| /* |
| * Copyright (C) 2020 CESNET, https://photonics.cesnet.cz/ |
| * |
| * Written by Václav Kubernát <kubernat@cesnet.cz> |
| * |
| */ |
| |
| #pragma once |
| |
| #include <boost/spirit/home/x3.hpp> |
| #include "ast_handlers.hpp" |
| #include "common_parsers.hpp" |
| #include "leaf_data.hpp" |
| |
| namespace x3 = boost::spirit::x3; |
| |
| x3::rule<cdPath_class, dataPath_> const cdPath = "cdPath"; |
| x3::rule<getPath_class, decltype(get_::m_path)> const getPath = "getPath"; |
| x3::rule<rpcPath_class, dataPath_> const rpcPath = "rpcPath"; |
| x3::rule<presenceContainerPath_class, dataPath_> const presenceContainerPath = "presenceContainerPath"; |
| x3::rule<listInstancePath_class, dataPath_> const listInstancePath = "listInstancePath"; |
| x3::rule<leafListElementPath_class, dataPath_> const leafListElementPath = "leafListElementPath"; |
| x3::rule<initializePath_class, x3::unused_type> const initializePath = "initializePath"; |
| x3::rule<trailingSlash_class, TrailingSlash> const trailingSlash = "trailingSlash"; |
| x3::rule<absoluteStart_class, Scope> const absoluteStart = "absoluteStart"; |
| x3::rule<keyValue_class, keyValue_> const keyValue = "keyValue"; |
| x3::rule<key_identifier_class, std::string> const key_identifier = "key_identifier"; |
| x3::rule<listSuffix_class, std::vector<keyValue_>> const listSuffix = "listSuffix"; |
| x3::rule<createKeySuggestions_class, x3::unused_type> const createKeySuggestions = "createKeySuggestions"; |
| x3::rule<createValueSuggestions_class, x3::unused_type> const createValueSuggestions = "createValueSuggestions"; |
| x3::rule<suggestKeysEnd_class, x3::unused_type> const suggestKeysEnd = "suggestKeysEnd"; |
| x3::rule<class leafListValue_class, leaf_data_> const leafListValue = "leafListValue"; |
| |
| enum class NodeParserMode { |
| CompleteDataNode, |
| IncompleteDataNode, |
| CompletionsOnly, |
| SchemaNode |
| }; |
| |
| template <auto> |
| struct ModeToAttribute; |
| template <> |
| struct ModeToAttribute<NodeParserMode::CompleteDataNode> { |
| using type = dataNode_; |
| }; |
| template <> |
| struct ModeToAttribute<NodeParserMode::IncompleteDataNode> { |
| using type = dataNode_; |
| }; |
| template <> |
| struct ModeToAttribute<NodeParserMode::SchemaNode> { |
| using type = schemaNode_; |
| }; |
| // The CompletionsOnly attribute is dataNode_ only because of convenience: |
| // having the same return type means we can get by without a ton of `if constexpr` stanzas. |
| // So the code will still "parse data into the target attr" for simplicity. |
| template <> |
| struct ModeToAttribute<NodeParserMode::CompletionsOnly> { |
| using type = dataNode_; |
| }; |
| |
| enum class CompletionMode { |
| Schema, |
| Data |
| }; |
| |
| template <NodeParserMode PARSER_MODE, CompletionMode COMPLETION_MODE> |
| struct NodeParser : x3::parser<NodeParser<PARSER_MODE, COMPLETION_MODE>> { |
| using attribute_type = typename ModeToAttribute<PARSER_MODE>::type; |
| |
| std::function<bool(const Schema&, const std::string& path)> m_filterFunction; |
| |
| NodeParser(const std::function<bool(const Schema&, const std::string& path)>& filterFunction) |
| : m_filterFunction(filterFunction) |
| { |
| } |
| |
| // GCC complains that `end` isn't used when doing completions only |
| // FIXME: GCC 10.1 doesn't emit a warning here. Remove [[maybe_unused]] when GCC 10 is available |
| template <typename It, typename Ctx, typename RCtx, typename Attr> |
| bool parse(It& begin, [[maybe_unused]] It end, Ctx const& ctx, RCtx& rctx, Attr& attr) const |
| { |
| std::string tableName; |
| if constexpr (std::is_same<attribute_type, schemaNode_>()) { |
| tableName = "schemaNode"; |
| } else { |
| tableName = "dataNode"; |
| } |
| x3::symbols<attribute_type> table(tableName); |
| |
| ParserContext& parserContext = x3::get<parser_context_tag>(ctx); |
| parserContext.m_suggestions.clear(); |
| for (const auto& child : parserContext.m_schema.availableNodes(parserContext.currentSchemaPath(), Recursion::NonRecursive)) { |
| attribute_type out; |
| std::string parseString; |
| if (child.first) { |
| out.m_prefix = module_{*child.first}; |
| parseString = *child.first + ":"; |
| } |
| parseString += child.second; |
| |
| if (!m_filterFunction(parserContext.m_schema, joinPaths(pathToSchemaString(parserContext.currentSchemaPath(), Prefixes::Always), parseString))) { |
| continue; |
| } |
| |
| switch (parserContext.m_schema.nodeType(parserContext.currentSchemaPath(), child)) { |
| case yang::NodeTypes::Container: |
| case yang::NodeTypes::PresenceContainer: |
| out.m_suffix = container_{child.second}; |
| parserContext.m_suggestions.emplace(Completion{parseString + "/"}); |
| break; |
| case yang::NodeTypes::Leaf: |
| out.m_suffix = leaf_{child.second}; |
| parserContext.m_suggestions.emplace(Completion{parseString + " "}); |
| break; |
| case yang::NodeTypes::List: |
| if constexpr (std::is_same<attribute_type, schemaNode_>()) { |
| out.m_suffix = list_{child.second}; |
| } else { |
| out.m_suffix = listElement_{child.second, {}}; |
| } |
| |
| if constexpr (COMPLETION_MODE == CompletionMode::Schema) { |
| parserContext.m_suggestions.emplace(Completion{parseString + "/"}); |
| } else { |
| parserContext.m_suggestions.emplace(Completion{parseString, "[", Completion::WhenToAdd::IfFullMatch}); |
| } |
| break; |
| case yang::NodeTypes::LeafList: |
| if constexpr (std::is_same<attribute_type, schemaNode_>()) { |
| out.m_suffix = leafList_{child.second}; |
| } else { |
| out.m_suffix = leafListElement_{child.second, {}}; |
| } |
| |
| if constexpr (COMPLETION_MODE == CompletionMode::Schema) { |
| parserContext.m_suggestions.emplace(Completion{parseString + "/"}); |
| } else { |
| parserContext.m_suggestions.emplace(Completion{parseString, "[", Completion::WhenToAdd::IfFullMatch}); |
| } |
| break; |
| case yang::NodeTypes::Rpc: |
| out.m_suffix = rpcNode_{child.second}; |
| parserContext.m_suggestions.emplace(Completion{parseString + "/"}); |
| break; |
| case yang::NodeTypes::Action: |
| out.m_suffix = actionNode_{child.second}; |
| parserContext.m_suggestions.emplace(Completion{parseString + "/"}); |
| break; |
| case yang::NodeTypes::AnyXml: |
| case yang::NodeTypes::Notification: |
| continue; |
| } |
| table.add(parseString, out); |
| if (!child.first) { |
| auto topLevelModule = parserContext.currentSchemaPath().m_nodes.begin()->m_prefix; |
| out.m_prefix = topLevelModule; |
| table.add(topLevelModule->m_name + ":" + parseString, out); |
| } |
| } |
| table.add("..", attribute_type{nodeup_{}}); |
| parserContext.m_completionIterator = begin; |
| |
| if constexpr (PARSER_MODE == NodeParserMode::CompletionsOnly) { |
| return true; |
| } else { |
| It saveIter; |
| // GCC complains that I assign saveIter because I use it only if NodeType is dataNode_ |
| // FIXME: GCC 10.1 doesn't emit a warning here. Make this unconditional when GCC 10 is available. |
| if constexpr (std::is_same<attribute_type, dataNode_>()) { |
| saveIter = begin; |
| } |
| |
| auto res = table.parse(begin, end, ctx, rctx, attr); |
| |
| if (std::holds_alternative<leaf_>(attr.m_suffix)) { |
| parserContext.m_tmpListKeyLeafPath.m_location = parserContext.currentSchemaPath(); |
| ModuleNodePair node{attr.m_prefix.flat_map([](const auto& it) { |
| return boost::optional<std::string>{it.m_name}; |
| }), |
| std::get<leaf_>(attr.m_suffix).m_name}; |
| parserContext.m_tmpListKeyLeafPath.m_node = node; |
| } |
| |
| if constexpr (std::is_same<attribute_type, dataNode_>()) { |
| if (std::holds_alternative<listElement_>(attr.m_suffix)) { |
| parserContext.m_tmpListPath = parserContext.currentDataPath(); |
| auto tmpList = list_{std::get<listElement_>(attr.m_suffix).m_name}; |
| parserContext.m_tmpListPath.m_nodes.emplace_back(attr.m_prefix, tmpList); |
| |
| res = listSuffix.parse(begin, end, ctx, rctx, std::get<listElement_>(attr.m_suffix).m_keys); |
| |
| // FIXME: think of a better way to do this, that is, get rid of manual iterator reverting |
| if (!res) { |
| // If listSuffix didn't succeed, we check, if we allow incomplete nodes. If we do, then we replace listElement_ with list_. |
| // If we don't, we fail the whole symbol table. |
| if constexpr (PARSER_MODE == NodeParserMode::IncompleteDataNode) { |
| res = true; |
| attr.m_suffix = list_{std::get<listElement_>(attr.m_suffix).m_name}; |
| } else { |
| begin = saveIter; |
| } |
| } |
| } |
| |
| if (std::holds_alternative<leafListElement_>(attr.m_suffix)) { |
| parserContext.m_tmpListKeyLeafPath.m_location = parserContext.currentSchemaPath(); |
| ModuleNodePair node{attr.m_prefix.flat_map([](const auto& it) { |
| return boost::optional<std::string>{it.m_name}; |
| }), |
| std::get<leafListElement_>(attr.m_suffix).m_name}; |
| parserContext.m_tmpListKeyLeafPath.m_node = node; |
| res = leafListValue.parse(begin, end, ctx, rctx, std::get<leafListElement_>(attr.m_suffix).m_value); |
| |
| if (!res) { |
| if constexpr (PARSER_MODE == NodeParserMode::IncompleteDataNode) { |
| res = true; |
| attr.m_suffix = leafList_{std::get<leafListElement_>(attr.m_suffix).m_name}; |
| } else { |
| begin = saveIter; |
| } |
| } |
| } |
| } |
| |
| if (res) { |
| parserContext.pushPathFragment(attr); |
| } |
| |
| return res; |
| } |
| } |
| }; |
| |
| template <CompletionMode COMPLETION_MODE> using schemaNode = NodeParser<NodeParserMode::SchemaNode, COMPLETION_MODE>; |
| template <CompletionMode COMPLETION_MODE> using dataNode = NodeParser<NodeParserMode::CompleteDataNode, COMPLETION_MODE>; |
| template <CompletionMode COMPLETION_MODE> using incompleteDataNode = NodeParser<NodeParserMode::IncompleteDataNode, COMPLETION_MODE>; |
| template <CompletionMode COMPLETION_MODE> using pathCompletions = NodeParser<NodeParserMode::CompletionsOnly, COMPLETION_MODE>; |
| |
| using AnyPath = boost::variant<schemaPath_, dataPath_>; |
| |
| enum class PathParserMode { |
| AnyPath, |
| DataPath, |
| DataPathListEnd |
| }; |
| |
| template <> |
| struct ModeToAttribute<PathParserMode::AnyPath> { |
| using type = AnyPath; |
| }; |
| |
| template <> |
| struct ModeToAttribute<PathParserMode::DataPath> { |
| using type = dataPath_; |
| }; |
| |
| template <> |
| struct ModeToAttribute<PathParserMode::DataPathListEnd> { |
| using type = dataPath_; |
| }; |
| |
| template <PathParserMode PARSER_MODE, CompletionMode COMPLETION_MODE> |
| struct PathParser : x3::parser<PathParser<PARSER_MODE, COMPLETION_MODE>> { |
| using attribute_type = ModeToAttribute<PARSER_MODE>; |
| std::function<bool(const Schema&, const std::string& path)> m_filterFunction; |
| |
| PathParser(const std::function<bool(const Schema&, const std::string& path)>& filterFunction = [] (const auto&, const auto&) {return true;}) |
| : m_filterFunction(filterFunction) |
| { |
| } |
| |
| template <typename It, typename Ctx, typename RCtx, typename Attr> |
| bool parse(It& begin, It end, Ctx const& ctx, RCtx& rctx, Attr& attr) const |
| { |
| initializePath.parse(begin, end, ctx, rctx, x3::unused); |
| dataPath_ attrData; |
| |
| auto pathEnd = x3::rule<class PathEnd>{"pathEnd"} = &space_separator | x3::eoi; |
| // absoluteStart has to be separate from the dataPath parser, |
| // otherwise, if the "dataNode % '/'" parser fails, the begin iterator |
| // gets reverted to before the starting slash. |
| auto res = (-absoluteStart).parse(begin, end, ctx, rctx, attrData.m_scope); |
| auto dataPath = x3::attr(attrData.m_scope) |
| >> (dataNode<COMPLETION_MODE>{m_filterFunction} % '/' | pathEnd >> x3::attr(std::vector<dataNode_>{})) |
| >> -trailingSlash; |
| res = dataPath.parse(begin, end, ctx, rctx, attrData); |
| |
| // If we allow data paths with a list at the end, we just try to parse that separately. |
| if constexpr (PARSER_MODE == PathParserMode::DataPathListEnd || PARSER_MODE == PathParserMode::AnyPath) { |
| if (!res || !pathEnd.parse(begin, end, ctx, rctx, x3::unused)) { |
| dataNode_ attrNodeList; |
| res = incompleteDataNode<COMPLETION_MODE>{m_filterFunction}.parse(begin, end, ctx, rctx, attrNodeList); |
| if (res) { |
| attrData.m_nodes.emplace_back(attrNodeList); |
| // If the trailing slash matches, no more nodes are parsed. |
| // That means no more completion. So, I generate them |
| // manually. |
| res = (-(trailingSlash >> x3::omit[pathCompletions<COMPLETION_MODE>{m_filterFunction}])).parse(begin, end, ctx, rctx, attrData.m_trailingSlash); |
| } |
| } |
| } |
| |
| attr = attrData; |
| if constexpr (PARSER_MODE == PathParserMode::AnyPath) { |
| // If our data path already has some listElement_ fragments, we can't parse rest of the path as a schema path |
| auto hasLists = std::any_of(attrData.m_nodes.begin(), attrData.m_nodes.end(), |
| [] (const auto& node) { return std::holds_alternative<listElement_>(node.m_suffix); }); |
| // If parsing failed, or if there's more input we try parsing schema nodes. |
| if (!hasLists) { |
| if (!res || !pathEnd.parse(begin, end, ctx, rctx, x3::unused)) { |
| // If dataPath parsed some nodes, they will be saved in `attrData`. We have to keep these. |
| schemaPath_ attrSchema = dataPathToSchemaPath(attrData); |
| auto schemaPath = schemaNode<COMPLETION_MODE>{m_filterFunction} % '/'; |
| // The schemaPath parser continues where the dataPath parser ended. |
| res = schemaPath.parse(begin, end, ctx, rctx, attrSchema.m_nodes); |
| auto trailing = -trailingSlash >> pathEnd; |
| res = trailing.parse(begin, end, ctx, rctx, attrSchema.m_trailingSlash); |
| attr = attrSchema; |
| } |
| } |
| } |
| return res; |
| } |
| }; |
| |
| // Need to use these wrappers so that my PathParser class gets the proper |
| // attribute. Otherwise, Spirit injects the attribute of the outer parser that |
| // uses my PathParser. |
| // Example grammar: anyPath | module. |
| // The PathParser class would get a boost::variant as the attribute, but I |
| // don't want to deal with that, so I use these wrappers to ensure the |
| // attribute I want (and let Spirit deal with boost::variant). |
| auto const anyPath = x3::rule<class anyPath_class, AnyPath>{"anyPath"} = PathParser<PathParserMode::AnyPath, CompletionMode::Schema>{}; |
| auto const dataPath = x3::rule<class dataPath_class, dataPath_>{"dataPath"} = PathParser<PathParserMode::DataPath, CompletionMode::Data>{}; |
| auto const dataPathListEnd = x3::rule<class dataPath_class, dataPath_>{"dataPath"} = PathParser<PathParserMode::DataPathListEnd, CompletionMode::Data>{}; |
| |
| #if __clang__ |
| #pragma GCC diagnostic push |
| #pragma GCC diagnostic ignored "-Woverloaded-shift-op-parentheses" |
| #endif |
| |
| struct SuggestLeafListEnd : x3::parser<SuggestLeafListEnd> { |
| using attribute_type = x3::unused_type; |
| template <typename It, typename Ctx, typename RCtx, typename Attr> |
| bool parse(It& begin, It, Ctx const& ctx, RCtx&, Attr&) const |
| { |
| auto& parserContext = x3::get<parser_context_tag>(ctx); |
| parserContext.m_completionIterator = begin; |
| parserContext.m_suggestions = {Completion{"]"}}; |
| |
| return true; |
| } |
| } const suggestLeafListEnd; |
| |
| auto const leafListValue_def = |
| '[' >> leaf_data >> suggestLeafListEnd >> ']'; |
| |
| auto const rest = |
| x3::omit[x3::no_skip[+(x3::char_ - '/' - space_separator)]]; |
| |
| auto const key_identifier_def = |
| x3::lexeme[ |
| ((x3::alpha | char_("_")) >> *(x3::alnum | char_("_") | char_("-") | char_("."))) |
| ]; |
| |
| auto const createKeySuggestions_def = |
| x3::eps; |
| |
| auto const createValueSuggestions_def = |
| x3::eps; |
| |
| auto const suggestKeysEnd_def = |
| x3::eps; |
| |
| auto const keyValue_def = |
| key_identifier > '=' > createValueSuggestions > leaf_data; |
| |
| auto const keyValueWrapper = |
| x3::no_skip['[' > createKeySuggestions > keyValue > suggestKeysEnd > ']']; |
| |
| // even though we don't allow no keys to be supplied, the star allows me to check which keys are missing |
| auto const listSuffix_def = |
| *keyValueWrapper; |
| |
| auto const list_def = |
| node_identifier >> !char_('['); |
| |
| auto const absoluteStart_def = |
| x3::omit['/'] >> x3::attr(Scope::Absolute); |
| |
| auto const trailingSlash_def = |
| x3::omit['/'] >> x3::attr(TrailingSlash::Present); |
| |
| auto const filterConfigFalse = [] (const Schema& schema, const std::string& path) { |
| return schema.isConfig(path); |
| }; |
| |
| // A WritableOps value is injected through the `x3::with` with this tag (see usage of the tag). It controls whether |
| // `config: false` data can be set with the `set` command. This is used by yang-cli because that tool needs modeling of |
| // the full datastore, including the "read-only" data. |
| struct writableOps_tag; |
| |
| PathParser<PathParserMode::DataPath, CompletionMode::Data> const dataPathFilterConfigFalse{filterConfigFalse}; |
| |
| struct WritableLeafPath : x3::parser<WritableLeafPath> { |
| using attribute_type = dataPath_; |
| template <typename It, typename Ctx, typename RCtx, typename Attr> |
| static bool parse(It& begin, It end, Ctx const& ctx, RCtx& rctx, Attr& attr) |
| { |
| bool res; |
| if (x3::get<writableOps_tag>(ctx) == WritableOps::Yes) { |
| res = dataPath.parse(begin, end, ctx, rctx, attr); |
| } else { |
| res = dataPathFilterConfigFalse.parse(begin, end, ctx, rctx, attr); |
| } |
| if (!res) { |
| return false; |
| } |
| |
| if (attr.m_nodes.empty() || !std::holds_alternative<leaf_>(attr.m_nodes.back().m_suffix)) { |
| auto& parserContext = x3::get<parser_context_tag>(ctx); |
| parserContext.m_errorMsg = "This is not a path to leaf."; |
| return false; |
| } |
| |
| return true; |
| } |
| |
| } writableLeafPath; |
| |
| struct RpcActionPath : x3::parser<RpcActionPath> { |
| using attribute_type = dataPath_; |
| template <typename It, typename Ctx, typename RCtx, typename Attr> |
| static bool parse(It& begin, It end, Ctx const& ctx, RCtx& rctx, Attr& attr) |
| { |
| bool res = dataPath.parse(begin, end, ctx, rctx, attr); |
| if (!res) { |
| return false; |
| } |
| |
| if (attr.m_nodes.empty() |
| || (!std::holds_alternative<rpcNode_>(attr.m_nodes.back().m_suffix) && !std::holds_alternative<actionNode_>(attr.m_nodes.back().m_suffix))) { |
| auto& parserContext = x3::get<parser_context_tag>(ctx); |
| parserContext.m_errorMsg = "This is not a path to an RPC/action."; |
| return false; |
| } |
| |
| return true; |
| } |
| }; |
| |
| auto const rpcActionPath = as<dataPath_>[RpcActionPath()]; |
| |
| auto const noRpcOrAction = [] (const Schema& schema, const std::string& path) { |
| auto nodeType = schema.nodeType(path); |
| return nodeType != yang::NodeTypes::Rpc && nodeType != yang::NodeTypes::Action; |
| }; |
| |
| auto const getPath_def = |
| PathParser<PathParserMode::DataPathListEnd, CompletionMode::Data>{noRpcOrAction} | |
| PathParser<PathParserMode::DataPath, CompletionMode::Data>{noRpcOrAction} | |
| (module >> "*"); |
| |
| auto const cdPath_def = |
| PathParser<PathParserMode::DataPath, CompletionMode::Data>{noRpcOrAction}; |
| |
| auto const presenceContainerPath_def = |
| dataPath; |
| |
| auto const listInstancePath_def = |
| dataPath; |
| |
| auto const leafListElementPath_def = |
| dataPath; |
| |
| // A "nothing" parser, which is used to indicate we tried to parse a path |
| auto const initializePath_def = |
| x3::eps; |
| |
| |
| |
| #if __clang__ |
| #pragma GCC diagnostic pop |
| #endif |
| |
| BOOST_SPIRIT_DEFINE(keyValue) |
| BOOST_SPIRIT_DEFINE(key_identifier) |
| BOOST_SPIRIT_DEFINE(listSuffix) |
| BOOST_SPIRIT_DEFINE(cdPath) |
| BOOST_SPIRIT_DEFINE(getPath) |
| BOOST_SPIRIT_DEFINE(presenceContainerPath) |
| BOOST_SPIRIT_DEFINE(listInstancePath) |
| BOOST_SPIRIT_DEFINE(leafListElementPath) |
| BOOST_SPIRIT_DEFINE(initializePath) |
| BOOST_SPIRIT_DEFINE(createKeySuggestions) |
| BOOST_SPIRIT_DEFINE(createValueSuggestions) |
| BOOST_SPIRIT_DEFINE(suggestKeysEnd) |
| BOOST_SPIRIT_DEFINE(leafListValue) |
| BOOST_SPIRIT_DEFINE(absoluteStart) |
| BOOST_SPIRIT_DEFINE(trailingSlash) |