and tools (such as binding generators). * The compiler is able to automatically warn when a switch is not exhaustive; but this is currently suppressed by the default: ssassert(false, ...) idiom. * Integers and plain enums are weakly type checked: they implicitly convert into each other. This can hide bugs where type conversion is performed but not intended. Enum classes are strongly type checked. * Plain enums pollute parent namespaces; enum classes do not. Almost every defined enum we have already has a kind of ad-hoc namespacing via `NAMESPACE_`, which is now explicit. * Plain enums do not have a well-defined ABI size, which is important for bindings. Enum classes can have it, if specified. We specify the base type for all enums as uint32_t, which is a safe choice and allows us to not change the numeric values of any variants. This commit introduces absolutely no functional change to the code, just renaming and change of types. It handles almost all cases, except GraphicsWindow::pending.operation, which needs minor functional change.
2016-05-20 08:31:20 +00:00
case Op::PARAM_PTR:
case Op::PARAM:
case Op::CONSTANT:
case Op::VARIABLE:
break;
Convert all enumerations to use `enum class`. Specifically, take the old code that looks like this: class Foo { enum { X = 1, Y = 2 }; int kind; } ... foo.kind = Foo::X; ... and convert it to this: class Foo { enum class Kind : uint32_t { X = 1, Y = 2 }; Kind kind; } ... foo.kind = Foo::Kind::X; (In some cases the enumeration would not be in the class namespace, such as when it is generally useful.) The benefits are as follows: * The type of the field gives a clear indication of intent, both to humans and tools (such as binding generators). * The compiler is able to automatically warn when a switch is not exhaustive; but this is currently suppressed by the default: ssassert(false, ...) idiom. * Integers and plain enums are weakly type checked: they implicitly convert into each other. This can hide bugs where type conversion is performed but not intended. Enum classes are strongly type checked. * Plain enums pollute parent namespaces; enum classes do not. Almost every defined enum we have already has a kind of ad-hoc namespacing via `NAMESPACE_`, which is now explicit. * Plain enums do not have a well-defined ABI size, which is important for bindings. Enum classes can have it, if specified. We specify the base type for all enums as uint32_t, which is a safe choice and allows us to not change the numeric values of any variants. This commit introduces absolutely no functional change to the code, just renaming and change of types. It handles almost all cases, except GraphicsWindow::pending.operation, which needs minor functional change.
2016-05-20 08:31:20 +00:00
case Op::MINUS:
case Op::TIMES:
case Op::DIV:
case Op::PLUS:
// If both ops are known, then we can evaluate immediately
Convert all enumerations to use `enum class`. Specifically, take the old code that looks like this: class Foo { enum { X = 1, Y = 2 }; int kind; } ... foo.kind = Foo::X; ... and convert it to this: class Foo { enum class Kind : uint32_t { X = 1, Y = 2 }; Kind kind; } ... foo.kind = Foo::Kind::X; (In some cases the enumeration would not be in the class namespace, such as when it is generally useful.) The benefits are as follows: * The type of the field gives a clear indication of intent, both to humans and tools (such as binding generators). * The compiler is able to automatically warn when a switch is not exhaustive; but this is currently suppressed by the default: ssassert(false, ...) idiom. * Integers and plain enums are weakly type checked: they implicitly convert into each other. This can hide bugs where type conversion is performed but not intended. Enum classes are strongly type checked. * Plain enums pollute parent namespaces; enum classes do not. Almost every defined enum we have already has a kind of ad-hoc namespacing via `NAMESPACE_`, which is now explicit. * Plain enums do not have a well-defined ABI size, which is important for bindings. Enum classes can have it, if specified. We specify the base type for all enums as uint32_t, which is a safe choice and allows us to not change the numeric values of any variants. This commit introduces absolutely no functional change to the code, just renaming and change of types. It handles almost all cases, except GraphicsWindow::pending.operation, which needs minor functional change.
2016-05-20 08:31:20 +00:00
if(n->a->op == Op::CONSTANT && n->b->op == Op::CONSTANT) {
double nv = n->Eval();
Convert all enumerations to use `enum class`. Specifically, take the old code that looks like this: class Foo { enum { X = 1, Y = 2 }; int kind; } ... foo.kind = Foo::X; ... and convert it to this: class Foo { enum class Kind : uint32_t { X = 1, Y = 2 }; Kind kind; } ... foo.kind = Foo::Kind::X; (In some cases the enumeration would not be in the class namespace, such as when it is generally useful.) The benefits are as follows: * The type of the field gives a clear indication of intent, both to humans and tools (such as binding generators). * The compiler is able to automatically warn when a switch is not exhaustive; but this is currently suppressed by the default: ssassert(false, ...) idiom. * Integers and plain enums are weakly type checked: they implicitly convert into each other. This can hide bugs where type conversion is performed but not intended. Enum classes are strongly type checked. * Plain enums pollute parent namespaces; enum classes do not. Almost every defined enum we have already has a kind of ad-hoc namespacing via `NAMESPACE_`, which is now explicit. * Plain enums do not have a well-defined ABI size, which is important for bindings. Enum classes can have it, if specified. We specify the base type for all enums as uint32_t, which is a safe choice and allows us to not change the numeric values of any variants. This commit introduces absolutely no functional change to the code, just renaming and change of types. It handles almost all cases, except GraphicsWindow::pending.operation, which needs minor functional change.
2016-05-20 08:31:20 +00:00
n->op = Op::CONSTANT;
n->v = nv;
break;
}
// x + 0 = 0 + x = x
Convert all enumerations to use `enum class`. Specifically, take the old code that looks like this: class Foo { enum { X = 1, Y = 2 }; int kind; } ... foo.kind = Foo::X; ... and convert it to this: class Foo { enum class Kind : uint32_t { X = 1, Y = 2 }; Kind kind; } ... foo.kind = Foo::Kind::X; (In some cases the enumeration would not be in the class namespace, such as when it is generally useful.) The benefits are as follows: * The type of the field gives a clear indication of intent, both to humans and tools (such as binding generators). * The compiler is able to automatically warn when a switch is not exhaustive; but this is currently suppressed by the default: ssassert(false, ...) idiom. * Integers and plain enums are weakly type checked: they implicitly convert into each other. This can hide bugs where type conversion is performed but not intended. Enum classes are strongly type checked. * Plain enums pollute parent namespaces; enum classes do not. Almost every defined enum we have already has a kind of ad-hoc namespacing via `NAMESPACE_`, which is now explicit. * Plain enums do not have a well-defined ABI size, which is important for bindings. Enum classes can have it, if specified. We specify the base type for all enums as uint32_t, which is a safe choice and allows us to not change the numeric values of any variants. This commit introduces absolutely no functional change to the code, just renaming and change of types. It handles almost all cases, except GraphicsWindow::pending.operation, which needs minor functional change.
2016-05-20 08:31:20 +00:00
if(op == Op::PLUS && n->b->op == Op::CONSTANT && Tol(n->b->v, 0)) {
*n = *(n->a); break;
}
Convert all enumerations to use `enum class`. Specifically, take the old code that looks like this: class Foo { enum { X = 1, Y = 2 }; int kind; } ... foo.kind = Foo::X; ... and convert it to this: class Foo { enum class Kind : uint32_t { X = 1, Y = 2 }; Kind kind; } ... foo.kind = Foo::Kind::X; (In some cases the enumeration would not be in the class namespace, such as when it is generally useful.) The benefits are as follows: * The type of the field gives a clear indication of intent, both to humans and tools (such as binding generators). * The compiler is able to automatically warn when a switch is not exhaustive; but this is currently suppressed by the default: ssassert(false, ...) idiom. * Integers and plain enums are weakly type checked: they implicitly convert into each other. This can hide bugs where type conversion is performed but not intended. Enum classes are strongly type checked. * Plain enums pollute parent namespaces; enum classes do not. Almost every defined enum we have already has a kind of ad-hoc namespacing via `NAMESPACE_`, which is now explicit. * Plain enums do not have a well-defined ABI size, which is important for bindings. Enum classes can have it, if specified. We specify the base type for all enums as uint32_t, which is a safe choice and allows us to not change the numeric values of any variants. This commit introduces absolutely no functional change to the code, just renaming and change of types. It handles almost all cases, except GraphicsWindow::pending.operation, which needs minor functional change.
2016-05-20 08:31:20 +00:00
if(op == Op::PLUS && n->a->op == Op::CONSTANT && Tol(n->a->v, 0)) {
*n = *(n->b); break;
}
// 1*x = x*1 = x
Convert all enumerations to use `enum class`. Specifically, take the old code that looks like this: class Foo { enum { X = 1, Y = 2 }; int kind; } ... foo.kind = Foo::X; ... and convert it to this: class Foo { enum class Kind : uint32_t { X = 1, Y = 2 }; Kind kind; } ... foo.kind = Foo::Kind::X; (In some cases the enumeration would not be in the class namespace, such as when it is generally useful.) The benefits are as follows: * The type of the field gives a clear indication of intent, both to humans and tools (such as binding generators). * The compiler is able to automatically warn when a switch is not exhaustive; but this is currently suppressed by the default: ssassert(false, ...) idiom. * Integers and plain enums are weakly type checked: they implicitly convert into each other. This can hide bugs where type conversion is performed but not intended. Enum classes are strongly type checked. * Plain enums pollute parent namespaces; enum classes do not. Almost every defined enum we have already has a kind of ad-hoc namespacing via `NAMESPACE_`, which is now explicit. * Plain enums do not have a well-defined ABI size, which is important for bindings. Enum classes can have it, if specified. We specify the base type for all enums as uint32_t, which is a safe choice and allows us to not change the numeric values of any variants. This commit introduces absolutely no functional change to the code, just renaming and change of types. It handles almost all cases, except GraphicsWindow::pending.operation, which needs minor functional change.
2016-05-20 08:31:20 +00:00
if(op == Op::TIMES && n->b->op == Op::CONSTANT && Tol(n->b->v, 1)) {
*n = *(n->a); break;
}
Convert all enumerations to use `enum class`. Specifically, take the old code that looks like this: class Foo { enum { X = 1, Y = 2 }; int kind; } ... foo.kind = Foo::X; ... and convert it to this: class Foo { enum class Kind : uint32_t { X = 1, Y = 2 }; Kind kind; } ... foo.kind = Foo::Kind::X; (In some cases the enumeration would not be in the class namespace, such as when it is generally useful.) The benefits are as follows: * The type of the field gives a clear indication of intent, both to humans and tools (such as binding generators). * The compiler is able to automatically warn when a switch is not exhaustive; but this is currently suppressed by the default: ssassert(false, ...) idiom. * Integers and plain enums are weakly type checked: they implicitly convert into each other. This can hide bugs where type conversion is performed but not intended. Enum classes are strongly type checked. * Plain enums pollute parent namespaces; enum classes do not. Almost every defined enum we have already has a kind of ad-hoc namespacing via `NAMESPACE_`, which is now explicit. * Plain enums do not have a well-defined ABI size, which is important for bindings. Enum classes can have it, if specified. We specify the base type for all enums as uint32_t, which is a safe choice and allows us to not change the numeric values of any variants. This commit introduces absolutely no functional change to the code, just renaming and change of types. It handles almost all cases, except GraphicsWindow::pending.operation, which needs minor functional change.
2016-05-20 08:31:20 +00:00
if(op == Op::TIMES && n->a->op == Op::CONSTANT && Tol(n->a->v, 1)) {
*n = *(n->b); break;
}
// 0*x = x*0 = 0
Convert all enumerations to use `enum class`. Specifically, take the old code that looks like this: class Foo { enum { X = 1, Y = 2 }; int kind; } ... foo.kind = Foo::X; ... and convert it to this: class Foo { enum class Kind : uint32_t { X = 1, Y = 2 }; Kind kind; } ... foo.kind = Foo::Kind::X; (In some cases the enumeration would not be in the class namespace, such as when it is generally useful.) The benefits are as follows: * The type of the field gives a clear indication of intent, both to humans and tools (such as binding generators). * The compiler is able to automatically warn when a switch is not exhaustive; but this is currently suppressed by the default: ssassert(false, ...) idiom. * Integers and plain enums are weakly type checked: they implicitly convert into each other. This can hide bugs where type conversion is performed but not intended. Enum classes are strongly type checked. * Plain enums pollute parent namespaces; enum classes do not. Almost every defined enum we have already has a kind of ad-hoc namespacing via `NAMESPACE_`, which is now explicit. * Plain enums do not have a well-defined ABI size, which is important for bindings. Enum classes can have it, if specified. We specify the base type for all enums as uint32_t, which is a safe choice and allows us to not change the numeric values of any variants. This commit introduces absolutely no functional change to the code, just renaming and change of types. It handles almost all cases, except GraphicsWindow::pending.operation, which needs minor functional change.
2016-05-20 08:31:20 +00:00
if(op == Op::TIMES && n->b->op == Op::CONSTANT && Tol(n->b->v, 0)) {
n->op = Op::CONSTANT; n->v = 0; break;
}
Convert all enumerations to use `enum class`. Specifically, take the old code that looks like this: class Foo { enum { X = 1, Y = 2 }; int kind; } ... foo.kind = Foo::X; ... and convert it to this: class Foo { enum class Kind : uint32_t { X = 1, Y = 2 }; Kind kind; } ... foo.kind = Foo::Kind::X; (In some cases the enumeration would not be in the class namespace, such as when it is generally useful.) The benefits are as follows: * The type of the field gives a clear indication of intent, both to humans and tools (such as binding generators). * The compiler is able to automatically warn when a switch is not exhaustive; but this is currently suppressed by the default: ssassert(false, ...) idiom. * Integers and plain enums are weakly type checked: they implicitly convert into each other. This can hide bugs where type conversion is performed but not intended. Enum classes are strongly type checked. * Plain enums pollute parent namespaces; enum classes do not. Almost every defined enum we have already has a kind of ad-hoc namespacing via `NAMESPACE_`, which is now explicit. * Plain enums do not have a well-defined ABI size, which is important for bindings. Enum classes can have it, if specified. We specify the base type for all enums as uint32_t, which is a safe choice and allows us to not change the numeric values of any variants. This commit introduces absolutely no functional change to the code, just renaming and change of types. It handles almost all cases, except GraphicsWindow::pending.operation, which needs minor functional change.
2016-05-20 08:31:20 +00:00
if(op == Op::TIMES && n->a->op == Op::CONSTANT && Tol(n->a->v, 0)) {
n->op = Op::CONSTANT; n->v = 0; break;
}
break;
Convert all enumerations to use `enum class`. Specifically, take the old code that looks like this: class Foo { enum { X = 1, Y = 2 }; int kind; } ... foo.kind = Foo::X; ... and convert it to this: class Foo { enum class Kind : uint32_t { X = 1, Y = 2 }; Kind kind; } ... foo.kind = Foo::Kind::X; (In some cases the enumeration would not be in the class namespace, such as when it is generally useful.) The benefits are as follows: * The type of the field gives a clear indication of intent, both to humans and tools (such as binding generators). * The compiler is able to automatically warn when a switch is not exhaustive; but this is currently suppressed by the default: ssassert(false, ...) idiom. * Integers and plain enums are weakly type checked: they implicitly convert into each other. This can hide bugs where type conversion is performed but not intended. Enum classes are strongly type checked. * Plain enums pollute parent namespaces; enum classes do not. Almost every defined enum we have already has a kind of ad-hoc namespacing via `NAMESPACE_`, which is now explicit. * Plain enums do not have a well-defined ABI size, which is important for bindings. Enum classes can have it, if specified. We specify the base type for all enums as uint32_t, which is a safe choice and allows us to not change the numeric values of any variants. This commit introduces absolutely no functional change to the code, just renaming and change of types. It handles almost all cases, except GraphicsWindow::pending.operation, which needs minor functional change.
2016-05-20 08:31:20 +00:00
case Op::SQRT:
case Op::SQUARE:
case Op::NEGATE:
case Op::SIN:
case Op::COS:
case Op::ASIN:
case Op::ACOS:
if(n->a->op == Op::CONSTANT) {
double nv = n->Eval();
Convert all enumerations to use `enum class`. Specifically, take the old code that looks like this: class Foo { enum { X = 1, Y = 2 }; int kind; } ... foo.kind = Foo::X; ... and convert it to this: class Foo { enum class Kind : uint32_t { X = 1, Y = 2 }; Kind kind; } ... foo.kind = Foo::Kind::X; (In some cases the enumeration would not be in the class namespace, such as when it is generally useful.) The benefits are as follows: * The type of the field gives a clear indication of intent, both to humans and tools (such as binding generators). * The compiler is able to automatically warn when a switch is not exhaustive; but this is currently suppressed by the default: ssassert(false, ...) idiom. * Integers and plain enums are weakly type checked: they implicitly convert into each other. This can hide bugs where type conversion is performed but not intended. Enum classes are strongly type checked. * Plain enums pollute parent namespaces; enum classes do not. Almost every defined enum we have already has a kind of ad-hoc namespacing via `NAMESPACE_`, which is now explicit. * Plain enums do not have a well-defined ABI size, which is important for bindings. Enum classes can have it, if specified. We specify the base type for all enums as uint32_t, which is a safe choice and allows us to not change the numeric values of any variants. This commit introduces absolutely no functional change to the code, just renaming and change of types. It handles almost all cases, except GraphicsWindow::pending.operation, which needs minor functional change.
2016-05-20 08:31:20 +00:00
n->op = Op::CONSTANT;
n->v = nv;
}
break;
}
return n;
}
void Expr::Substitute(hParam oldh, hParam newh) {
Convert all enumerations to use `enum class`. Specifically, take the old code that looks like this: class Foo { enum { X = 1, Y = 2 }; int kind; } ... foo.kind = Foo::X; ... and convert it to this: class Foo { enum class Kind : uint32_t { X = 1, Y = 2 }; Kind kind; } ... foo.kind = Foo::Kind::X; (In some cases the enumeration would not be in the class namespace, such as when it is generally useful.) The benefits are as follows: * The type of the field gives a clear indication of intent, both to humans and tools (such as binding generators). * The compiler is able to automatically warn when a switch is not exhaustive; but this is currently suppressed by the default: ssassert(false, ...) idiom. * Integers and plain enums are weakly type checked: they implicitly convert into each other. This can hide bugs where type conversion is performed but not intended. Enum classes are strongly type checked. * Plain enums pollute parent namespaces; enum classes do not. Almost every defined enum we have already has a kind of ad-hoc namespacing via `NAMESPACE_`, which is now explicit. * Plain enums do not have a well-defined ABI size, which is important for bindings. Enum classes can have it, if specified. We specify the base type for all enums as uint32_t, which is a safe choice and allows us to not change the numeric values of any variants. This commit introduces absolutely no functional change to the code, just renaming and change of types. It handles almost all cases, except GraphicsWindow::pending.operation, which needs minor functional change.
2016-05-20 08:31:20 +00:00
ssassert(op != Op::PARAM_PTR, "Expected an expression that refer to params via handles");
if(op == Op::PARAM && parh == oldh) {
parh = newh;
}
int c = Children();
if(c >= 1) a->Substitute(oldh, newh);
if(c >= 2) b->Substitute(oldh, newh);
}
//-----------------------------------------------------------------------------
// If the expression references only one parameter that appears in pl, then
// return that parameter. If no param is referenced, then return NO_PARAMS.
// If multiple params are referenced, then return MULTIPLE_PARAMS.
//-----------------------------------------------------------------------------
const hParam Expr::NO_PARAMS = { 0 };
const hParam Expr::MULTIPLE_PARAMS = { 1 };
hParam Expr::ReferencedParams(ParamList *pl) const {
Convert all enumerations to use `enum class`. Specifically, take the old code that looks like this: class Foo { enum { X = 1, Y = 2 }; int kind; } ... foo.kind = Foo::X; ... and convert it to this: class Foo { enum class Kind : uint32_t { X = 1, Y = 2 }; Kind kind; } ... foo.kind = Foo::Kind::X; (In some cases the enumeration would not be in the class namespace, such as when it is generally useful.) The benefits are as follows: * The type of the field gives a clear indication of intent, both to humans and tools (such as binding generators). * The compiler is able to automatically warn when a switch is not exhaustive; but this is currently suppressed by the default: ssassert(false, ...) idiom. * Integers and plain enums are weakly type checked: they implicitly convert into each other. This can hide bugs where type conversion is performed but not intended. Enum classes are strongly type checked. * Plain enums pollute parent namespaces; enum classes do not. Almost every defined enum we have already has a kind of ad-hoc namespacing via `NAMESPACE_`, which is now explicit. * Plain enums do not have a well-defined ABI size, which is important for bindings. Enum classes can have it, if specified. We specify the base type for all enums as uint32_t, which is a safe choice and allows us to not change the numeric values of any variants. This commit introduces absolutely no functional change to the code, just renaming and change of types. It handles almost all cases, except GraphicsWindow::pending.operation, which needs minor functional change.
2016-05-20 08:31:20 +00:00
if(op == Op::PARAM) {
if(pl->FindByIdNoOops(parh)) {
return parh;
} else {
return NO_PARAMS;
}
}
Convert all enumerations to use `enum class`. Specifically, take the old code that looks like this: class Foo { enum { X = 1, Y = 2 }; int kind; } ... foo.kind = Foo::X; ... and convert it to this: class Foo { enum class Kind : uint32_t { X = 1, Y = 2 }; Kind kind; } ... foo.kind = Foo::Kind::X; (In some cases the enumeration would not be in the class namespace, such as when it is generally useful.) The benefits are as follows: * The type of the field gives a clear indication of intent, both to humans and tools (such as binding generators). * The compiler is able to automatically warn when a switch is not exhaustive; but this is currently suppressed by the default: ssassert(false, ...) idiom. * Integers and plain enums are weakly type checked: they implicitly convert into each other. This can hide bugs where type conversion is performed but not intended. Enum classes are strongly type checked. * Plain enums pollute parent namespaces; enum classes do not. Almost every defined enum we have already has a kind of ad-hoc namespacing via `NAMESPACE_`, which is now explicit. * Plain enums do not have a well-defined ABI size, which is important for bindings. Enum classes can have it, if specified. We specify the base type for all enums as uint32_t, which is a safe choice and allows us to not change the numeric values of any variants. This commit introduces absolutely no functional change to the code, just renaming and change of types. It handles almost all cases, except GraphicsWindow::pending.operation, which needs minor functional change.
2016-05-20 08:31:20 +00:00
ssassert(op != Op::PARAM_PTR, "Expected an expression that refer to params via handles");
int c = Children();
if(c == 0) {
return NO_PARAMS;
} else if(c == 1) {
return a->ReferencedParams(pl);
} else if(c == 2) {
hParam pa, pb;
pa = a->ReferencedParams(pl);
pb = b->ReferencedParams(pl);
if(pa == NO_PARAMS) {
return pb;
} else if(pb == NO_PARAMS) {
return pa;
} else if(pa == pb) {
return pa; // either, doesn't matter
} else {
return MULTIPLE_PARAMS;
}
} else ssassert(false, "Unexpected children count");
}
//-----------------------------------------------------------------------------
// Routines to pretty-print an expression. Mostly for debugging.
//-----------------------------------------------------------------------------
std::string Expr::Print() const {
char c;
switch(op) {
Convert all enumerations to use `enum class`. Specifically, take the old code that looks like this: class Foo { enum { X = 1, Y = 2 }; int kind; } ... foo.kind = Foo::X; ... and convert it to this: class Foo { enum class Kind : uint32_t { X = 1, Y = 2 }; Kind kind; } ... foo.kind = Foo::Kind::X; (In some cases the enumeration would not be in the class namespace, such as when it is generally useful.) The benefits are as follows: * The type of the field gives a clear indication of intent, both to humans and tools (such as binding generators). * The compiler is able to automatically warn when a switch is not exhaustive; but this is currently suppressed by the default: ssassert(false, ...) idiom. * Integers and plain enums are weakly type checked: they implicitly convert into each other. This can hide bugs where type conversion is performed but not intended. Enum classes are strongly type checked. * Plain enums pollute parent namespaces; enum classes do not. Almost every defined enum we have already has a kind of ad-hoc namespacing via `NAMESPACE_`, which is now explicit. * Plain enums do not have a well-defined ABI size, which is important for bindings. Enum classes can have it, if specified. We specify the base type for all enums as uint32_t, which is a safe choice and allows us to not change the numeric values of any variants. This commit introduces absolutely no functional change to the code, just renaming and change of types. It handles almost all cases, except GraphicsWindow::pending.operation, which needs minor functional change.
2016-05-20 08:31:20 +00:00
case Op::PARAM: return ssprintf("param(%08x)", parh.v);
case Op::PARAM_PTR: return ssprintf("param(p%08x)", parp->h.v);
Convert all enumerations to use `enum class`. Specifically, take the old code that looks like this: class Foo { enum { X = 1, Y = 2 }; int kind; } ... foo.kind = Foo::X; ... and convert it to this: class Foo { enum class Kind : uint32_t { X = 1, Y = 2 }; Kind kind; } ... foo.kind = Foo::Kind::X; (In some cases the enumeration would not be in the class namespace, such as when it is generally useful.) The benefits are as follows: * The type of the field gives a clear indication of intent, both to humans and tools (such as binding generators). * The compiler is able to automatically warn when a switch is not exhaustive; but this is currently suppressed by the default: ssassert(false, ...) idiom. * Integers and plain enums are weakly type checked: they implicitly convert into each other. This can hide bugs where type conversion is performed but not intended. Enum classes are strongly type checked. * Plain enums pollute parent namespaces; enum classes do not. Almost every defined enum we have already has a kind of ad-hoc namespacing via `NAMESPACE_`, which is now explicit. * Plain enums do not have a well-defined ABI size, which is important for bindings. Enum classes can have it, if specified. We specify the base type for all enums as uint32_t, which is a safe choice and allows us to not change the numeric values of any variants. This commit introduces absolutely no functional change to the code, just renaming and change of types. It handles almost all cases, except GraphicsWindow::pending.operation, which needs minor functional change.
2016-05-20 08:31:20 +00:00
case Op::CONSTANT: return ssprintf("%.3f", v);
case Op::VARIABLE: return "(var)";
Convert all enumerations to use `enum class`. Specifically, take the old code that looks like this: class Foo { enum { X = 1, Y = 2 }; int kind; } ... foo.kind = Foo::X; ... and convert it to this: class Foo { enum class Kind : uint32_t { X = 1, Y = 2 }; Kind kind; } ... foo.kind = Foo::Kind::X; (In some cases the enumeration would not be in the class namespace, such as when it is generally useful.) The benefits are as follows: * The type of the field gives a clear indication of intent, both to humans and tools (such as binding generators). * The compiler is able to automatically warn when a switch is not exhaustive; but this is currently suppressed by the default: ssassert(false, ...) idiom. * Integers and plain enums are weakly type checked: they implicitly convert into each other. This can hide bugs where type conversion is performed but not intended. Enum classes are strongly type checked. * Plain enums pollute parent namespaces; enum classes do not. Almost every defined enum we have already has a kind of ad-hoc namespacing via `NAMESPACE_`, which is now explicit. * Plain enums do not have a well-defined ABI size, which is important for bindings. Enum classes can have it, if specified. We specify the base type for all enums as uint32_t, which is a safe choice and allows us to not change the numeric values of any variants. This commit introduces absolutely no functional change to the code, just renaming and change of types. It handles almost all cases, except GraphicsWindow::pending.operation, which needs minor functional change.
2016-05-20 08:31:20 +00:00
case Op::PLUS: c = '+'; goto p;
case Op::MINUS: c = '-'; goto p;
case Op::TIMES: c = '*'; goto p;
case Op::DIV: c = '/'; goto p;
p:
return "(" + a->Print() + " " + c + " " + b->Print() + ")";
break;
Convert all enumerations to use `enum class`. Specifically, take the old code that looks like this: class Foo { enum { X = 1, Y = 2 }; int kind; } ... foo.kind = Foo::X; ... and convert it to this: class Foo { enum class Kind : uint32_t { X = 1, Y = 2 }; Kind kind; } ... foo.kind = Foo::Kind::X; (In some cases the enumeration would not be in the class namespace, such as when it is generally useful.) The benefits are as follows: * The type of the field gives a clear indication of intent, both to humans and tools (such as binding generators). * The compiler is able to automatically warn when a switch is not exhaustive; but this is currently suppressed by the default: ssassert(false, ...) idiom. * Integers and plain enums are weakly type checked: they implicitly convert into each other. This can hide bugs where type conversion is performed but not intended. Enum classes are strongly type checked. * Plain enums pollute parent namespaces; enum classes do not. Almost every defined enum we have already has a kind of ad-hoc namespacing via `NAMESPACE_`, which is now explicit. * Plain enums do not have a well-defined ABI size, which is important for bindings. Enum classes can have it, if specified. We specify the base type for all enums as uint32_t, which is a safe choice and allows us to not change the numeric values of any variants. This commit introduces absolutely no functional change to the code, just renaming and change of types. It handles almost all cases, except GraphicsWindow::pending.operation, which needs minor functional change.
2016-05-20 08:31:20 +00:00
case Op::NEGATE: return "(- " + a->Print() + ")";
case Op::SQRT: return "(sqrt " + a->Print() + ")";
case Op::SQUARE: return "(square " + a->Print() + ")";
case Op::SIN: return "(sin " + a->Print() + ")";
case Op::COS: return "(cos " + a->Print() + ")";
case Op::ASIN: return "(asin " + a->Print() + ")";
case Op::ACOS: return "(acos " + a->Print() + ")";
}
Enable exhaustive switch coverage warnings as an error, and use them. Specifically, this enables -Wswitch=error on GCC/Clang and its MSVC equivalent; the exact way it is handled varies slightly, but what they all have in common is that in a switch statement over an enumeration, any enumerand that is not explicitly (via case:) or implicitly (via default:) handled in the switch triggers an error. Moreover, we also change the switch statements in three ways: * Switch statements that ought to be extended every time a new enumerand is added (e.g. Entity::DrawOrGetDistance(), are changed to explicitly list every single enumerand, and not have a default: branch. Note that the assertions are kept because it is legal for a enumeration to have a value unlike any of its defined enumerands, and we can e.g. read garbage from a file, or an uninitialized variable. This requires some rearranging if a default: branch is undesired. * Switch statements that ought to only ever see a few select enumerands, are changed to always assert in the default: branch. * Switch statements that do something meaningful for a few enumerands, and ignore everything else, are changed to do nothing in a default: branch, under the assumption that changing them every time an enumerand is added or removed would just result in noise and catch no bugs. This commit also removes the {Request,Entity,Constraint}::UNKNOWN and Entity::DATUM_POINT enumerands, as those were just fancy names for zeroes. They mess up switch exhaustiveness checks and most of the time were not the best way to implement what they did anyway.
2016-05-25 06:55:50 +00:00
ssassert(false, "Unexpected operation");
}
//-----------------------------------------------------------------------------
// A parser; convert a string to an expression. Infix notation, with the
// usual shift/reduce approach. I had great hopes for user-entered eq
// constraints, but those don't seem very useful, so right now this is just
// to provide calculator type functionality wherever numbers are entered.
//-----------------------------------------------------------------------------
class ExprParser {
public:
enum class TokenType {
ERROR = 0,
PAREN_LEFT,
PAREN_RIGHT,
BINARY_OP,
UNARY_OP,
OPERAND,
END,
};
class Token {
public:
TokenType type;
Expr *expr;
static Token From(TokenType type = TokenType::ERROR, Expr *expr = NULL);
static Token From(TokenType type, Expr::Op op);
bool IsError() const { return type == TokenType::ERROR; }
};
std::string::const_iterator it, end;
std::vector<Token> stack;
char ReadChar();
char PeekChar();
std::string ReadWord();
void SkipSpace();
Token PopOperator(std::string *error);
Token PopOperand(std::string *error);
int Precedence(Token token);
Token LexNumber(std::string *error);
Token Lex(std::string *error);
bool Reduce(std::string *error);
bool Parse(std::string *error, size_t reduceUntil = 0);
static Expr *Parse(const std::string &input, std::string *error);
};
ExprParser::Token ExprParser::Token::From(TokenType type, Expr *expr) {
Token t;
t.type = type;
t.expr = expr;
return t;
}
ExprParser::Token ExprParser::Token::From(TokenType type, Expr::Op op) {
Token t;
t.type = type;
t.expr = Expr::AllocExpr();
t.expr->op = op;
return t;
}
char ExprParser::ReadChar() {
return *it++;
}
char ExprParser::PeekChar() {
if(it == end) {
return '\0';
} else {
return *it;
}
}
std::string ExprParser::ReadWord() {
std::string s;
while(char c = PeekChar()) {
if(!isalnum(c)) break;
s.push_back(ReadChar());
}
return s;
}
void ExprParser::SkipSpace() {
while(char c = PeekChar()) {
if(!isspace(c)) break;
ReadChar();
}
}
ExprParser::Token ExprParser::LexNumber(std::string *error) {
std::string s;
while(char c = PeekChar()) {
if(!((c >= '0' && c <= '9') || c == 'e' || c == 'E' || c == '.' || c == '_')) break;
if(c == '_') {
ReadChar();
continue;
}
s.push_back(ReadChar());
}
char *endptr;
double d = strtod(s.c_str(), &endptr);
Token t = Token::From();
if(endptr == s.c_str() + s.size()) {
t = Token::From(TokenType::OPERAND, Expr::Op::CONSTANT);
t.expr->v = d;
} else {
*error = "'" + s + "' is not a valid number";
}
return t;
}
ExprParser::Token ExprParser::Lex(std::string *error) {
SkipSpace();
Token t = Token::From();
char c = PeekChar();
if(isupper(c)) {
std::string n = ReadWord();
t = Token::From(TokenType::OPERAND, Expr::Op::VARIABLE);
} else if(isalpha(c)) {
std::string s = ReadWord();
if(s == "sqrt") {
t = Token::From(TokenType::UNARY_OP, Expr::Op::SQRT);
} else if(s == "square") {
t = Token::From(TokenType::UNARY_OP, Expr::Op::SQUARE);
} else if(s == "sin") {
t = Token::From(TokenType::UNARY_OP, Expr::Op::SIN);
} else if(s == "cos") {
t = Token::From(TokenType::UNARY_OP, Expr::Op::COS);
} else if(s == "asin") {
t = Token::From(TokenType::UNARY_OP, Expr::Op::ASIN);
} else if(s == "acos") {
t = Token::From(TokenType::UNARY_OP, Expr::Op::ACOS);
} else if(s == "pi") {
t = Token::From(TokenType::OPERAND, Expr::Op::CONSTANT);
t.expr->v = PI;
} else {
*error = "'" + s + "' is not a valid variable, function or constant";
}
} else if(isdigit(c) || c == '.') {
return LexNumber(error);
} else if(ispunct(c)) {
ReadChar();
if(c == '+') {
t = Token::From(TokenType::BINARY_OP, Expr::Op::PLUS);
} else if(c == '-') {
t = Token::From(TokenType::BINARY_OP, Expr::Op::MINUS);
} else if(c == '*') {
t = Token::From(TokenType::BINARY_OP, Expr::Op::TIMES);
} else if(c == '/') {
t = Token::From(TokenType::BINARY_OP, Expr::Op::DIV);
} else if(c == '(') {
t = Token::From(TokenType::PAREN_LEFT);
} else if(c == ')') {
t = Token::From(TokenType::PAREN_RIGHT);
} else {
*error = "'" + std::string(1, c) + "' is not a valid operator";
}
} else if(c == '\0') {
t = Token::From(TokenType::END);
} else {
*error = "Unexpected character '" + std::string(1, c) + "'";
}
return t;
}
ExprParser::Token ExprParser::PopOperand(std::string *error) {
Token t = Token::From();
if(stack.empty() || stack.back().type != TokenType::OPERAND) {
*error = "Expected an operand";
} else {
t = stack.back();
stack.pop_back();
}
return t;
}
ExprParser::Token ExprParser::PopOperator(std::string *error) {
Token t = Token::From();
if(stack.empty() || (stack.back().type != TokenType::UNARY_OP &&
stack.back().type != TokenType::BINARY_OP)) {
*error = "Expected an operator";
} else {
t = stack.back();
stack.pop_back();
}
return t;
}
int ExprParser::Precedence(Token t) {
ssassert(t.type == TokenType::BINARY_OP ||
t.type == TokenType::UNARY_OP ||
t.type == TokenType::OPERAND,
"Unexpected token type");
if(t.type == TokenType::UNARY_OP) {
return 30;
} else if(t.expr->op == Expr::Op::TIMES ||
t.expr->op == Expr::Op::DIV) {
return 20;
} else if(t.expr->op == Expr::Op::PLUS ||
t.expr->op == Expr::Op::MINUS) {
return 10;
} else if(t.type == TokenType::OPERAND) {
return 0;
} else ssassert(false, "Unexpected operator");
}
bool ExprParser::Reduce(std::string *error) {
Token a = PopOperand(error);
if(a.IsError()) return false;
Token op = PopOperator(error);
if(op.IsError()) return false;
Token r = Token::From(TokenType::OPERAND);
switch(op.type) {
case TokenType::BINARY_OP: {
Token b = PopOperand(error);
if(b.IsError()) return false;
r.expr = b.expr->AnyOp(op.expr->op, a.expr);
break;
}
case TokenType::UNARY_OP: {
Expr *e = a.expr;
switch(op.expr->op) {
case Expr::Op::NEGATE: e = e->Negate(); break;
case Expr::Op::SQRT: e = e->Sqrt(); break;
case Expr::Op::SQUARE: e = e->Times(e); break;
case Expr::Op::SIN: e = e->Times(Expr::From(PI/180))->Sin(); break;
case Expr::Op::COS: e = e->Times(Expr::From(PI/180))->Cos(); break;
case Expr::Op::ASIN: e = e->ASin()->Times(Expr::From(180/PI)); break;
case Expr::Op::ACOS: e = e->ACos()->Times(Expr::From(180/PI)); break;
default: ssassert(false, "Unexpected unary operator");
}
r.expr = e;
break;
}
default: ssassert(false, "Unexpected operator");
}
stack.push_back(r);
return true;
}
bool ExprParser::Parse(std::string *error, size_t reduceUntil) {
while(true) {
Token t = Lex(error);
switch(t.type) {
case TokenType::ERROR:
return false;
case TokenType::END:
case TokenType::PAREN_RIGHT:
while(stack.size() > 1 + reduceUntil) {
if(!Reduce(error)) return false;
}
if(t.type == TokenType::PAREN_RIGHT) {
stack.push_back(t);
}
return true;
case TokenType::PAREN_LEFT: {
// sub-expression
if(!Parse(error, /*reduceUntil=*/stack.size())) return false;
if(stack.empty() || stack.back().type != TokenType::PAREN_RIGHT) {
*error = "Expected ')'";
return false;
}
stack.pop_back();
break;
}
case TokenType::BINARY_OP:
if((stack.size() > reduceUntil && stack.back().type != TokenType::OPERAND) ||
stack.size() == reduceUntil) {
if(t.expr->op == Expr::Op::MINUS) {
t.type = TokenType::UNARY_OP;
t.expr->op = Expr::Op::NEGATE;
stack.push_back(t);
break;
}
}
while(stack.size() > 1 + reduceUntil &&
Precedence(t) <= Precedence(stack[stack.size() - 2])) {
if(!Reduce(error)) return false;
}
stack.push_back(t);
break;
case TokenType::UNARY_OP:
case TokenType::OPERAND:
stack.push_back(t);
break;
}
}
return true;
}
Expr *ExprParser::Parse(const std::string &input, std::string *error) {
ExprParser parser;
parser.it = input.cbegin();
parser.end = input.cend();
if(!parser.Parse(error)) return NULL;
Token r = parser.PopOperand(error);
if(r.IsError()) return NULL;
return r.expr;
}
Expr *Expr::Parse(const std::string &input, std::string *error) {
return ExprParser::Parse(input, error);
}
Expr *Expr::From(const std::string &input, bool popUpError) {
std::string error;
Expr *e = ExprParser::Parse(input, &error);
if(!e) {
dbp("Parse/lex error: %s", error.c_str());
if(popUpError) {
Error("Not a valid number or expression: '%s'.\n%s.",
input.c_str(), error.c_str());
}
}
return e;
}