How to implement a visitor pattern over an AST
Hello, I have an AST successfully building, generating text file, and subsequently Spirit Qi parsing for verification purposes. Now I want to implement a visitor pattern over that AST, somehow... For "calculators" and such, "evaluating" those expressions is one thing; however, this AST is not that. It is closer akin to a parsed Xml or Json model/tree. In other words, I'd like for it to feel sort of like an iterator if that's even possible. It should have contextual awareness where it is at all times, etc, perhaps "incrementing" is a depth first analysis. In particular I have in mind to iterate the expected and actual AST results in order to perform the verification. In other words, I might do something like this: auto expected_it = ast_visitor{&expected}; auto actual_it = ast_visitor{&actual}; REQUIRE(expected_it); REQUIRE(actual_it); REQUIRE(*expected_it == *actual_it); // Rinse and repeat ++expected_it; ++actual_it; REQUIRE(*expected_it == *actual_it); I wonder if something like this is even possible from the Boost libraries? Or interested to hear any insights on the subject. For that matter, I wonder it is more appropriate to simply implement the logical operators. Thanks! Michael Powell
AMDG On 12/10/2018 12:04 PM, Michael Powell via Boost-users wrote:
I have an AST successfully building, generating text file, and subsequently Spirit Qi parsing for verification purposes.
Now I want to implement a visitor pattern over that AST, somehow... For "calculators" and such, "evaluating" those expressions is one thing; however, this AST is not that. It is closer akin to a parsed Xml or Json model/tree.
In other words, I'd like for it to feel sort of like an iterator if that's even possible. It should have contextual awareness where it is at all times, etc, perhaps "incrementing" is a depth first analysis.
It's possible to do this, but making an iterator over a tree is quite a bit more complex than just using a depth-first visitation whenever you need to process the tree.
In particular I have in mind to iterate the expected and actual AST results in order to perform the verification.
For this case, the simplest method is a binary visitor. For example, variant equality looks something like this: struct equal_to_visitor { using result_type = void; template<class T> bool operator()(const T& lhs, const T& rhs) { return lhs == rhs; } template<class T, class U> bool operator()(const T&, const U&) { return false; } }; ... bool operator==(const variant<...>& lhs, const variant<...> rhs) { return boost::apply_visitor(equal_to_visitor{}, lhs, rhs); } Note that since we always return false if the types are different, a slight variation can make this work with a unary visitor.
In other words, I might do something like this:
auto expected_it = ast_visitor{&expected}; auto actual_it = ast_visitor{&actual};
REQUIRE(expected_it); REQUIRE(actual_it);
REQUIRE(*expected_it == *actual_it);
// Rinse and repeat ++expected_it; ++actual_it; REQUIRE(*expected_it == *actual_it);
I wonder if something like this is even possible from the Boost libraries? Or interested to hear any insights on the subject.
For that matter, I wonder it is more appropriate to simply implement the logical operators.
In Christ, Steven Watanabe
On Mon, Dec 10, 2018 at 3:00 PM Steven Watanabe via Boost-users <boost-users@lists.boost.org> wrote:
AMDG
On 12/10/2018 12:04 PM, Michael Powell via Boost-users wrote:
I have an AST successfully building, generating text file, and subsequently Spirit Qi parsing for verification purposes.
Now I want to implement a visitor pattern over that AST, somehow... For "calculators" and such, "evaluating" those expressions is one thing; however, this AST is not that. It is closer akin to a parsed Xml or Json model/tree.
In other words, I'd like for it to feel sort of like an iterator if that's even possible. It should have contextual awareness where it is at all times, etc, perhaps "incrementing" is a depth first analysis.
It's possible to do this, but making an iterator over a tree is quite a bit more complex than just using a depth-first visitation whenever you need to process the tree.
It is; I actually have a similar pattern in a "string rendering visitor" in which I have internally specialized template methods for the particular nodes of visitation. One might call that a unary visitor, given a single operand. In this case, I rinse and repeat given the binary case, two operands: template<typename T> bool equals(const T& x, const T& y) const { const auto message = "Comparator type '" + std::string(typeid(T).name()) + "' unsupported"; throw std::exception(message.c_str()); } template<> bool equals<ast::TopLevelType>(const ast::TopLevelType& x, const ast::TopLevelType& y) const { // ... } And so on... Which also guarantees that I am also walking both tree's at the same moment and in like fashion.
In particular I have in mind to iterate the expected and actual AST results in order to perform the verification.
For this case, the simplest method is a binary visitor.
For example, variant equality looks something like this:
struct equal_to_visitor { using result_type = void; template<class T> bool operator()(const T& lhs, const T& rhs) { return lhs == rhs; } template<class T, class U> bool operator()(const T&, const U&) { return false; } }; ... bool operator==(const variant<...>& lhs, const variant<...> rhs) { return boost::apply_visitor(equal_to_visitor{}, lhs, rhs); }
Note that since we always return false if the types are different, a slight variation can make this work with a unary visitor.
In other words, I might do something like this:
auto expected_it = ast_visitor{&expected}; auto actual_it = ast_visitor{&actual};
REQUIRE(expected_it); REQUIRE(actual_it);
REQUIRE(*expected_it == *actual_it);
// Rinse and repeat ++expected_it; ++actual_it; REQUIRE(*expected_it == *actual_it);
I wonder if something like this is even possible from the Boost libraries? Or interested to hear any insights on the subject.
For that matter, I wonder it is more appropriate to simply implement the logical operators.
In Christ, Steven Watanabe _______________________________________________ Boost-users mailing list Boost-users@lists.boost.org https://lists.boost.org/mailman/listinfo.cgi/boost-users
AMDG On 12/10/2018 01:14 PM, Michael Powell via Boost-users wrote:
On Mon, Dec 10, 2018 at 3:00 PM Steven Watanabe via Boost-users <boost-users@lists.boost.org> wrote:
On 12/10/2018 12:04 PM, Michael Powell via Boost-users wrote:
I have an AST successfully building, generating text file, and subsequently Spirit Qi parsing for verification purposes.
Now I want to implement a visitor pattern over that AST, somehow... For "calculators" and such, "evaluating" those expressions is one thing; however, this AST is not that. It is closer akin to a parsed Xml or Json model/tree.
In other words, I'd like for it to feel sort of like an iterator if that's even possible. It should have contextual awareness where it is at all times, etc, perhaps "incrementing" is a depth first analysis.
It's possible to do this, but making an iterator over a tree is quite a bit more complex than just using a depth-first visitation whenever you need to process the tree.
It is; I actually have a similar pattern in a "string rendering visitor" in which I have internally specialized template methods for the particular nodes of visitation. One might call that a unary visitor, given a single operand.
In this case, I rinse and repeat given the binary case, two operands:
template<typename T> bool equals(const T& x, const T& y) const { const auto message = "Comparator type '" + std::string(typeid(T).name()) + "' unsupported"; throw std::exception(message.c_str()); }
template<> bool equals<ast::TopLevelType>(const ast::TopLevelType& x, const ast::TopLevelType& y) const { // ... }
And so on... Which also guarantees that I am also walking both tree's at the same moment and in like fashion.
Yep. Although you can implement tree iterators using coroutines as demonstrated here: https://www.boost.org/doc/libs/1_68_0/libs/coroutine/doc/html/coroutine/moti... you lose information about the tree structure. This is a good thing in that example, but not so much for your use case. In Christ, Steven Watanabe
On 12/10/18 2:14 PM, Michael Powell via Boost-users wrote:
On Mon, Dec 10, 2018 at 3:00 PM Steven Watanabe via Boost-users <boost-users@lists.boost.org> wrote:
AMDG
On 12/10/2018 12:04 PM, Michael Powell via Boost-users wrote:
I have an AST successfully building, generating text file, and subsequently Spirit Qi parsing for verification purposes.
Now I want to implement a visitor pattern over that AST, somehow... For "calculators" and such, "evaluating" those expressions is one thing; however, this AST is not that. It is closer akin to a parsed Xml or Json model/tree.
In other words, I'd like for it to feel sort of like an iterator if that's even possible. It should have contextual awareness where it is at all times, etc, perhaps "incrementing" is a depth first analysis.
It's possible to do this, but making an iterator over a tree is quite a bit more complex than just using a depth-first visitation whenever you need to process the tree.
It is; I actually have a similar pattern in a "string rendering visitor" in which I have internally specialized template methods for the particular nodes of visitation. One might call that a unary visitor, given a single operand.
In this case, I rinse and repeat given the binary case, two operands:
template<typename T> bool equals(const T& x, const T& y) const { const auto message = "Comparator type '" + std::string(typeid(T).name()) + "' unsupported"; throw std::exception(message.c_str()); }
template<> bool equals<ast::TopLevelType>(const ast::TopLevelType& x, const ast::TopLevelType& y) const { // ... }
And so on... Which also guarantees that I am also walking both tree's at the same moment and in like fashion.
[snip] Hi Michael, I'm spitballing here, but my novice view of karma is that it takes an AST and, like your "string rendering visitor" turns it into a string on std::cout or some other ostream. So, I'm guessing you could look at karma to see how it does it, and duplicate that method, only instead of rendering to a ostream, you'd just "render" as a bool result. Again, I could largely misunderstand what karma does, so take the above with a "large grain of salt". -regards, Larry
On Mon, Dec 10, 2018 at 7:19 PM Larry Evans via Boost-users <boost-users@lists.boost.org> wrote:
On 12/10/18 2:14 PM, Michael Powell via Boost-users wrote:
On Mon, Dec 10, 2018 at 3:00 PM Steven Watanabe via Boost-users <boost-users@lists.boost.org> wrote:
AMDG
On 12/10/2018 12:04 PM, Michael Powell via Boost-users wrote:
I have an AST successfully building, generating text file, and subsequently Spirit Qi parsing for verification purposes.
Now I want to implement a visitor pattern over that AST, somehow... For "calculators" and such, "evaluating" those expressions is one thing; however, this AST is not that. It is closer akin to a parsed Xml or Json model/tree.
In other words, I'd like for it to feel sort of like an iterator if that's even possible. It should have contextual awareness where it is at all times, etc, perhaps "incrementing" is a depth first analysis.
It's possible to do this, but making an iterator over a tree is quite a bit more complex than just using a depth-first visitation whenever you need to process the tree.
It is; I actually have a similar pattern in a "string rendering visitor" in which I have internally specialized template methods for the particular nodes of visitation. One might call that a unary visitor, given a single operand.
In this case, I rinse and repeat given the binary case, two operands:
template<typename T> bool equals(const T& x, const T& y) const { const auto message = "Comparator type '" + std::string(typeid(T).name()) + "' unsupported"; throw std::exception(message.c_str()); }
template<> bool equals<ast::TopLevelType>(const ast::TopLevelType& x, const ast::TopLevelType& y) const { // ... }
And so on... Which also guarantees that I am also walking both tree's at the same moment and in like fashion.
[snip] Hi Michael,
I'm spitballing here, but my novice view of karma is that it takes an AST and, like your "string rendering visitor" turns it into a string on std::cout or some other ostream. So, I'm guessing you could look at karma to see how it does it, and duplicate that method, only instead of rendering to a ostream, you'd just "render" as a bool result.
Again, I could largely misunderstand what karma does, so take the above with a "large grain of salt".
No worries, yes, I did consider Karma for this purpose, but I figured it may be "simple enough" to approach it otherwise. I may change my mind downstream, however. ;-)
-regards, Larry
_______________________________________________ Boost-users mailing list Boost-users@lists.boost.org https://lists.boost.org/mailman/listinfo.cgi/boost-users
On Mon, Dec 10, 2018 at 3:15 PM Michael Powell via Boost-users < boost-users@lists.boost.org> wrote: And so on... Which also guarantees that I am also walking both tree's
at the same moment and in like fashion.
For comparison, you might consider the use of Boost.Coroutine to allow you to interleave recursive steps through your two trees: https://www.youtube.com/watch?v=3SvkWY7JSeY (originally discussed here:) https://groups.google.com/forum/#!msg/boost-list/AZQgtk6jMho/fIFowJ72AgAJ
On Tue, Dec 18, 2018 at 6:02 PM Nat Goodspeed via Boost-users <boost-users@lists.boost.org> wrote:
On Mon, Dec 10, 2018 at 3:15 PM Michael Powell via Boost-users <boost-users@lists.boost.org> wrote:
And so on... Which also guarantees that I am also walking both tree's at the same moment and in like fashion.
For comparison, you might consider the use of Boost.Coroutine to allow you to interleave recursive steps through your two trees: https://www.youtube.com/watch?v=3SvkWY7JSeY (originally discussed here:) https://groups.google.com/forum/#!msg/boost-list/AZQgtk6jMho/fIFowJ72AgAJ
That's an interesting idea, but I'm afraid it was my fault for not communicating that clearly enough. There is little (if any) similarity between "nodes" in my tree/graph, apart from thin slices, like "has name" or "has number", etc. Additionally, rather than an explicit connection, "edge", the connectivity between AST nodes is implicit; the existence of which via either boost::variant or std::vector inclusion. Along these lines. Appreciate the feedback, however.
_______________________________________________ Boost-users mailing list Boost-users@lists.boost.org https://lists.boost.org/mailman/listinfo.cgi/boost-users
participants (5)
-
Larry Evans -
Michael Powell -
Nat Goodspeed -
Steven Watanabe