diff options
Diffstat (limited to 'libbuild2/build/script/parser.cxx')
| -rw-r--r-- | libbuild2/build/script/parser.cxx | 1693 |
1 files changed, 1422 insertions, 271 deletions
diff --git a/libbuild2/build/script/parser.cxx b/libbuild2/build/script/parser.cxx index d9bfef6..3ecf23d 100644 --- a/libbuild2/build/script/parser.cxx +++ b/libbuild2/build/script/parser.cxx @@ -15,6 +15,8 @@ #include <libbuild2/algorithm.hxx> #include <libbuild2/make-parser.hxx> +#include <libbuild2/adhoc-rule-buildscript.hxx> + #include <libbuild2/script/run.hxx> #include <libbuild2/build/script/lexer.hxx> @@ -45,7 +47,7 @@ namespace build2 { path_ = &pn; - pre_parse_ = true; + top_pre_parse_ = pre_parse_ = true; lexer l (is, *path_, line, lexer_mode::command_line); set_lexer (&l); @@ -59,7 +61,7 @@ namespace build2 pbase_ = scope_->src_path_; - file_based_ = tt.is_a<file> (); + file_based_ = tt.is_a<file> () || tt.is_a<group> (); perform_update_ = find (as.begin (), as.end (), perform_update_id) != as.end (); @@ -91,15 +93,31 @@ namespace build2 info << "consider using 'depdb' builtin to track its result " << "changes"; - // Diagnose absent/ambigous script name. + // Diagnose computed variable exansions. + // + if (computed_var_) + fail (*computed_var_) + << "expansion of computed variable is only allowed in depdb " + << "preamble" << + info << "consider using 'depdb' builtin to track its value " + << "changes"; + + // Diagnose absent/ambiguous script name. But try to deduce an absent + // name from the script operation first. // { diag_record dr; - if (!diag_name_ && !diag_line_) + if (!diag_name_ && diag_preamble_.empty ()) { - dr << fail (s.start_loc) - << "unable to deduce low-verbosity script diagnostics name"; + if (as.size () == 1) + { + diag_name_ = make_pair (ctx->operation_table[as[0].operation ()], + location ()); + } + else + dr << fail (s.start_loc) + << "unable to deduce low-verbosity script diagnostics name"; } else if (diag_name2_) { @@ -125,20 +143,22 @@ namespace build2 // Save the script name or custom diagnostics line. // - assert (diag_name_.has_value () != diag_line_.has_value ()); + assert (diag_name_.has_value () == diag_preamble_.empty ()); if (diag_name_) s.diag_name = move (diag_name_->first); else - s.diag_line = move (diag_line_->first); + s.diag_preamble = move (diag_preamble_); // Save the custom dependency change tracking lines, if present. // s.depdb_clear = depdb_clear_.has_value (); + s.depdb_value = depdb_value_; if (depdb_dyndep_) { s.depdb_dyndep = depdb_dyndep_->second; s.depdb_dyndep_byproduct = depdb_dyndep_byproduct_; + s.depdb_dyndep_dyn_target = depdb_dyndep_dyn_target_; } s.depdb_preamble = move (depdb_preamble_); @@ -181,13 +201,27 @@ namespace build2 } } + // Parse a logical line, handling the flow control constructs + // recursively. + // + // If the flow control construct type is specified, then this line is + // assumed to belong to such a construct. + // void parser:: - pre_parse_line (token& t, type& tt, bool if_line) + pre_parse_line (token& t, type& tt, optional<line_type> fct) { + // enter: next token is peeked at (type in tt) + // leave: newline + + assert (!fct || + *fct == line_type::cmd_if || + *fct == line_type::cmd_while || + *fct == line_type::cmd_for_stream || + *fct == line_type::cmd_for_args); + // Determine the line type/start token. // - line_type lt ( - pre_parse_line_start (t, tt, lexer_mode::second_token)); + line_type lt (pre_parse_line_start (t, tt, lexer_mode::second_token)); line ln; @@ -220,22 +254,148 @@ namespace build2 break; } + // + // See pre_parse_line_start() for details. + // + case line_type::cmd_for_args: assert (false); break; + case line_type::cmd_for_stream: + { + // First we need to sense the next few tokens and detect which + // form of the loop we are dealing with, the first (for x: ...) + // or the third (x <...) one. Note that the second form (... | for + // x) is handled separately. + // + // If the next token doesn't look like a variable name, then this + // is the third form. Otherwise, if colon follows the variable + // name, potentially after the attributes, then this is the first + // form and the third form otherwise. + // + // Note that for the third form we will need to pass the 'for' + // token as a program name to the command expression parsing + // function since it will be gone from the token stream by that + // time. Thus, we save it. We also need to make sure the sensing + // always leaves the variable name token in t/tt. + // + // Note also that in this model it won't be possible to support + // options in the first form. + // + token pt (t); + assert (pt.type == type::word && pt.value == "for"); + + mode (lexer_mode::for_loop); + next (t, tt); + + // Note that we also consider special variable names (those that + // don't clash with the command line elements like redirects, etc) + // to later fail gracefully. + // + string& n (t.value); + + if (tt == type::word && t.qtype == quote_type::unquoted && + (n[0] == '_' || alpha (n[0]) || // Variable. + n == "~")) // Special variable. + { + // Detect patterns analogous to parse_variable_name() (so we + // diagnose `for x[string]: ...`). + // + if (n.find_first_of ("[*?") != string::npos) + fail (t) << "expected variable name instead of " << n; + + if (special_variable (n)) + fail (t) << "attempt to set '" << n << "' special variable"; + + // Parse out the element attributes, if present. + // + if (lexer_->peek_char ().first == '[') + { + // Save the variable name token before the attributes parsing + // and restore it afterwards. Also make sure that the token + // which follows the attributes stays in the stream. + // + token vt (move (t)); + next_with_attributes (t, tt); + + attributes_push (t, tt, + true /* standalone */, + false /* next_token */); + + t = move (vt); + tt = t.type; + } + + if (lexer_->peek_char ().first == ':') + lt = line_type::cmd_for_args; + } + + if (lt == line_type::cmd_for_stream) // for x <... + { + // At this point t/tt contains the variable name token. Now + // pre-parse the command expression in the command_line lexer + // mode starting from this position and also passing the 'for' + // token as a program name. + // + // Note that the fact that the potential attributes are already + // parsed doesn't affect the command expression pre-parsing. + // Also note that they will be available during the execution + // phase being replayed. + // + expire_mode (); // Expire the for-loop lexer mode. + + parse_command_expr_result r ( + parse_command_expr (t, tt, + lexer::redirect_aliases, + move (pt))); + + assert (r.for_loop); + + if (tt != type::newline) + fail (t) << "expected newline instead of " << t; + + parse_here_documents (t, tt, r); + } + else // for x: ... + { + next (t, tt); + + assert (tt == type::colon); + + expire_mode (); // Expire the for-loop lexer mode. + + // Parse the value similar to the var line type (see above). + // + mode (lexer_mode::variable_line); + parse_variable_line (t, tt); + + if (tt != type::newline) + fail (t) << "expected newline instead of " << t << " after for"; + } + + ln.var = nullptr; + ++level_; + break; + } case line_type::cmd_elif: case line_type::cmd_elifn: case line_type::cmd_else: - case line_type::cmd_end: { - if (!if_line) - { + if (!fct || *fct != line_type::cmd_if) fail (t) << lt << " without preceding 'if'"; - } + } + // Fall through. + case line_type::cmd_end: + { + if (!fct) + fail (t) << lt << " without preceding 'if', 'for', or 'while'"; } // Fall through. case line_type::cmd_if: case line_type::cmd_ifn: + case line_type::cmd_while: next (t, tt); // Skip to start of command. - if (lt == line_type::cmd_if || lt == line_type::cmd_ifn) + if (lt == line_type::cmd_if || + lt == line_type::cmd_ifn || + lt == line_type::cmd_while) ++level_; else if (lt == line_type::cmd_end) --level_; @@ -243,15 +403,24 @@ namespace build2 // Fall through. case line_type::cmd: { - pair<command_expr, here_docs> p; + parse_command_expr_result r; if (lt != line_type::cmd_else && lt != line_type::cmd_end) - p = parse_command_expr (t, tt, lexer::redirect_aliases); + r = parse_command_expr (t, tt, lexer::redirect_aliases); + + if (r.for_loop) + { + lt = line_type::cmd_for_stream; + ln.var = nullptr; + + ++level_; + } if (tt != type::newline) fail (t) << "expected newline instead of " << t; - parse_here_documents (t, tt, p); + parse_here_documents (t, tt, r); + break; } } @@ -269,22 +438,76 @@ namespace build2 *save_line_ = move (ln); } - if (lt == line_type::cmd_if || lt == line_type::cmd_ifn) + switch (lt) { - tt = peek (lexer_mode::first_token); + case line_type::cmd_if: + case line_type::cmd_ifn: + { + tt = peek (lexer_mode::first_token); + + pre_parse_if_else (t, tt); + break; + } + case line_type::cmd_while: + case line_type::cmd_for_stream: + case line_type::cmd_for_args: + { + tt = peek (lexer_mode::first_token); - pre_parse_if_else (t, tt); + pre_parse_loop (t, tt, lt); + break; + } + default: break; } } + // Pre-parse the flow control construct block line. + // + void parser:: + pre_parse_block_line (token& t, type& tt, line_type bt) + { + // enter: peeked first token of the line (type in tt) + // leave: newline + + const location ll (get_location (peeked ())); + + if (tt == type::eos) + fail (ll) << "expected closing 'end'"; + + line_type fct; // Flow control type the block type relates to. + + switch (bt) + { + case line_type::cmd_if: + case line_type::cmd_ifn: + case line_type::cmd_elif: + case line_type::cmd_elifn: + case line_type::cmd_else: + { + fct = line_type::cmd_if; + break; + } + case line_type::cmd_while: + case line_type::cmd_for_stream: + case line_type::cmd_for_args: + { + fct = bt; + break; + } + default: assert(false); + } + + pre_parse_line (t, tt, fct); + assert (tt == type::newline); + } + void parser:: pre_parse_if_else (token& t, type& tt) { // enter: peeked first token of next line (type in tt) // leave: newline - // Parse lines until we see closing 'end'. Nested if-else blocks are - // handled recursively. + // Parse lines until we see closing 'end'. // for (line_type bt (line_type::cmd_if); // Current block. ; @@ -292,25 +515,21 @@ namespace build2 { const location ll (get_location (peeked ())); - if (tt == type::eos) - fail (ll) << "expected closing 'end'"; - // Parse one line. Note that this one line can still be multiple - // lines in case of if-else. In this case we want to view it as - // cmd_if, not cmd_end. Thus remember the start position of the - // next logical line. + // lines in case of a flow control construct. In this case we want + // to view it as cmd_if, not cmd_end. Thus remember the start + // position of the next logical line. // size_t i (script_->body.size ()); - pre_parse_line (t, tt, true /* if_line */); - assert (tt == type::newline); + pre_parse_block_line (t, tt, bt); line_type lt (script_->body[i].type); // First take care of 'end'. // if (lt == line_type::cmd_end) - return; + break; // Check if-else block sequencing. // @@ -334,6 +553,29 @@ namespace build2 } } + void parser:: + pre_parse_loop (token& t, type& tt, line_type lt) + { + // enter: peeked first token of next line (type in tt) + // leave: newline + + assert (lt == line_type::cmd_while || + lt == line_type::cmd_for_stream || + lt == line_type::cmd_for_args); + + // Parse lines until we see closing 'end'. + // + for (;; tt = peek (lexer_mode::first_token)) + { + size_t i (script_->body.size ()); + + pre_parse_block_line (t, tt, lt); + + if (script_->body[i].type == line_type::cmd_end) + break; + } + } + command_expr parser:: parse_command_line (token& t, type& tt) { @@ -344,12 +586,12 @@ namespace build2 // assert (!pre_parse_); - pair<command_expr, here_docs> p ( + parse_command_expr_result pr ( parse_command_expr (t, tt, lexer::redirect_aliases)); assert (tt == type::newline); - parse_here_documents (t, tt, p); + parse_here_documents (t, tt, pr); assert (tt == type::newline); // @@ Note that currently running programs via a runner (e.g., see @@ -362,7 +604,7 @@ namespace build2 // passed to the environment constructor, similar to passing the // script deadline. // - return move (p.first); + return move (pr.expr); } // @@ -423,6 +665,12 @@ namespace build2 fail (l) << "'" << v << "' call via 'env' builtin"; }; + auto diag_loc = [this] () + { + assert (!diag_preamble_.empty ()); + return diag_preamble_.back ().tokens[0].location (); + }; + if (v == "diag") { verify (); @@ -439,24 +687,41 @@ namespace build2 } else // Custom diagnostics. { - assert (diag_line_); - fail (l) << "multiple 'diag' builtin calls" << - info (diag_line_->second) << "previous call is here"; + info (diag_loc ()) << "previous call is here"; } } - // Instruct the parser to save the diag builtin line separately - // from the script lines, when it is fully parsed. Note that it - // will be executed prior to the script body execution to obtain - // the custom diagnostics. + // Move the script body to the end of the diag preamble. // - diag_line_ = make_pair (line (), l); - save_line_ = &diag_line_->first; - diag_weight_ = 4; + // Note that we move into the preamble whatever is there and delay + // the check until the execution (see the depdb preamble + // collecting for the reasoning). + // + lines& ls (script_->body); + diag_preamble_.insert (diag_preamble_.end (), + make_move_iterator (ls.begin ()), + make_move_iterator (ls.end ())); + ls.clear (); + + // Also move the body_temp_dir flag, if it is true. + // + if (script_->body_temp_dir) + { + script_->diag_preamble_temp_dir = true; + script_->body_temp_dir = false; + } - diag_name_ = nullopt; - diag_name2_ = nullopt; + // Similar to the depdb preamble collection, instruct the parser + // to save the depdb builtin line separately from the script + // lines. + // + diag_preamble_.push_back (line ()); + save_line_ = &diag_preamble_.back (); + + diag_weight_ = 4; + diag_name_ = nullopt; + diag_name2_ = nullopt; // Note that the rest of the line contains the builtin argument to // be printed, thus we parse it in the value lexer mode. @@ -478,17 +743,16 @@ namespace build2 { if (a != perform_update_id) fail (l) << "'depdb' builtin cannot be used to " - << ctx.meta_operation_table[a.meta_operation ()].name - << ' ' << ctx.operation_table[a.operation ()]; + << ctx->meta_operation_table[a.meta_operation ()].name + << ' ' << ctx->operation_table[a.operation ()]; } if (!file_based_) - fail (l) << "'depdb' builtin can only be used for file-based " - << "targets"; + fail (l) << "'depdb' builtin can only be used for file- or " + << "file group-based targets"; - if (diag_line_) - fail (diag_line_->second) - << "'diag' builtin call before 'depdb' call" << + if (!diag_preamble_.empty ()) + fail (diag_loc ()) << "'diag' builtin call before 'depdb' call" << info (l) << "'depdb' call is here"; // Note that the rest of the line contains the builtin command @@ -567,8 +831,18 @@ namespace build2 fail (l) << "multiple 'depdb dyndep' calls" << info (depdb_dyndep_->first) << "previous call is here"; - if (peek () == type::word && peeked ().value == "--byproduct") - depdb_dyndep_byproduct_ = true; + if (peek () == type::word) + { + const string& v (peeked ().value); + + // Note: --byproduct and --dyn-target are mutually + // exclusive. + // + if (v == "--byproduct") + depdb_dyndep_byproduct_ = true; + else if (v == "--dyn-target") + depdb_dyndep_dyn_target_ = true; + } } else { @@ -577,6 +851,8 @@ namespace build2 info (depdb_dyndep_->first) << "'depdb dyndep' call is here"; } + depdb_value_ = depdb_value_ || (v == "string" || v == "hash"); + // Move the script body to the end of the depdb preamble. // // Note that at this (pre-parsing) stage we cannot evaluate if @@ -601,10 +877,12 @@ namespace build2 script_->body_temp_dir = false; } - // Reset the impure function call info since it's valid for the - // depdb preamble. + // Reset the impure function call and computed variable + // expansion tracking since both are valid for the depdb + // preamble. // impure_func_ = nullopt; + computed_var_ = nullopt; // Instruct the parser to save the depdb builtin line separately // from the script lines, when it is fully parsed. Note that the @@ -656,10 +934,53 @@ namespace build2 // // This is also the reason why we add a diag frame. // + // The problem turned out to be worse than originally thought: we + // may call a function (for example, as part of if) with invalid + // arguments. And this could happen in the depdb preamble, which + // means we cannot fix this by moving the depdb builtin (which must + // come after the preamble). So let's peek at what's ahead and omit + // the expansion if it's anything iffy, namely, eval context or + // function call. + // + bool skip_diag (false); if (pre_parse_ && diag_weight_ != 4) { - pre_parse_ = false; // Make parse_names() perform expansions. - pre_parse_suspended_ = true; + // Based on the buildfile expansion parsing logic. + // + if (tt == type::lparen) // Evaluation context. + skip_diag = true; + else if (tt == type::dollar) + { + type ptt (peek (lexer_mode::variable)); + + if (!peeked ().separated) + { + if (ptt == type::lparen) + { + // While strictly speaking this can also be a function call, + // this is highly unusual and we will assume it's a variable + // expansion. + } + else if (ptt == type::word) + { + pair<char, bool> r (lexer_->peek_char ()); + + if (r.first == '(' && !r.second) // Function call. + skip_diag = true; + } + } + } + + if (!skip_diag) + { + // Sanity check: we should not be suspending the pre-parse mode + // turned on by the base parser. + // + assert (top_pre_parse_); + + pre_parse_ = false; // Make parse_names() perform expansions. + pre_parse_suspended_ = true; + } } auto df = make_diag_frame ( @@ -687,7 +1008,7 @@ namespace build2 pre_parse_ = true; } - if (pre_parse_ && diag_weight_ == 4) + if (pre_parse_ && (diag_weight_ == 4 || skip_diag)) return nullopt; } @@ -709,6 +1030,19 @@ namespace build2 return nullopt; } + // If this is a value of the special cmdline type, then only do + // certain tests below if the value is not quoted and doesn't contain + // any characters that would be consumed by re-lexing. + // + // This is somewhat of a hack but handling this properly would not + // only require unquoting but also keeping track of which special + // characters were quoted (and thus should be treated literally) and + // which were not (and thus should act as separators, etc). + // + bool qs (pr.type != nullptr && + pr.type->is_a<cmdline> () && + need_cmdline_relex (ns[0].value)); + // We have to handle process_path[_ex] and executable target. The // process_path[_ex] we may have to recognize syntactically because // of the loss of type, for example: @@ -742,10 +1076,14 @@ namespace build2 pp_vt = pr.type; ns.clear (); } - else if (ns[0].file ()) + else if (ns[0].file () && !qs) { // Find the end of the value. // + // Note that here we ignore the whole cmdline issue (see above) + // for the further values assuming that they are unquoted and + // don't contain any special characters. + // auto b (ns.begin ()); auto i (value_traits<process_path_ex>::find_end (ns)); @@ -812,40 +1150,45 @@ namespace build2 // else if (!ns[0].simple ()) { - if (const target* t = search_existing ( - ns[0], *scope_, ns[0].pair ? ns[1].dir : empty_dir_path)) + if (!qs) { - if (const auto* et = t->is_a<exe> ()) + // This could be a script from src so search like a prerequisite. + // + if (const target* t = search_existing ( + ns[0], *scope_, ns[0].pair ? ns[1].dir : empty_dir_path)) { - if (pre_parse_) + if (const auto* et = t->is_a<exe> ()) { - if (auto* n = et->lookup_metadata<string> ("name")) + if (pre_parse_) { - set_diag (*n, 3); - return nullopt; + if (auto* n = et->lookup_metadata<string> ("name")) + { + set_diag (*n, 3); + return nullopt; + } + // Fall through. } - // Fall through. - } - else - { - process_path pp (et->process_path ()); + else + { + process_path pp (et->process_path ()); - if (pp.empty ()) - fail (l) << "target " << *et << " is out of date" << - info << "consider specifying it as a prerequisite of " - << environment_->target; + if (pp.empty ()) + fail (l) << "target " << *et << " is out of date" << + info << "consider specifying it as a prerequisite of " + << environment_->target; - ns.erase (ns.begin (), ns.begin () + (ns[0].pair ? 2 : 1)); - return optional<process_path> (move (pp)); + ns.erase (ns.begin (), ns.begin () + (ns[0].pair ? 2 : 1)); + return optional<process_path> (move (pp)); + } } - } - if (pre_parse_) - { - diag_record dr (fail (l)); - dr << "unable to deduce low-verbosity script diagnostics name " - << "from target " << *t; - suggest_diag (dr); + if (pre_parse_) + { + diag_record dr (fail (l)); + dr << "unable to deduce low-verbosity script diagnostics name " + << "from target " << *t; + suggest_diag (dr); + } } } @@ -863,26 +1206,29 @@ namespace build2 { // If we are here, the name is simple and is not part of a pair. // - string& v (ns[0].value); + if (!qs) + { + string& v (ns[0].value); - // Try to interpret the name as a builtin. - // - const builtin_info* bi (builtins.find (v)); + // Try to interpret the name as a builtin. + // + const builtin_info* bi (builtins.find (v)); - if (bi != nullptr) - { - set_diag (move (v), bi->weight); - return nullopt; - } - // - // Try to interpret the name as a pseudo-builtin. - // - // Note that both of them has the zero weight and cannot be picked - // up as a script name. - // - else if (v == "set" || v == "exit") - { - return nullopt; + if (bi != nullptr) + { + set_diag (move (v), bi->weight); + return nullopt; + } + // + // Try to interpret the name as a pseudo-builtin. + // + // Note that both of them has the zero weight and cannot be picked + // up as a script name. + // + else if (v == "set" || v == "exit") + { + return nullopt; + } } diag_record dr (fail (l)); @@ -907,8 +1253,9 @@ namespace build2 // Note that we rely on "small function object" optimization here. // auto exec_cmd = [this] (token& t, build2::script::token_type& tt, - size_t li, + const iteration_index* ii, size_t li, bool single, + const function<command_function>& cf, const location& ll) { // We use the 0 index to signal that this is the only command. @@ -919,7 +1266,7 @@ namespace build2 command_expr ce ( parse_command_line (t, static_cast<token_type&> (tt))); - runner_->run (*environment_, ce, li, ll); + runner_->run (*environment_, ce, ii, li, cf, ll); }; exec_lines (s.body, exec_cmd); @@ -928,11 +1275,30 @@ namespace build2 runner_->leave (e, s.end_loc); } + // Return true if the specified expression executes the set builtin or + // is a for-loop. + // + static bool + valid_preamble_cmd (const command_expr& ce, + const function<command_function>& cf) + { + return find_if ( + ce.begin (), ce.end (), + [&cf] (const expr_term& et) + { + const process_path& p (et.pipe.back ().program); + return p.initial == nullptr && + (p.recall.string () == "set" || + (cf != nullptr && p.recall.string () == "for")); + }) != ce.end (); + } + void parser:: - exec_depdb_preamble (action a, const scope& bs, const file& t, + exec_depdb_preamble (action a, const scope& bs, const target& t, environment& e, const script& s, runner& r, lines_iterator begin, lines_iterator end, depdb& dd, + dynamic_targets* dyn_targets, bool* update, optional<timestamp> mt, bool* deferred_failure, @@ -955,28 +1321,34 @@ namespace build2 action a; const scope& bs; - const file& t; + const target& t; environment& env; const script& scr; depdb& dd; + dynamic_targets* dyn_targets; bool* update; bool* deferred_failure; optional<timestamp> mt; dyndep_byproduct* byp; - } data {trace, a, bs, t, e, s, dd, update, deferred_failure, mt, byp}; + } data { + trace, + a, bs, t, + e, s, + dd, dyn_targets, update, deferred_failure, mt, byp}; auto exec_cmd = [this, &data] (token& t, build2::script::token_type& tt, - size_t li, + const iteration_index* ii, size_t li, bool /* single */, + const function<command_function>& cf, const location& ll) { // Note that we never reset the line index to zero (as we do in - // execute_body()) assuming that there are some script body - // commands to follow. + // execute_body()) assuming that there are some script body commands + // to follow. // if (tt == type::word && t.value == "depdb") { @@ -995,8 +1367,9 @@ namespace build2 // exec_depdb_dyndep (t, tt, li, ll, - data.a, data.bs, const_cast<file&> (data.t), + data.a, data.bs, const_cast<target&> (data.t), data.dd, + *data.dyn_targets, *data.update, *data.mt, *data.deferred_failure, @@ -1006,35 +1379,29 @@ namespace build2 { names ns (exec_special (t, tt, true /* skip <cmd> */)); + string v; + const char* w (nullptr); if (cmd == "hash") { sha256 cs; for (const name& n: ns) to_checksum (cs, n); - if (data.dd.expect (cs.string ()) != nullptr) - l4 ([&] { - data.trace (ll) - << "'depdb hash' argument change forcing update of " - << data.t;}); + v = cs.string (); + w = "argument"; } else if (cmd == "string") { - string s; try { - s = convert<string> (move (ns)); + v = convert<string> (move (ns)); } catch (const invalid_argument& e) { fail (ll) << "invalid 'depdb string' argument: " << e; } - if (data.dd.expect (s) != nullptr) - l4 ([&] { - data.trace (ll) - << "'depdb string' argument change forcing update of " - << data.t;}); + w = "argument"; } else if (cmd == "env") { @@ -1055,14 +1422,32 @@ namespace build2 fail (ll) << pf << e; } - if (data.dd.expect (cs.string ()) != nullptr) - l4 ([&] { - data.trace (ll) - << "'depdb env' environment change forcing update of " - << data.t;}); + v = cs.string (); + w = "environment"; } else assert (false); + + // Prefix the value with the type letter. This serves two + // purposes: + // + // 1. It makes sure the result is never a blank line. We use + // blank lines as anchors to skip directly to certain entries + // (e.g., dynamic targets). + // + // 2. It allows us to detect the beginning of prerequisites + // since an absolute path will be distinguishable from these + // entries (in the future we may want to add an explicit + // blank after such custom entries to make this easier). + // + v.insert (0, 1, ' '); + v.insert (0, 1, cmd[0]); // `h`, `s`, or `e` + + if (data.dd.expect (v) != nullptr) + l4 ([&] { + data.trace (ll) + << "'depdb " << cmd << "' " << w << " change forcing " + << "update of " << data.t;}); } } else @@ -1070,15 +1455,7 @@ namespace build2 command_expr ce ( parse_command_line (t, static_cast<token_type&> (tt))); - // Verify that this expression executes the set builtin. - // - if (find_if (ce.begin (), ce.end (), - [] (const expr_term& et) - { - const process_path& p (et.pipe.back ().program); - return p.initial == nullptr && - p.recall.string () == "set"; - }) == ce.end ()) + if (!valid_preamble_cmd (ce, cf)) { const replay_tokens& rt (data.scr.depdb_preamble.back ().tokens); assert (!rt.empty ()); @@ -1089,20 +1466,93 @@ namespace build2 info (rt[0].location ()) << "depdb preamble ends here"; } - runner_->run (*environment_, ce, li, ll); + runner_->run (*environment_, ce, ii, li, cf, ll); } }; exec_lines (begin, end, exec_cmd); } + pair<names, location> parser:: + execute_diag_preamble (const scope& rs, const scope& bs, + environment& e, const script& s, runner& r, + bool diag, bool enter, bool leave) + { + tracer trace ("execute_diag_preamble"); + + assert (!s.diag_preamble.empty ()); + + const line& dl (s.diag_preamble.back ()); // Diag builtin line. + + pre_exec (rs, bs, e, &s, &r); + + if (enter) + runner_->enter (e, s.start_loc); + + // Perform the variable assignments. + // + auto exec_cmd = [&dl, this] (token& t, + build2::script::token_type& tt, + const iteration_index* ii, size_t li, + bool /* single */, + const function<command_function>& cf, + const location& ll) + { + // Note that we never reset the line index to zero (as we do in + // execute_body()) assuming that there are some script body commands + // to follow. + // + command_expr ce ( + parse_command_line (t, static_cast<token_type&> (tt))); + + if (!valid_preamble_cmd (ce, cf)) + { + const replay_tokens& rt (dl.tokens); + assert (!rt.empty ()); + + fail (ll) << "disallowed command in diag preamble" << + info << "only variable assignments are allowed in diag preamble" + << info (rt[0].location ()) << "diag preamble ends here"; + } + + runner_->run (*environment_, ce, ii, li, cf, ll); + }; + + exec_lines (s.diag_preamble.begin (), s.diag_preamble.end () - 1, + exec_cmd); + + // Execute the diag line, if requested. + // + names ns; + + if (diag) + { + // Copy the tokens and start playing. + // + replay_data (replay_tokens (dl.tokens)); + + token t; + build2::script::token_type tt; + next (t, tt); + + ns = exec_special (t, tt, true /* skip_first */); + + replay_stop (); + } + + if (leave) + runner_->leave (e, s.end_loc); + + return make_pair (ns, dl.tokens.front ().location ()); + } + void parser:: pre_exec (const scope& rs, const scope& bs, environment& e, const script* s, runner* r) { path_ = nullptr; // Set by replays. - pre_parse_ = false; + top_pre_parse_ = pre_parse_ = false; set_lexer (nullptr); @@ -1152,22 +1602,37 @@ namespace build2 apply_value_attributes (&var, lhs, move (rhs), kind); }; - auto exec_if = [this] (token& t, build2::script::token_type& tt, - size_t li, - const location& ll) + auto exec_cond = [this] (token& t, build2::script::token_type& tt, + const iteration_index* ii, size_t li, + const location& ll) { command_expr ce ( parse_command_line (t, static_cast<token_type&> (tt))); - // Assume if-else always involves multiple commands. + // Assume a flow control construct always involves multiple + // commands. // - return runner_->run_if (*environment_, ce, li, ll); + return runner_->run_cond (*environment_, ce, ii, li, ll); }; - build2::script::parser::exec_lines (begin, end, - exec_set, exec_cmd, exec_if, - environment_->exec_line, - &environment_->var_pool); + auto exec_for = [this] (const variable& var, + value&& val, + const attributes& val_attrs, + const location&) + { + value& lhs (environment_->assign (var)); + + attributes_.push_back (val_attrs); + + apply_value_attributes (&var, lhs, move (val), type::assign); + }; + + build2::script::parser::exec_lines ( + begin, end, + exec_set, exec_cmd, exec_cond, exec_for, + nullptr /* iteration_index */, + environment_->exec_line, + &environment_->var_pool); } names parser:: @@ -1184,33 +1649,12 @@ namespace build2 : names (); } - names parser:: - execute_special (const scope& rs, const scope& bs, - environment& e, - const line& ln, - bool omit_builtin) - { - pre_exec (rs, bs, e, nullptr /* script */, nullptr /* runner */); - - // Copy the tokens and start playing. - // - replay_data (replay_tokens (ln.tokens)); - - token t; - build2::script::token_type tt; - next (t, tt); - - names r (exec_special (t, tt, omit_builtin)); - - replay_stop (); - return r; - } - void parser:: exec_depdb_dyndep (token& lt, build2::script::token_type& ltt, size_t li, const location& ll, - action a, const scope& bs, file& t, + action a, const scope& bs, target& t, depdb& dd, + dynamic_targets& dyn_targets, bool& update, timestamp mt, bool& deferred_failure, @@ -1223,6 +1667,7 @@ namespace build2 depdb_dyndep_options ops; bool prog (false); bool byprod (false); + bool dyn_tgt (false); // Prerequisite update filter (--update-*). // @@ -1265,11 +1710,9 @@ namespace build2 next (t, tt); // Skip the 'dyndep' command. - if (tt == type::word && t.value == "--byproduct") - { - byprod = true; + if (tt == type::word && ((byprod = (t.value == "--byproduct")) || + (dyn_tgt = (t.value == "--dyn-target")))) next (t, tt); - } assert (byprod == (byprod_result != nullptr)); @@ -1488,10 +1931,23 @@ namespace build2 continue; } - // Handle --byproduct in the wrong place. + // Handle --byproduct and --dyn-target in the wrong place. // if (strcmp (a, "--byproduct") == 0) - fail (ll) << "depdb dyndep: --byproduct must be first option"; + { + fail (ll) << "depdb dyndep: " + << (dyn_tgt + ? "--byproduct specified with --dyn-target" + : "--byproduct must be first option"); + } + + if (strcmp (a, "--dyn-target") == 0) + { + fail (ll) << "depdb dyndep: " + << (byprod + ? "--dyn-target specified with --byproduct" + : "--dyn-target must be first option"); + } // Handle non-literal --update-*. // @@ -1516,29 +1972,31 @@ namespace build2 } } - // --what - // - const char* what (ops.what_specified () - ? ops.what ().c_str () - : "file"); - // --format // dyndep_format format (dyndep_format::make); - if (ops.format_specified ()) { const string& f (ops.format ()); - if (f != "make") + if (f == "lines") format = dyndep_format::lines; + else if (f != "make") fail (ll) << "depdb dyndep: invalid --format option value '" << f << "'"; } + // Prerequisite-specific options. + // + + // --what + // + const char* what (ops.what_specified () + ? ops.what ().c_str () + : "file"); + // --cwd // optional<dir_path> cwd; - if (ops.cwd_specified ()) { if (!byprod) @@ -1558,28 +2016,6 @@ namespace build2 fail (ll) << "depdb dyndep: -I specified with --byproduct"; } - // --file - // - // Note that if --file is specified without a program, then we assume - // it is one of the static prerequisites. - // - optional<path> file; - - if (ops.file_specified ()) - { - file = move (ops.file ()); - - if (file->relative ()) - { - if (!cwd) - fail (ll) << "depdb dyndep: relative path specified with --file"; - - *file = *cwd / *file; - } - } - else if (!prog) - fail (ll) << "depdb dyndep: program or --file expected"; - // --default-type // // Get the default prerequisite type falling back to file{} if not @@ -1591,7 +2027,7 @@ namespace build2 // translation unit would want to make sure it resolves extracted // system headers to h{} targets analogous to the c module's rule. // - const target_type* def_pt; + const target_type* def_pt (&file::static_type); if (ops.default_type_specified ()) { const string& t (ops.default_type ()); @@ -1599,10 +2035,8 @@ namespace build2 def_pt = bs.find_target_type (t); if (def_pt == nullptr) fail (ll) << "depdb dyndep: unknown target type '" << t - << "' specific with --default-type"; + << "' specified with --default-type"; } - else - def_pt = &file::static_type; // --adhoc // @@ -1612,6 +2046,93 @@ namespace build2 fail (ll) << "depdb dyndep: --adhoc specified with --byproduct"; } + // Target-specific options. + // + + // --target-what + // + const char* what_tgt ("file"); + if (ops.target_what_specified ()) + { + if (!dyn_tgt) + fail (ll) << "depdb dyndep: --target-what specified without " + << "--dyn-target"; + + what_tgt = ops.target_what ().c_str (); + } + + // --target-cwd + // + optional<dir_path> cwd_tgt; + if (ops.target_cwd_specified ()) + { + if (!dyn_tgt) + fail (ll) << "depdb dyndep: --target-cwd specified without " + << "--dyn-target"; + + cwd_tgt = move (ops.target_cwd ()); + + if (cwd_tgt->relative ()) + fail (ll) << "depdb dyndep: relative path specified with " + << "--target-cwd"; + } + + // --target-default-type + // + const target_type* def_tt (&file::static_type); + if (ops.target_default_type_specified ()) + { + if (!dyn_tgt) + fail (ll) << "depdb dyndep: --target-default-type specified " + << "without --dyn-target"; + + const string& t (ops.target_default_type ()); + + def_tt = bs.find_target_type (t); + if (def_tt == nullptr) + fail (ll) << "depdb dyndep: unknown target type '" << t + << "' specified with --target-default-type"; + } + + map<string, const target_type*> map_tt; + if (ops.target_extension_type_specified ()) + { + if (!dyn_tgt) + fail (ll) << "depdb dyndep: --target-extension-type specified " + << "without --dyn-target"; + + for (pair<const string, string>& p: ops.target_extension_type ()) + { + const target_type* tt (bs.find_target_type (p.second)); + if (tt == nullptr) + fail (ll) << "depdb dyndep: unknown target type '" << p.second + << "' specified with --target-extension-type"; + + map_tt[p.first] = tt; + } + } + + // --file (last since need --*cwd) + // + // Note that if --file is specified without a program, then we assume + // it is one of the static prerequisites. + // + optional<path> file; + if (ops.file_specified ()) + { + file = move (ops.file ()); + + if (file->relative ()) + { + if (!cwd && !cwd_tgt) + fail (ll) << "depdb dyndep: relative path specified with --file"; + + *file = (cwd ? *cwd : *cwd_tgt) / *file; + } + } + else if (!prog) + fail (ll) << "depdb dyndep: program or --file expected"; + // Update prerequisite targets. // using dyndep = dyndep_rule; @@ -1632,7 +2153,7 @@ namespace build2 // So there is a nuanced interaction between update=match and // --update-*. // - if ((p.include & 4) != 0) + if ((p.include & adhoc_buildscript_rule::include_unmatch) != 0) { l6 ([&]{trace << "skipping unmatched " << *pt;}); continue; @@ -1705,6 +2226,11 @@ namespace build2 update = dyndep::update ( trace, a, *pt, update ? timestamp_unknown : mt) || update; + // While implicit, it is for a static prerequisite, so marking it + // feels correct. + // + p.include |= prerequisite_target::include_udm; + // Mark as updated (see execute_update_prerequisites() for // details. // @@ -1737,6 +2263,10 @@ namespace build2 return; } + const scope& rs (*bs.root_scope ()); + + group* g (t.is_a<group> ()); // If not group then file. + // This code is based on the prior work in the cc module (specifically // extract_headers()) where you can often find more detailed rationale // for some of the steps performed. @@ -1834,9 +2364,29 @@ namespace build2 command_expr cmd; srcout_map so_map; + // Save/restore script cleanups. + // + struct cleanups + { + build2::script::cleanups ordinary; + paths special; + }; + optional<cleanups> script_cleanups; + + auto cleanups_guard = make_guard ( + [this, &script_cleanups] () + { + if (script_cleanups) + { + swap (environment_->cleanups, script_cleanups->ordinary); + swap (environment_->special_cleanups, script_cleanups->special); + } + }); + auto init_run = [this, &ctx, <, <t, &ll, - prog, &file, &ops, &cmd, &so_map] () + prog, &file, &ops, + &cmd, &so_map, &script_cleanups] () { // Populate the srcout map with the -I$out_base -I$src_base pairs. // @@ -1849,6 +2399,10 @@ namespace build2 if (prog) { + script_cleanups = cleanups {}; + swap (environment_->cleanups, script_cleanups->ordinary); + swap (environment_->special_cleanups, script_cleanups->special); + cmd = parse_command_line (lt, static_cast<token_type&> (ltt)); // If the output goes to stdout, then this should be a single @@ -1864,15 +2418,10 @@ namespace build2 // they include the line index in their names to avoid clashes // between lines). // - // Cleanups are not an issue, they will simply replaced. And + // Cleanups are not an issue, they will simply be replaced. And // overriding the contents of the special files seems harmless and // consistent with what would happen if the command redirects its // output to a non-special file. - // - if (file) - environment_->clean ( - {build2::script::cleanup_type::always, *file}, - true /* implicit */); } }; @@ -1882,7 +2431,7 @@ namespace build2 size_t skip_count (0); auto add = [this, &trace, what, - a, &bs, &t, &pts, pts_n = pts.size (), + a, &bs, &t, g, &pts, pts_n = pts.size (), &ops, &map_ext, def_pt, &pfx_map, &so_map, &dd, &skip_count] (path fp, size_t* skip, @@ -1892,6 +2441,61 @@ namespace build2 bool cache (skip == nullptr); + // Handle fsdir{} prerequisite separately. + // + // Note: inspired by inject_fsdir(). + // + if (fp.to_directory ()) + { + if (!cache) + { + // Note: already absolute since cannot be non-existent. + // + fp.normalize (); + } + + const fsdir* dt (&search<fsdir> (t, + path_cast<dir_path> (fp), + dir_path (), + string (), nullptr, nullptr)); + + // Subset of code for file below. + // + if (!cache) + { + for (size_t i (0); i != pts_n; ++i) + { + const prerequisite_target& p (pts[i]); + + if (const target* pt = + (p.target != nullptr ? p.target : + p.adhoc () ? reinterpret_cast<target*> (p.data) : + nullptr)) + { + if (dt == pt) + return false; + } + } + + if (*skip != 0) + { + --(*skip); + return false; + } + } + + match_sync (a, *dt); + pts.push_back ( + prerequisite_target ( + nullptr, true /* adhoc */, reinterpret_cast<uintptr_t> (dt))); + + if (!cache) + dd.expect (fp.representation ()); + + skip_count++; + return false; + } + // We can only defer the failure if we will be running the recipe // body. // @@ -1946,13 +2550,26 @@ namespace build2 // Skip if this is one of the targets. // + // Note that for dynamic targets this only works if we see the + // targets before prerequisites (like in the make dependency + // format). + // if (ops.drop_cycles ()) { - for (const target* m (&t); m != nullptr; m = m->adhoc_member) + if (g != nullptr) { - if (ft == m) + auto& ms (g->members); + if (find (ms.begin (), ms.end (), ft) != ms.end ()) return false; } + else + { + for (const target* m (&t); m != nullptr; m = m->adhoc_member) + { + if (ft == m) + return false; + } + } } // Skip until where we left off. @@ -1981,10 +2598,15 @@ namespace build2 { prerequisite_target& pt (pts.back ()); + // Note: set the include_target flag for consistency (the + // updated_during_match() check does not apply since it's a + // dynamic prerequisite). + // if (pt.adhoc ()) { pt.data = reinterpret_cast<uintptr_t> (pt.target); pt.target = nullptr; + pt.include |= prerequisite_target::include_target; } else pt.data = 1; // Already updated. @@ -2018,13 +2640,27 @@ namespace build2 << t; }); + // While in the make format targets come before prerequisites, in + // depdb we store them after since any change to prerequisites can + // invalidate the set of targets. So we save them first and process + // later. + // + // Note also that we need to return them to the caller in case we are + // updating. + // If nothing so far has invalidated the dependency database, then try // the cached data before running the program. // bool cache (!update); + bool skip_blank (false); for (bool restart (true), first_run (true); restart; cache = false) { + // Clear the state in case we are restarting. + // + if (dyn_tgt) + dyn_targets.clear (); + restart = false; if (cache) @@ -2033,7 +2669,8 @@ namespace build2 // assert (skip_count == 0); - // We should always end with a blank line. + // We should always end with a blank line after the list of + // dynamic prerequisites. // for (;;) { @@ -2047,8 +2684,11 @@ namespace build2 break; } - if (l->empty ()) // Done, nothing changed. - return; + if (l->empty ()) // Done with prerequisites, nothing changed. + { + skip_blank = true; + break; + } if (optional<bool> r = add (path (move (*l)), nullptr, mt)) { @@ -2070,6 +2710,52 @@ namespace build2 return; } } + + if (!restart) // Nothing changed. + { + if (dyn_tgt) + { + // We should always end with a blank line after the list of + // dynamic targets. + // + for (;;) + { + string* l (dd.read ()); + + // If the line is invalid, run the compiler. + // + if (l == nullptr) + { + restart = true; + break; + } + + if (l->empty ()) // Done with targets. + break; + + // Split into type and path (see below for background). + // + size_t p (l->find (' ')); + if (p == string::npos || // Invalid format. + p == 0 || // Empty type. + p + 1 == l->size ()) // Empty path. + { + dd.write (); // Invalidate this line. + restart = true; + break; + } + + string t (*l, 0, p); + l->erase (0, p + 1); + + dyn_targets.push_back ( + dynamic_target {move (t), path (move (*l))}); + } + } + + if (!restart) // Done, nothing changed. + break; // Break earliy to keep cache=true. + } } else { @@ -2078,9 +2764,16 @@ namespace build2 init_run (); first_run = false; } - else if (!prog) + else { - fail (ll) << "generated " << what << " without program to retry"; + if (!prog) + fail (ll) << "generated " << what << " without program to retry"; + + // Drop dyndep cleanups accumulated on the previous run. + // + assert (script_cleanups); // Sanity check. + environment_->cleanups.clear (); + environment_->special_cleanups.clear (); } // Save the timestamp just before we run the command. If we depend @@ -2101,18 +2794,50 @@ namespace build2 istringstream iss; if (prog) { - string s; - build2::script::run (*environment_, - cmd, - li, - ll, - !file ? &s : nullptr); - + // Note: depdb is disallowed inside flow control constructs. + // if (!file) { - iss.str (move (s)); + function<command_function> cf ( + [&iss] + (build2::script::environment&, + const strings&, + auto_fd in, + pipe_command* pipe, + const optional<deadline>& dl, + const location& ll) + { + read (move (in), + false /* whitespace */, + false /* newline */, + true /* exact */, + [&iss] (string&& s) {iss.str (move (s));}, + pipe, + dl, + ll, + "depdb-dyndep"); + }); + + build2::script::run (*environment_, + cmd, + nullptr /* iteration_index */, li, + ll, + cf, false /* last_cmd */); + iss.exceptions (istream::badbit); } + else + { + build2::script::run ( + *environment_, cmd, nullptr /* iteration_index */, li, ll); + + // Note: make it a maybe-cleanup in case the command cleans it + // up itself. + // + environment_->clean ( + {build2::script::cleanup_type::maybe, *file}, + true /* implicit */); + } } ifdstream ifs (ifdstream::badbit); @@ -2180,32 +2905,72 @@ namespace build2 if (r.second.empty ()) continue; - // @@ TODO: what should we do about targets? + // Skip targets unless requested to extract. // - // Note that if we take GCC as an example, things are + // BTW, if you are wondering why don't we extract targets + // by default, take GCC as an example, where things are // quite messed up: by default it ignores -o and just // takes the source file name and replaces the extension // with a platform-appropriate object file extension. One // can specify a custom target (or even multiple targets) - // with -MT or with -MQ (quoting). Though MinGW GCC still - // does not quote `:` with -MQ. So in this case it's + // with -MT or with -MQ (quoting). So in this case it's // definitely easier for the user to ignore the targets // and just specify everything in the buildfile. // - // On the other hand, other tools are likely to produce - // more sensible output (except perhaps for quoting). - // - // @@ Maybe in the lax mode we should only recognize `:` - // if it's separated on at least one side? - // - // Alternatively, we could detect Windows drives in - // paths and "handle" them (I believe this is what GNU - // make does). Maybe we should have three formats: - // make-lax, make, make-strict? - // if (r.first == make_type::target) + { + // NOTE: similar code below. + // + if (dyn_tgt) + { + path& f (r.second); + + if (f.relative ()) + { + if (!cwd_tgt) + fail (il) << "relative " << what_tgt + << " target path '" << f + << "' in make dependency declaration" << + info << "consider using --target-cwd to specify " + << "relative path base"; + + f = *cwd_tgt / f; + } + + // Note that unlike prerequisites, here we don't need + // normalize_external() since we expect the targets to + // be within this project. + // + try + { + f.normalize (); + } + catch (const invalid_path&) + { + fail (il) << "invalid " << what_tgt << " target " + << "path '" << f.string () << "'"; + } + + // The target must be within this project. + // + if (!f.sub (rs.out_path ())) + { + fail (il) << what_tgt << " target path " << f + << " must be inside project output " + << "directory " << rs.out_path (); + } + + // Note: type is resolved later. + // + dyn_targets.push_back ( + dynamic_target {string (), move (f)}); + } + continue; + } + // NOTE: similar code below. + // if (optional<bool> u = add (move (r.second), &skip, rmt)) { restart = *u; @@ -2233,20 +2998,380 @@ namespace build2 break; } - break; + break; // case + } + case dyndep_format::lines: + { + bool tgt (dyn_tgt); // Reading targets or prerequisites. + + for (string l; !restart; ++il.line) // Reuse the buffer. + { + if (eof (getline (is, l))) + break; + + if (l.empty ()) + { + if (!tgt) + fail (il) << "blank line in prerequisites list"; + + tgt = false; // Targets/prerequisites separating blank. + continue; + } + + // See if this line start with space to indicate a non- + // existent prerequisite. This variable serves both as a + // flag and as a position of the beginning of the path. + // + size_t n (l.front () == ' ' ? 1 : 0); + + if (tgt) + { + // NOTE: similar code above. + // + path f; + try + { + // Non-existent target doesn't make sense. + // + if (n) + throw invalid_path (""); + + f = path (l); + + if (f.relative ()) + { + if (!cwd_tgt) + fail (il) << "relative " << what_tgt + << " target path '" << f + << "' in lines dependency declaration" << + info << "consider using --target-cwd to specify " + << "relative path base"; + + f = *cwd_tgt / f; + } + + // Note that unlike prerequisites, here we don't need + // normalize_external() since we expect the targets to + // be within this project. + // + f.normalize (); + } + catch (const invalid_path&) + { + fail (il) << "invalid " << what_tgt << " target path '" + << l << "'"; + } + + // The target must be within this project. + // + if (!f.sub (rs.out_path ())) + { + fail (il) << what_tgt << " target path " << f + << " must be inside project output directory " + << rs.out_path (); + } + + // Note: type is resolved later. + // + dyn_targets.push_back ( + dynamic_target {string (), move (f)}); + } + else + { + path f; + try + { + f = path (l.c_str () + n, l.size () - n); + + if (f.empty () || + (n && f.to_directory ())) // Non-existent fsdir{}. + throw invalid_path (""); + + if (f.relative ()) + { + if (!n) + { + if (!cwd) + fail (il) << "relative " << what + << " prerequisite path '" << f + << "' in lines dependency declaration" << + info << "consider using --cwd to specify " + << "relative path base"; + + f = *cwd / f; + } + } + else if (n) + { + // @@ TODO: non-existent absolute paths. + // + throw invalid_path (""); + } + } + catch (const invalid_path&) + { + fail (il) << "invalid " << what << " prerequisite path '" + << l << "'"; + } + + // NOTE: similar code above. + // + if (optional<bool> u = add (move (f), &skip, rmt)) + { + restart = *u; + + if (restart) + { + update = true; + l6 ([&]{trace << "restarting";}); + } + } + else + { + // Trigger recompilation, mark as expected to fail, and + // bail out. + // + update = true; + deferred_failure = true; + break; + } + } + } + + break; // case } } + if (file) + ifs.close (); + // Bail out early if we have deferred a failure. // if (deferred_failure) return; + + // Clean after each depdb-dyndep execution. + // + if (prog) + clean (*environment_, ll); } } - // Add the terminating blank line (we are updating depdb). + // Add the dynamic prerequisites terminating blank line if we are + // updating depdb and unless it's already there. + // + if (!cache && !skip_blank) + dd.expect (""); + + // Handle dynamic targets. // - dd.expect (""); + if (dyn_tgt) + { + if (g != nullptr && g->members_static == 0 && dyn_targets.empty ()) + fail (ll) << "group " << *g << " has no static or dynamic members"; + + // There is one more level (at least that we know of) to this rabbit + // hole: if the set of dynamic targets changes between clean and + // update and we do a `clean update` batch, then we will end up with + // old targets (as entered by clean from old depdb information) + // being present during update. So we need to clean them out. + // + // Optimize this for a first/single batch (common case) by noticing + // that there are only real targets to start with. + // + // Note that this doesn't affect explicit groups where we reset the + // members on each update (see adhoc_rule_buildscript::apply()). + // + optional<vector<const target*>> dts; + if (g == nullptr) + { + for (const target* m (&t); m != nullptr; m = m->adhoc_member) + { + if (m->decl != target_decl::real) + dts = vector<const target*> (); + } + } + + struct map_ext_data + { + const char* what_tgt; + const map<string, const target_type*>& map_tt; + const path* f; // Updated on each iteration. + } d {what_tgt, map_tt, nullptr}; + + function<dyndep::map_extension_func> map_ext ( + [this, &d] (const scope& bs, const string& n, const string& e) + { + small_vector<const target_type*, 2> tts; + + // Check the custom mapping first. + // + auto i (d.map_tt.find (e)); + if (i != d.map_tt.end ()) + tts.push_back (i->second); + else + { + tts = dyndep::map_extension (bs, n, e, nullptr); + + // Issue custom diagnostics suggesting --target-extension-type. + // + if (tts.size () > 1) + { + diag_record dr (fail); + + dr << "mapping of " << d.what_tgt << " target path " << *d.f + << " to target type is ambiguous"; + + for (const target_type* tt: tts) + dr << info << "can be " << tt->name << "{}"; + + dr << info << "use --target-extension-type to provide custom " + << "mapping"; + } + } + + return tts; + }); + + function<dyndep::group_filter_func> filter; + if (g != nullptr) + { + // Skip static/duplicate members in explicit group. + // + filter = [] (mtime_target& g, const build2::file& m) + { + auto& ms (g.as<group> ().members); + return find (ms.begin (), ms.end (), &m) == ms.end (); + }; + } + + // Unlike for prerequisites, for targets we store in depdb both the + // resolved target type and path. The target type is used in clean + // (see adhoc_rule_buildscript::apply()) where we cannot easily get + // hold of all the dyndep options to map the path to target type. + // So the format of the target line is: + // + // <type> <path> + // + string l; // Reuse the buffer. + for (dynamic_target& dt: dyn_targets) + { + const path& f (dt.path); + + d.f = &f; // Current file being mapped. + + // Note that this logic should be consistent with what we have in + // adhoc_buildscript_rule::apply() for perform_clean. + // + const build2::file* ft (nullptr); + if (g != nullptr) + { + pair<const build2::file&, bool> r ( + dyndep::inject_group_member ( + what_tgt, + a, bs, *g, + f, // Can't move since need to return dyn_targets. + map_ext, *def_tt, filter)); + + // Note: no target_decl shenanigans since reset the members on + // each update. + // + if (!r.second) + { + dt.type.clear (); // Static indicator. + continue; + } + + ft = &r.first; + + // Note: we only currently support dynamic file members so it + // will be file if first. + // + g->members.push_back (ft); + } + else + { + pair<const build2::file&, bool> r ( + dyndep::inject_adhoc_group_member ( + what_tgt, + a, bs, t, + f, // Can't move since need to return dyn_targets. + map_ext, *def_tt)); + + // Note that we have to track the dynamic target even if it was + // already a member (think `b update && b clean update`). + // + if (!r.second && r.first.decl == target_decl::real) + { + dt.type.clear (); // Static indicator. + continue; + } + + ft = &r.first; + + if (dts) + dts->push_back (ft); + } + + const char* tn (ft->type ().name); + + if (dt.type.empty ()) + dt.type = tn; + else if (dt.type != tn) + { + // This can, for example, happen if the user changed the + // extension to target type mapping. Say swapped extension + // variable values of two target types. + // + fail << "mapping of " << what_tgt << " target path " << f + << " to target type has changed" << + info << "previously mapped to " << dt.type << "{}" << + info << "now mapped to " << tn << "{}" << + info << "perform from scratch rebuild of " << t; + } + + if (!cache) + { + l = dt.type; + l += ' '; + l += f.string (); + dd.expect (l); + } + } + + // Add the dynamic targets terminating blank line. + // + if (!cache) + dd.expect (""); + + // Clean out old dynamic targets (skip the primary member). + // + if (dts) + { + assert (g == nullptr); + + for (target* p (&t); p->adhoc_member != nullptr; ) + { + target* m (p->adhoc_member); + + if (m->decl != target_decl::real) + { + // While there could be quite a few dynamic targets (think + // something like Doxygen), this will hopefully be optimized + // down to a contiguous memory region scan for an integer and + // so should be fast. + // + if (find (dts->begin (), dts->end (), m) == dts->end ()) + { + p->adhoc_member = m->adhoc_member; // Drop m. + continue; + } + } + + p = m; + } + } + } // Reload $< and $> to make sure they contain the newly discovered // prerequisites and targets. @@ -2255,7 +3380,8 @@ namespace build2 environment_->set_special_variables (a); } - // When add a special variable don't forget to update lexer::word(). + // When add a special variable don't forget to update lexer::word() and + // for-loop parsing in pre_parse_line(). // bool parser:: special_variable (const string& n) noexcept @@ -2264,15 +3390,24 @@ namespace build2 } lookup parser:: - lookup_variable (name&& qual, string&& name, const location& loc) + lookup_variable (names&& qual, string&& name, const location& loc) { // In the pre-parse mode collect the referenced variable names for the // script semantics change tracking. // + // Note that during pre-parse a computed (including qualified) name + // is signalled as an empty name. + // if (pre_parse_ || pre_parse_suspended_) { lookup r; + // Note that pre-parse can be switched on by the base parser even + // during execute. + // + if (!top_pre_parse_) + return r; + // Add the variable name skipping special variables and suppressing // duplicates, unless the default variables change tracking is // canceled with `depdb clear`. While at it, check if the script @@ -2288,10 +3423,8 @@ namespace build2 { if (pre_parse_suspended_) { - const variable* pvar (scope_->ctx.var_pool.find (name)); - - if (pvar != nullptr) - r = (*scope_)[*pvar]; + if (const variable* var = scope_->var_pool ().find (name)) + r = (*scope_)[*var]; } if (!depdb_clear_) @@ -2302,12 +3435,27 @@ namespace build2 vars.push_back (move (name)); } } + else + { + // What about pre_parse_suspended_? Don't think it makes sense to + // diagnose this since it can be indirect (that is, via an + // intermediate variable). + // + if (perform_update_ && file_based_ && !computed_var_) + computed_var_ = loc; + } return r; } if (!qual.empty ()) - fail (loc) << "qualified variable name"; + { + // Qualified variable is computed and we expect the user to track + // its changes manually. + // + return build2::script::parser::lookup_variable ( + move (qual), move (name), loc); + } lookup r (environment_->lookup (name)); @@ -2318,13 +3466,13 @@ namespace build2 // diag builtin argument change (which can be affected by such a // variable expansion) doesn't affect the script semantics and the // depdb argument is specifically used for the script semantics change - // tracking. We also omit this check it the depdb builtin is used in - // the script, assuming that such variables are tracked manually, if - // required. + // tracking. We also omit this check if the depdb "value" (string, + // hash) builtin is used in the script, assuming that such variables + // are tracked manually, if required. // if (script_ != nullptr && !script_->depdb_clear && - script_->depdb_preamble.empty ()) + !script_->depdb_value) { if (r.defined () && !r.belongs (*environment_)) { @@ -2342,9 +3490,12 @@ namespace build2 void parser:: lookup_function (string&& name, const location& loc) { - if (perform_update_ && file_based_ && !impure_func_) + // Note that pre-parse can be switched on by the base parser even + // during execute. + // + if (top_pre_parse_ && perform_update_ && file_based_ && !impure_func_) { - const function_overloads* f (ctx.functions.find (name)); + const function_overloads* f (ctx->functions.find (name)); if (f != nullptr && !f->pure) impure_func_ = make_pair (move (name), loc); |