From 54870fb76b5f59cc2e6d69a8c7a8ef17853a0029 Mon Sep 17 00:00:00 2001 From: Boris Kolpackov Date: Mon, 5 Dec 2016 15:09:04 +0200 Subject: Add comma, ternary, logical operators support in eval context --- build2/lexer.cxx | 51 +++-- build2/parser | 25 ++- build2/parser.cxx | 489 ++++++++++++++++++++++++++++++------------- build2/test/script/token | 2 - build2/test/script/token.cxx | 3 - build2/token | 46 ++-- build2/token.cxx | 17 +- build2/variable | 11 + build2/variable.txx | 14 ++ 9 files changed, 463 insertions(+), 195 deletions(-) (limited to 'build2') diff --git a/build2/lexer.cxx b/build2/lexer.cxx index 8918740..81cfea5 100644 --- a/build2/lexer.cxx +++ b/build2/lexer.cxx @@ -65,8 +65,8 @@ namespace build2 } case lexer_mode::eval: { - s1 = ":<>=! $(){}[]#\t\n"; - s2 = " == "; + s1 = ":<>=!&|?, $(){}[]#\t\n"; + s2 = " = &| "; break; } case lexer_mode::single_quoted: @@ -233,32 +233,51 @@ namespace build2 case '[': return make_token (type::lsbrace); case ']': return make_token (type::rsbrace); case '$': return make_token (type::dollar); + case '?': return make_token (type::question); + case ',': return make_token (type::comma); case '(': return make_token (type::lparen); case ')': { state_.pop (); // Expire eval mode. return make_token (type::rparen); } + // Potentially two-character tokens. + // case '=': case '!': - { - if (peek () == '=') - { - get (); - return make_token (c == '=' ? type::equal : type::not_equal); - } - break; - } case '<': case '>': + case '|': + case '&': { - bool e (peek () == '='); - if (e) - get (); + xchar p (peek ()); + + type r (type::eos); + switch (c) + { + case '|': if (p == '|') r = type::log_or; break; + case '&': if (p == '&') r = type::log_and; break; + + case '<': r = (p == '=' ? type::less_equal : type::less); break; + case '>': r = (p == '=' ? type::greater_equal : type::greater); break; + + case '=': if (p == '=') r = type::equal; break; + + case '!': r = (p == '=' ? type::not_equal : type::log_not); break; + } + + if (r == type::eos) + break; + + switch (r) + { + case type::less: + case type::greater: + case type::log_not: break; + default: get (); + } - return make_token (c == '<' - ? e ? type::less_equal : type::less - : e ? type::greater_equal : type::greater); + return make_token (r); } } diff --git a/build2/parser b/build2/parser index e1f6e77..45f5b16 100644 --- a/build2/parser +++ b/build2/parser @@ -110,15 +110,30 @@ namespace build2 value&& rhs, token_type kind); - // Return the value (can be NULL/typed) as well as the indication of - // whether this is a non-empty eval context (i.e., '()' potentially with - // whitespace in between). + // Return the value pack (values can be NULL/typed). Note that for an + // empty eval context ('()' potentially with whitespaces in between) the + // result is an empty pack, not a pack of one empty. // - pair + values parse_eval (token&, token_type&); + values + parse_eval_comma (token&, token_type&, bool = false); + + value + parse_eval_ternary (token&, token_type&, bool = false); + + value + parse_eval_or (token&, token_type&, bool = false); + + value + parse_eval_and (token&, token_type&, bool = false); + + value + parse_eval_comp (token&, token_type&, bool = false); + value - parse_eval_trailer (token&, token_type&); + parse_eval_value (token&, token_type&, bool = false); // Attributes stack. We can have nested attributes, for example: // diff --git a/build2/parser.cxx b/build2/parser.cxx index 5d4ff0a..4126aff 100644 --- a/build2/parser.cxx +++ b/build2/parser.cxx @@ -1738,181 +1738,367 @@ namespace build2 } } - pair parser:: + values parser:: parse_eval (token& t, type& tt) { - mode (lexer_mode::eval, '@'); + // enter: lparen + // leave: rparen + + mode (lexer_mode::eval, '@'); // Auto-expires at rparen. next (t, tt); - return tt != type::rparen - ? make_pair (parse_eval_trailer (t, tt), true) - : make_pair (value (names ()), false); + + if (tt == type::rparen) + return values (); + + values r (parse_eval_comma (t, tt, true)); + + if (tt != type::rparen) + fail (t) << "unexpected " << t; // E.g., stray ':'. + + return r; + } + + values parser:: + parse_eval_comma (token& t, type& tt, bool first) + { + // enter: first token of LHS + // leave: next token after last RHS + + // Left-associative: parse in a loop for as long as we can. + // + values r; + value lhs (parse_eval_ternary (t, tt, first)); + + if (!pre_parse_) + r.push_back (move (lhs)); + + while (tt == type::comma) + { + next (t, tt); + value rhs (parse_eval_ternary (t, tt)); + + if (!pre_parse_) + r.push_back (move (rhs)); + } + + return r; } value parser:: - parse_eval_trailer (token& t, type& tt) + parse_eval_ternary (token& t, type& tt, bool first) { - // Parse the next value (if any) handling its attributes. + // enter: first token of LHS + // leave: next token after last RHS + + // Right-associative (kind of): we parse what's between ?: without + // regard for priority and we recurse on what's after :. Here is an + // example: + // + // a ? x ? y : z : b ? c : d + // + // This should be parsed/evaluated as: + // + // a ? (x ? y : z) : (b ? c : d) + // + location l (get_location (t)); + value lhs (parse_eval_or (t, tt, first)); + + if (tt != type::question) + return lhs; + + // Use the pre-parse mechanism to implement short-circuit. // - auto parse = [&t, &tt, this] () -> value + bool pp (pre_parse_); + + bool q; + try { - // Parse value attributes if any. Note that it's ok not to have anything - // after the attributes, as in, ($foo == [null]), or even ([null]) - // - auto at (attributes_push (t, tt, true)); + q = pp ? true : convert (move (lhs)); + } + catch (const invalid_argument& e) { fail (l) << e.what (); } - // Note that if parse_value() gets called, it expects to see a value. - // - value r (tt != type::rparen && - tt != type::colon && - tt != type::equal && - tt != type::not_equal && - tt != type::less && - tt != type::less_equal && - tt != type::greater && - tt != type::greater_equal - ? parse_value (t, tt) - : value (names ())); + if (!pp) + pre_parse_ = !q; // Short-circuit middle? - if (pre_parse_) - return r; // Empty. + next (t, tt); + value mhs (parse_eval_ternary (t, tt)); - // Process attributes if any. - // - if (!at.first) + if (tt != type::colon) + fail (t) << "expected ':' instead of " << t; + + if (!pp) + pre_parse_ = q; // Short-circuit right? + + next (t, tt); + value rhs (parse_eval_ternary (t, tt)); + + pre_parse_ = pp; + return q ? move (mhs) : move (rhs); + } + + value parser:: + parse_eval_or (token& t, type& tt, bool first) + { + // enter: first token of LHS + // leave: next token after last RHS + + // Left-associative: parse in a loop for as long as we can. + // + location l (get_location (t)); + value lhs (parse_eval_and (t, tt, first)); + + // Use the pre-parse mechanism to implement short-circuit. + // + bool pp (pre_parse_); + + while (tt == type::log_or) + { + try { - attributes_pop (); - return r; + if (!pre_parse_ && convert (move (lhs))) + pre_parse_ = true; + + next (t, tt); + l = get_location (t); + value rhs (parse_eval_and (t, tt)); + + if (pre_parse_) + continue; + + // Store the result as bool value. + // + lhs = convert (move (rhs)); } + catch (const invalid_argument& e) { fail (l) << e.what (); } + } - value v; - apply_value_attributes (nullptr, v, move (r), type::assign); - return v; - }; + pre_parse_ = pp; + return lhs; + } - value lhs (parse ()); + value parser:: + parse_eval_and (token& t, type& tt, bool first) + { + // enter: first token of LHS + // leave: next token after last RHS - // We continue evaluating from left to right until we reach ')', storing - // the result in lhs (think 'a == b == false'). + // Left-associative: parse in a loop for as long as we can. // - while (tt != type::rparen) - { - const location l (get_location (t)); + location l (get_location (t)); + value lhs (parse_eval_comp (t, tt, first)); - // Remember to update parse_value above if adding any new name - // separators here. - // - switch (tt) + // Use the pre-parse mechanism to implement short-circuit. + // + bool pp (pre_parse_); + + while (tt == type::log_and) + { + try { - case type::colon: - { - // Later, when we support '?:', we will need to decide which case - // this is. But for now ':' is always a scope/target qualified name - // which we represent as a special ':'-style pair. - // - next (t, tt); - value rhs (parse_eval_trailer (t, tt)); + if (!pre_parse_ && !convert (move (lhs))) + pre_parse_ = true; - if (tt != type::rparen) - fail (l) << "variable name expected after ':'"; + next (t, tt); + l = get_location (t); + value rhs (parse_eval_comp (t, tt)); - if (pre_parse_) - break; + if (pre_parse_) + continue; + + // Store the result as bool value. + // + lhs = convert (move (rhs)); + } + catch (const invalid_argument& e) { fail (l) << e.what (); } + } - if (lhs.type != nullptr || !lhs || lhs.empty ()) - fail (l) << "scope/target expected before ':'"; + pre_parse_ = pp; + return lhs; + } - if (rhs.type != nullptr || !rhs || rhs.empty ()) - fail (l) << "variable name expected after ':'"; + value parser:: + parse_eval_comp (token& t, type& tt, bool first) + { + // enter: first token of LHS + // leave: next token after last RHS - names& ns (lhs.as ()); - ns.back ().pair = ':'; + // Left-associative: parse in a loop for as long as we can. + // + value lhs (parse_eval_value (t, tt, first)); + + while (tt == type::equal || + tt == type::not_equal || + tt == type::less || + tt == type::less_equal || + tt == type::greater || + tt == type::greater_equal) + { + type op (tt); + location l (get_location (t)); - ns.insert (ns.end (), - make_move_iterator (rhs.as ().begin ()), - make_move_iterator (rhs.as ().end ())); + next (t, tt); + value rhs (parse_eval_value (t, tt)); - break; + if (pre_parse_) + continue; + + // Use (potentially typed) comparison via value. If one of the values is + // typed while the other is not, then try to convert the untyped one to + // the other's type instead of complaining. This seems like a reasonable + // thing to do and will allow us to write: + // + // if ($build.version > 30000) + // + // Rather than having to write: + // + // if ($build.version > [uint64] 30000) + // + if (lhs.type != rhs.type) + { + // @@ Would be nice to pass location for diagnostics. + // + if (lhs.type == nullptr) + { + if (lhs) + typify (lhs, *rhs.type, nullptr); } - case type::equal: - case type::not_equal: - case type::less: - case type::less_equal: - case type::greater: - case type::greater_equal: + else if (rhs.type == nullptr) { - type op (tt); + if (rhs) + typify (rhs, *lhs.type, nullptr); + } + else + fail (l) << "comparison between " << lhs.type->name << " and " + << rhs.type->name; + } - // These are left-associative, so get the rhs value and evaluate. - // - // @@ In C++ == and != have lower precedence than <, etc. - // - next (t, tt); - value rhs (parse ()); + bool r; + switch (op) + { + case type::equal: r = lhs == rhs; break; + case type::not_equal: r = lhs != rhs; break; + case type::less: r = lhs < rhs; break; + case type::less_equal: r = lhs <= rhs; break; + case type::greater: r = lhs > rhs; break; + case type::greater_equal: r = lhs >= rhs; break; + default: r = false; assert (false); + } - if (pre_parse_) - break; + // Store the result as a bool value. + // + lhs = value (r); + } - // Use (potentially typed) comparison via value. If one of the - // values is typed while the other is not, then try to convert the - // untyped one to the other's type instead of complaining. This - // seems like a reasonable thing to do and will allow us to write: - // - // if ($build.version > 30000) - // - // Rather than having to write: - // - // if ($build.version > [uint64] 30000) - // - if (lhs.type != rhs.type) - { - // @@ Would be nice to pass location for diagnostics. - // - if (lhs.type == nullptr) - { - if (lhs) - typify (lhs, *rhs.type, nullptr); - } - else if (rhs.type == nullptr) - { - if (rhs) - typify (rhs, *lhs.type, nullptr); - } - else - fail (l) << "comparison between " << lhs.type->name << " and " - << rhs.type->name; - } + return lhs; + } - bool r; - switch (op) - { - case type::equal: r = lhs == rhs; break; - case type::not_equal: r = lhs != rhs; break; - case type::less: r = lhs < rhs; break; - case type::less_equal: r = lhs <= rhs; break; - case type::greater: r = lhs > rhs; break; - case type::greater_equal: r = lhs >= rhs; break; - default: r = false; assert (false); - } + value parser:: + parse_eval_value (token& t, type& tt, bool first) + { + // enter: first token of value + // leave: next token after value - // Store the result as a bool value. - // - if (lhs.type != nullptr) - { - // Reset to NULL and untype. - // - lhs = nullptr; - lhs.type = nullptr; - } + // Parse value attributes if any. Note that it's ok not to have anything + // after the attributes, as in, ($foo == [null]), or even ([null]) + // + auto at (attributes_push (t, tt, true)); + + const location l (get_location (t)); + + value v; + switch (tt) + { + case type::log_not: + { + next (t, tt); + v = parse_eval_value (t, tt); - lhs = r; + if (pre_parse_) break; + + try + { + // Store the result as bool value. + // + v = !convert (move (v)); } - default: - fail (l) << "expected ')' instead of " << t; + catch (const invalid_argument& e) { fail (l) << e.what (); } + break; + } + default: + { + // If parse_value() gets called, it expects to see a value. Note that + // it will also handle nested eval contexts. + // + v = (tt != type::colon && + tt != type::question && + tt != type::comma && + + tt != type::rparen && + + tt != type::equal && + tt != type::not_equal && + tt != type::less && + tt != type::less_equal && + tt != type::greater && + tt != type::greater_equal && + + tt != type::log_or && + tt != type::log_and + + ? parse_value (t, tt) + : value (names ())); } } - return lhs; + // If this is the first expression then handle the eval-qual special case + // (scope/target qualified name represented as a special ':'-style pair). + // + if (first && tt == type::colon) + { + if (at.first) + fail (at.second) << "attributes before qualified variable name"; + + attributes_pop (); + + const location nl (get_location (t)); + next (t, tt); + value n (parse_value (t, tt)); + + if (tt != type::rparen) + fail (t) << "expected ')' after variable name"; + + if (pre_parse_) + return v; // Empty. + + if (v.type != nullptr || !v || v.as ().size () != 1) + fail (l) << "expected scope/target before ':'"; + + if (n.type != nullptr || !n || n.as ().size () != 1) + fail (nl) << "expected variable name after ':'"; + + names& ns (v.as ()); + ns.back ().pair = ':'; + ns.push_back (move (n.as ().back ())); + return v; + } + else + { + if (pre_parse_) + return v; // Empty. + + // Process attributes if any. + // + if (!at.first) + { + attributes_pop (); + return v; + } + + value r; + apply_value_attributes (nullptr, r, move (v), type::assign); + return r; + } } pair parser:: @@ -2536,10 +2722,15 @@ namespace build2 else if (tt == type::lparen) { expire_mode (); - value v (parse_eval (t, tt).first); //@@ OUT will parse @-pair and do well? + values vs (parse_eval (t, tt)); //@@ OUT will parse @-pair and do well? if (!pre_parse_) { + if (vs.size () != 1) + fail (loc) << "expected single variable/function name"; + + value& v (vs[0]); + if (!v) fail (loc) << "null variable/function name"; @@ -2553,7 +2744,7 @@ namespace build2 if (n > 2 || (n == 2 && ns[0].pair != ':') || !ns[n - 1].simple ()) - fail (loc) << "variable/function name expected instead of '" + fail (loc) << "expected variable/function name instead of '" << ns << "'"; if (n == 2) @@ -2570,7 +2761,7 @@ namespace build2 } } else - fail (t) << "variable/function name expected instead of " << t; + fail (t) << "expected variable/function name instead of " << t; if (!pre_parse_ && name.empty ()) fail (loc) << "empty variable/function name"; @@ -2590,21 +2781,15 @@ namespace build2 // @@ Should we use (target/scope) qualification (of name) as the // context in which to call the function? // - auto args (parse_eval (t, tt)); + values args (parse_eval (t, tt)); tt = peek (); if (pre_parse_) continue; // As if empty result. - // Note that we move args to call(). + // Note that we "move" args to call(). // - result_data = functions.call ( - name, - vector_view ( - args.second ? &args.first : nullptr, - args.second ? 1 : 0), - loc); - + result_data = functions.call (name, args, loc); what = "function call"; } else @@ -2629,13 +2814,19 @@ namespace build2 // loc = get_location (t); - result_data = parse_eval (t, tt).first; - + values vs (parse_eval (t, tt)); tt = peek (); if (pre_parse_) continue; // As if empty result. + switch (vs.size ()) + { + case 0: result_data = value (names ()); break; + case 1: result_data = move (vs[0]); break; + default: fail (loc) << "expected single value"; + } + what = "context evaluation"; } diff --git a/build2/test/script/token b/build2/test/script/token index 7f79746..117ccaa 100644 --- a/build2/test/script/token +++ b/build2/test/script/token @@ -31,8 +31,6 @@ namespace build2 pipe, // | clean, // &{?!} (modifiers in value) - log_and, // && - log_or, // || in_pass, // <+ in_null, // <- diff --git a/build2/test/script/token.cxx b/build2/test/script/token.cxx index a8ef5b4..5552994 100644 --- a/build2/test/script/token.cxx +++ b/build2/test/script/token.cxx @@ -29,10 +29,7 @@ namespace build2 case token_type::minus: os << q << '-' << q; break; case token_type::clean: os << q << '&' << v << q; break; - case token_type::pipe: os << q << '|' << q; break; - case token_type::log_and: os << q << "&&" << q; break; - case token_type::log_or: os << q << "||" << q; break; case token_type::in_pass: os << q << "<+" << q; break; case token_type::in_null: os << q << "<-" << q; break; diff --git a/build2/token b/build2/token index 3d580b9..2cc1fca 100644 --- a/build2/token +++ b/build2/token @@ -28,23 +28,35 @@ namespace build2 newline, word, pair_separator, - colon, - lcbrace, // { - rcbrace, // } - lsbrace, // [ - rsbrace, // ] - assign, // = - prepend, // =+ - append, // += - equal, // == - not_equal, // != - less, // < - greater, // > - less_equal, // <= - greater_equal, // >= - dollar, // $ - lparen, // ( - rparen, // ) + + colon, // : + dollar, // $ + question, // ? + comma, // , + + lparen, // ( + rparen, // ) + + lcbrace, // { + rcbrace, // } + + lsbrace, // [ + rsbrace, // ] + + assign, // = + prepend, // =+ + append, // += + + equal, // == + not_equal, // != + less, // < + greater, // > + less_equal, // <= + greater_equal, // >= + + log_or, // || + log_and, // && + log_not, // ! value_next }; diff --git a/build2/token.cxx b/build2/token.cxx index df0d8ce..caa775b 100644 --- a/build2/token.cxx +++ b/build2/token.cxx @@ -23,22 +23,33 @@ namespace build2 case token_type::word: os << '\'' << t.value << '\''; break; case token_type::colon: os << q << ':' << q; break; + case token_type::dollar: os << q << '$' << q; break; + case token_type::question: os << q << '?' << q; break; + case token_type::comma: os << q << ',' << q; break; + + case token_type::lparen: os << q << '(' << q; break; + case token_type::rparen: os << q << ')' << q; break; + case token_type::lcbrace: os << q << '{' << q; break; case token_type::rcbrace: os << q << '}' << q; break; + case token_type::lsbrace: os << q << '[' << q; break; case token_type::rsbrace: os << q << ']' << q; break; + case token_type::assign: os << q << '=' << q; break; case token_type::prepend: os << q << "=+" << q; break; case token_type::append: os << q << "+=" << q; break; + case token_type::equal: os << q << "==" << q; break; case token_type::not_equal: os << q << "!=" << q; break; case token_type::less: os << q << '<' << q; break; case token_type::greater: os << q << '>' << q; break; case token_type::less_equal: os << q << "<=" << q; break; case token_type::greater_equal: os << q << ">=" << q; break; - case token_type::dollar: os << q << '$' << q; break; - case token_type::lparen: os << q << '(' << q; break; - case token_type::rparen: os << q << ')' << q; break; + + case token_type::log_or: os << q << "||" << q; break; + case token_type::log_and: os << q << "&&" << q; break; + case token_type::log_not: os << q << '!' << q; break; default: assert (false); // Unhandled extended token. } diff --git a/build2/variable b/build2/variable index b773047..abff1d5 100644 --- a/build2/variable +++ b/build2/variable @@ -248,6 +248,12 @@ namespace build2 reset (); }; + // This is what we call a "value pack"; it can be created by the eval + // context and passed as arguments to functions. Usually we will have just + // one value. + // + using values = small_vector; + // The values should be of the same type (or both be untyped) except NULL // values can also be untyped. NULL values compare equal and a NULL value // is always less than a non-NULL. @@ -439,6 +445,11 @@ namespace build2 // //template T convert (names&&); (declaration causes ambiguity) + // Convert value to T. If value is already of type T, then simply cast it. + // Otherwise call convert(names) above. + // + template T convert (value&&); + // Default implementations of the dtor/copy_ctor/copy_assing callbacks for // types that are stored directly in value::data_ and the provide all the // necessary functions (copy/move ctor and assignment operator). diff --git a/build2/variable.txx b/build2/variable.txx index ab31a45..d566a04 100644 --- a/build2/variable.txx +++ b/build2/variable.txx @@ -35,6 +35,20 @@ namespace build2 } template + T + convert (value&& v) + { + if (v.type == nullptr) + return convert (move (v).as ()); + else if (v.type == &value_traits::value_type) + return move (v).as (); + + throw invalid_argument ( + string ("invalid ") + value_traits::type_name + + " value: conversion from " + v.type->name); + } + + template void default_dtor (value& v) { -- cgit v1.1