summaryrefslogtreecommitdiffstats
path: root/src/bin/netconf/netconf_lexer.ll
diff options
context:
space:
mode:
Diffstat (limited to '')
-rw-r--r--src/bin/netconf/netconf_lexer.ll842
1 files changed, 842 insertions, 0 deletions
diff --git a/src/bin/netconf/netconf_lexer.ll b/src/bin/netconf/netconf_lexer.ll
new file mode 100644
index 0000000..5f056a6
--- /dev/null
+++ b/src/bin/netconf/netconf_lexer.ll
@@ -0,0 +1,842 @@
+/* Copyright (C) 2018-2020 Internet Systems Consortium, Inc. ("ISC")
+
+ This Source Code Form is subject to the terms of the Mozilla Public
+ License, v. 2.0. If a copy of the MPL was not distributed with this
+ file, You can obtain one at http://mozilla.org/MPL/2.0/. */
+
+%{ /* -*- C++ -*- */
+
+/* Generated files do not make clang static analyser so happy */
+#ifndef __clang_analyzer__
+
+#include <cctype>
+#include <cerrno>
+#include <climits>
+#include <cstdlib>
+#include <string>
+#include <netconf/parser_context.h>
+#include <asiolink/io_address.h>
+#include <boost/lexical_cast.hpp>
+#include <exceptions/exceptions.h>
+#include <cc/dhcp_config_error.h>
+
+/* Please avoid C++ style comments (// ... eol) as they break flex 2.6.2 */
+
+/* Work around an incompatibility in flex (at least versions
+ 2.5.31 through 2.5.33): it generates code that does
+ not conform to C89. See Debian bug 333231
+ <http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=333231>. */
+# undef yywrap
+# define yywrap() 1
+
+namespace {
+
+bool start_token_flag = false;
+
+isc::netconf::ParserContext::ParserType start_token_value;
+unsigned int comment_start_line = 0;
+
+using namespace isc;
+using isc::netconf::NetconfParser;
+
+};
+
+/* To avoid the call to exit... oops! */
+#define YY_FATAL_ERROR(msg) isc::netconf::ParserContext::fatal(msg)
+%}
+
+/* noyywrap disables automatic rewinding for the next file to parse. Since we
+ always parse only a single string, there's no need to do any wraps. And
+ using yywrap requires linking with -lfl, which provides the default yywrap
+ implementation that always returns 1 anyway. */
+%option noyywrap
+
+/* nounput simplifies the lexer, by removing support for putting a character
+ back into the input stream. We never use such capability anyway. */
+%option nounput
+
+/* batch means that we'll never use the generated lexer interactively. */
+%option batch
+
+/* avoid to get static global variables to remain with C++. */
+/* in last resort %option reentrant */
+
+/* Enables debug mode. To see the debug messages, one needs to also set
+ yy_flex_debug to 1, then the debug messages will be printed on stderr. */
+%option debug
+
+/* I have no idea what this option does, except it was specified in the bison
+ examples and Postgres folks added it to remove gcc 4.3 warnings. Let's
+ be on the safe side and keep it. */
+%option noinput
+
+%x COMMENT
+%x DIR_ENTER DIR_INCLUDE DIR_EXIT
+
+/* These are not token expressions yet, just convenience expressions that
+ can be used during actual token definitions. Note some can match
+ incorrect inputs (e.g., IP addresses) which must be checked. */
+int \-?[0-9]+
+blank [ \t\r]
+
+UnicodeEscapeSequence u[0-9A-Fa-f]{4}
+JSONEscapeCharacter ["\\/bfnrt]
+JSONEscapeSequence {JSONEscapeCharacter}|{UnicodeEscapeSequence}
+JSONStandardCharacter [^\x00-\x1f"\\]
+JSONStringCharacter {JSONStandardCharacter}|\\{JSONEscapeSequence}
+JSONString \"{JSONStringCharacter}*\"
+
+/* for errors */
+
+BadUnicodeEscapeSequence u[0-9A-Fa-f]{0,3}[^0-9A-Fa-f"]
+BadJSONEscapeSequence [^"\\/bfnrtu]|{BadUnicodeEscapeSequence}
+ControlCharacter [\x00-\x1f]
+ControlCharacterFill [^"\\]|\\["\\/bfnrtu]
+
+%{
+/* This code run each time a pattern is matched. It updates the location
+ by moving it ahead by yyleng bytes. yyleng specifies the length of the
+ currently matched token. */
+#define YY_USER_ACTION driver.loc_.columns(yyleng);
+%}
+
+%%
+
+%{
+ /* This part of the code is copied over to the verbatim to the top
+ of the generated yylex function. Explanation:
+ http://www.gnu.org/software/bison/manual/html_node/Multiple-start_002dsymbols.html */
+
+ /* Code run each time yylex is called. */
+ driver.loc_.step();
+
+ /* We currently have 3 points of entries defined:
+ START_JSON - which expects any valid JSON
+ START_NETCONF - which expects full configuration (with outer map and Netconf
+ object in it.
+ START_SUB_NETCONF - which expects only content of the Netconf, this is
+ primarily useful for testing. */
+ if (start_token_flag) {
+ start_token_flag = false;
+ switch (start_token_value) {
+ case ParserContext::PARSER_JSON:
+ default:
+ return NetconfParser::make_START_JSON(driver.loc_);
+ case ParserContext::PARSER_NETCONF:
+ return NetconfParser::make_START_NETCONF(driver.loc_);
+ case ParserContext::PARSER_SUB_NETCONF:
+ return NetconfParser::make_START_SUB_NETCONF(driver.loc_);
+ }
+ }
+%}
+
+#.* ;
+
+"//"(.*) ;
+
+"/*" {
+ BEGIN(COMMENT);
+ comment_start_line = driver.loc_.end.line;;
+}
+
+<COMMENT>"*/" BEGIN(INITIAL);
+<COMMENT>. ;
+<COMMENT><<EOF>> {
+ isc_throw(ParseError, "Comment not closed. (/* in line " << comment_start_line);
+}
+
+"<?" BEGIN(DIR_ENTER);
+<DIR_ENTER>"include" BEGIN(DIR_INCLUDE);
+<DIR_INCLUDE>\"([^\"\n])+\" {
+ /* Include directive. */
+
+ /* Extract the filename. */
+ std::string tmp(yytext+1);
+ tmp.resize(tmp.size() - 1);
+
+ driver.includeFile(tmp);
+}
+<DIR_ENTER,DIR_INCLUDE,DIR_EXIT><<EOF>> {
+ isc_throw(ParseError, "Directive not closed.");
+}
+<DIR_EXIT>"?>" BEGIN(INITIAL);
+
+
+<*>{blank}+ {
+ /* Ok, we found a with space. Let's ignore it and update loc variable. */
+ driver.loc_.step();
+}
+
+<*>[\n]+ {
+ /* Newline found. Let's update the location and continue. */
+ driver.loc_.lines(yyleng);
+ driver.loc_.step();
+}
+
+
+\"Netconf\" {
+ switch(driver.ctx_) {
+ case ParserContext::CONFIG:
+ return NetconfParser::make_NETCONF(driver.loc_);
+ default:
+ return NetconfParser::make_STRING("Netconf", driver.loc_);
+ }
+}
+
+\"user-context\" {
+ switch(driver.ctx_) {
+ case ParserContext::NETCONF:
+ case ParserContext::SERVER:
+ case ParserContext::CONTROL_SOCKET:
+ case ParserContext::LOGGERS:
+ return NetconfParser::make_USER_CONTEXT(driver.loc_);
+ default:
+ return NetconfParser::make_STRING("user-context", driver.loc_);
+ }
+}
+
+\"comment\" {
+ switch(driver.ctx_) {
+ case ParserContext::NETCONF:
+ case ParserContext::SERVER:
+ case ParserContext::CONTROL_SOCKET:
+ case ParserContext::LOGGERS:
+ return NetconfParser::make_COMMENT(driver.loc_);
+ default:
+ return NetconfParser::make_STRING("comment", driver.loc_);
+ }
+}
+
+\"boot-update\" {
+ switch(driver.ctx_) {
+ case ParserContext::NETCONF:
+ case ParserContext::SERVER:
+ return NetconfParser::make_BOOT_UPDATE(driver.loc_);
+ default:
+ return NetconfParser::make_STRING("boot-update", driver.loc_);
+ }
+}
+
+\"subscribe-changes\" {
+ switch(driver.ctx_) {
+ case ParserContext::NETCONF:
+ case ParserContext::SERVER:
+ return NetconfParser::make_SUBSCRIBE_CHANGES(driver.loc_);
+ default:
+ return NetconfParser::make_STRING("subscribe-changes", driver.loc_);
+ }
+}
+
+\"validate-changes\" {
+ switch(driver.ctx_) {
+ case ParserContext::NETCONF:
+ case ParserContext::SERVER:
+ return NetconfParser::make_VALIDATE_CHANGES(driver.loc_);
+ default:
+ return NetconfParser::make_STRING("validate-changes", driver.loc_);
+ }
+}
+
+\"managed-servers\" {
+ switch(driver.ctx_) {
+ case ParserContext::NETCONF:
+ return NetconfParser::make_MANAGED_SERVERS(driver.loc_);
+ default:
+ return NetconfParser::make_STRING("managed-servers", driver.loc_);
+ }
+}
+
+\"dhcp4\" {
+ switch(driver.ctx_) {
+ case ParserContext::MANAGED_SERVERS:
+ return NetconfParser::make_DHCP4_SERVER(driver.loc_);
+ default:
+ return NetconfParser::make_STRING("dhcp4", driver.loc_);
+ }
+}
+
+\"dhcp6\" {
+ switch(driver.ctx_) {
+ case ParserContext::MANAGED_SERVERS:
+ return NetconfParser::make_DHCP6_SERVER(driver.loc_);
+ default:
+ return NetconfParser::make_STRING("dhcp6", driver.loc_);
+ }
+}
+
+\"d2\" {
+ switch(driver.ctx_) {
+ case ParserContext::MANAGED_SERVERS:
+ return NetconfParser::make_D2_SERVER(driver.loc_);
+ default:
+ return NetconfParser::make_STRING("d2", driver.loc_);
+ }
+}
+
+\"ca\" {
+ switch(driver.ctx_) {
+ case ParserContext::MANAGED_SERVERS:
+ return NetconfParser::make_CA_SERVER(driver.loc_);
+ default:
+ return NetconfParser::make_STRING("ca", driver.loc_);
+ }
+}
+
+\"model\" {
+ switch(driver.ctx_) {
+ case ParserContext::SERVER:
+ return NetconfParser::make_MODEL(driver.loc_);
+ default:
+ return NetconfParser::make_STRING("model", driver.loc_);
+ }
+}
+
+\"control-socket\" {
+ switch(driver.ctx_) {
+ case ParserContext::SERVER:
+ return NetconfParser::make_CONTROL_SOCKET(driver.loc_);
+ default:
+ return NetconfParser::make_STRING("control-socket", driver.loc_);
+ }
+}
+
+\"unix\" {
+ switch(driver.ctx_) {
+ case ParserContext::SOCKET_TYPE:
+ return NetconfParser::make_UNIX(driver.loc_);
+ default:
+ return NetconfParser::make_STRING("unix", driver.loc_);
+ }
+}
+
+\"http\" {
+ switch(driver.ctx_) {
+ case ParserContext::SOCKET_TYPE:
+ return NetconfParser::make_HTTP(driver.loc_);
+ default:
+ return NetconfParser::make_STRING("http", driver.loc_);
+ }
+}
+
+\"stdout\" {
+ switch(driver.ctx_) {
+ case ParserContext::SOCKET_TYPE:
+ return NetconfParser::make_STDOUT(driver.loc_);
+ default:
+ return NetconfParser::make_STRING("stdout", driver.loc_);
+ }
+}
+
+\"socket-type\" {
+ switch(driver.ctx_) {
+ case ParserContext::CONTROL_SOCKET:
+ return NetconfParser::make_SOCKET_TYPE(driver.loc_);
+ default:
+ return NetconfParser::make_STRING("socket-type", driver.loc_);
+ }
+}
+
+\"socket-name\" {
+ switch(driver.ctx_) {
+ case ParserContext::CONTROL_SOCKET:
+ return NetconfParser::make_SOCKET_NAME(driver.loc_);
+ default:
+ return NetconfParser::make_STRING("socket-name", driver.loc_);
+ }
+}
+
+\"socket-url\" {
+ switch(driver.ctx_) {
+ case ParserContext::CONTROL_SOCKET:
+ return NetconfParser::make_SOCKET_URL(driver.loc_);
+ default:
+ return NetconfParser::make_STRING("socket-url", driver.loc_);
+ }
+}
+
+\"hooks-libraries\" {
+ switch(driver.ctx_) {
+ case ParserContext::NETCONF:
+ return NetconfParser::make_HOOKS_LIBRARIES(driver.loc_);
+ default:
+ return NetconfParser::make_STRING("hooks-libraries", driver.loc_);
+ }
+}
+
+\"library\" {
+ switch(driver.ctx_) {
+ case ParserContext::HOOKS_LIBRARIES:
+ return NetconfParser::make_LIBRARY(driver.loc_);
+ default:
+ return NetconfParser::make_STRING("library", driver.loc_);
+ }
+}
+
+\"parameters\" {
+ switch(driver.ctx_) {
+ case ParserContext::HOOKS_LIBRARIES:
+ return NetconfParser::make_PARAMETERS(driver.loc_);
+ default:
+ return NetconfParser::make_STRING("parameters", driver.loc_);
+ }
+}
+
+\"loggers\" {
+ switch(driver.ctx_) {
+ case ParserContext::NETCONF:
+ return NetconfParser::make_LOGGERS(driver.loc_);
+ default:
+ return NetconfParser::make_STRING("loggers", driver.loc_);
+ }
+}
+
+\"name\" {
+ switch(driver.ctx_) {
+ case ParserContext::LOGGERS:
+ return NetconfParser::make_NAME(driver.loc_);
+ default:
+ return NetconfParser::make_STRING("name", driver.loc_);
+ }
+}
+
+\"output_options\" {
+ switch(driver.ctx_) {
+ case ParserContext::LOGGERS:
+ return NetconfParser::make_OUTPUT_OPTIONS(driver.loc_);
+ default:
+ return NetconfParser::make_STRING("output_options", driver.loc_);
+ }
+}
+
+\"output\" {
+ switch(driver.ctx_) {
+ case ParserContext::OUTPUT_OPTIONS:
+ return NetconfParser::make_OUTPUT(driver.loc_);
+ default:
+ return NetconfParser::make_STRING("output", driver.loc_);
+ }
+}
+
+\"flush\" {
+ switch(driver.ctx_) {
+ case ParserContext::OUTPUT_OPTIONS:
+ return NetconfParser::make_FLUSH(driver.loc_);
+ default:
+ return NetconfParser::make_STRING("flush", driver.loc_);
+ }
+}
+
+\"maxsize\" {
+ switch(driver.ctx_) {
+ case ParserContext::OUTPUT_OPTIONS:
+ return NetconfParser::make_MAXSIZE(driver.loc_);
+ default:
+ return NetconfParser::make_STRING("maxsize", driver.loc_);
+ }
+}
+
+\"maxver\" {
+ switch(driver.ctx_) {
+ case ParserContext::OUTPUT_OPTIONS:
+ return NetconfParser::make_MAXVER(driver.loc_);
+ default:
+ return NetconfParser::make_STRING("maxver", driver.loc_);
+ }
+}
+
+\"pattern\" {
+ switch(driver.ctx_) {
+ case ParserContext::OUTPUT_OPTIONS:
+ return NetconfParser::make_PATTERN(driver.loc_);
+ default:
+ return NetconfParser::make_STRING("pattern", driver.loc_);
+ }
+}
+
+\"debuglevel\" {
+ switch(driver.ctx_) {
+ case ParserContext::LOGGERS:
+ return NetconfParser::make_DEBUGLEVEL(driver.loc_);
+ default:
+ return NetconfParser::make_STRING("debuglevel", driver.loc_);
+ }
+}
+
+\"severity\" {
+ switch(driver.ctx_) {
+ case ParserContext::LOGGERS:
+ return NetconfParser::make_SEVERITY(driver.loc_);
+ default:
+ return NetconfParser::make_STRING("severity", driver.loc_);
+ }
+}
+
+{JSONString} {
+ /* A string has been matched. It contains the actual string and single quotes.
+ We need to get those quotes out of the way and just use its content, e.g.
+ for 'foo' we should get foo */
+ std::string raw(yytext+1);
+ size_t len = raw.size() - 1;
+ raw.resize(len);
+ std::string decoded;
+ decoded.reserve(len);
+ for (size_t pos = 0; pos < len; ++pos) {
+ int b = 0;
+ char c = raw[pos];
+ switch (c) {
+ case '"':
+ /* impossible condition */
+ driver.error(driver.loc_, "Bad quote in \"" + raw + "\"");
+ break;
+ case '\\':
+ ++pos;
+ if (pos >= len) {
+ /* impossible condition */
+ driver.error(driver.loc_, "Overflow escape in \"" + raw + "\"");
+ }
+ c = raw[pos];
+ switch (c) {
+ case '"':
+ case '\\':
+ case '/':
+ decoded.push_back(c);
+ break;
+ case 'b':
+ decoded.push_back('\b');
+ break;
+ case 'f':
+ decoded.push_back('\f');
+ break;
+ case 'n':
+ decoded.push_back('\n');
+ break;
+ case 'r':
+ decoded.push_back('\r');
+ break;
+ case 't':
+ decoded.push_back('\t');
+ break;
+ case 'u':
+ /* support only \u0000 to \u00ff */
+ ++pos;
+ if (pos + 4 > len) {
+ /* impossible condition */
+ driver.error(driver.loc_,
+ "Overflow unicode escape in \"" + raw + "\"");
+ }
+ if ((raw[pos] != '0') || (raw[pos + 1] != '0')) {
+ driver.error(driver.loc_,
+ "Unsupported unicode escape in \"" + raw + "\"",
+ pos + 1);
+ }
+ pos += 2;
+ c = raw[pos];
+ if ((c >= '0') && (c <= '9')) {
+ b = (c - '0') << 4;
+ } else if ((c >= 'A') && (c <= 'F')) {
+ b = (c - 'A' + 10) << 4;
+ } else if ((c >= 'a') && (c <= 'f')) {
+ b = (c - 'a' + 10) << 4;
+ } else {
+ /* impossible condition */
+ driver.error(driver.loc_, "Not hexadecimal in unicode escape in \"" + raw + "\"");
+ }
+ pos++;
+ c = raw[pos];
+ if ((c >= '0') && (c <= '9')) {
+ b |= c - '0';
+ } else if ((c >= 'A') && (c <= 'F')) {
+ b |= c - 'A' + 10;
+ } else if ((c >= 'a') && (c <= 'f')) {
+ b |= c - 'a' + 10;
+ } else {
+ /* impossible condition */
+ driver.error(driver.loc_, "Not hexadecimal in unicode escape in \"" + raw + "\"");
+ }
+ decoded.push_back(static_cast<char>(b & 0xff));
+ break;
+ default:
+ /* impossible condition */
+ driver.error(driver.loc_, "Bad escape in \"" + raw + "\"");
+ }
+ break;
+ default:
+ if ((c >= 0) && (c < 0x20)) {
+ /* impossible condition */
+ driver.error(driver.loc_, "Invalid control in \"" + raw + "\"");
+ }
+ decoded.push_back(c);
+ }
+ }
+
+ return NetconfParser::make_STRING(decoded, driver.loc_);
+}
+
+\"{JSONStringCharacter}*{ControlCharacter}{ControlCharacterFill}*\" {
+ /* Bad string with a forbidden control character inside */
+ std::string raw(yytext+1);
+ size_t len = raw.size() - 1;
+ size_t pos = 0;
+ for (; pos < len; ++pos) {
+ char c = raw[pos];
+ if ((c >= 0) && (c < 0x20)) {
+ break;
+ }
+ }
+ driver.error(driver.loc_,
+ "Invalid control in " + std::string(yytext),
+ pos + 1);
+}
+
+\"{JSONStringCharacter}*\\{BadJSONEscapeSequence}[^"]*\" {
+ /* Bad string with a bad escape inside */
+ std::string raw(yytext+1);
+ size_t len = raw.size() - 1;
+ size_t pos = 0;
+ bool found = false;
+ for (; pos < len; ++pos) {
+ if (found) {
+ break;
+ }
+ char c = raw[pos];
+ if (c == '\\') {
+ ++pos;
+ c = raw[pos];
+ switch (c) {
+ case '"':
+ case '\\':
+ case '/':
+ case 'b':
+ case 'f':
+ case 'n':
+ case 'r':
+ case 't':
+ break;
+ case 'u':
+ if ((pos + 4 > len) ||
+ !std::isxdigit(raw[pos + 1]) ||
+ !std::isxdigit(raw[pos + 2]) ||
+ !std::isxdigit(raw[pos + 3]) ||
+ !std::isxdigit(raw[pos + 4])) {
+ found = true;
+ }
+ break;
+ default:
+ found = true;
+ break;
+ }
+ }
+ }
+ /* The rule stops on the first " including on \" so add ... in this case */
+ std::string trailer = "";
+ if (raw[len - 1] == '\\') {
+ trailer = "...";
+ }
+ driver.error(driver.loc_,
+ "Bad escape in " + std::string(yytext) + trailer,
+ pos);
+}
+
+\"{JSONStringCharacter}*\\\" {
+ /* Bad string with an open escape at the end */
+ std::string raw(yytext+1);
+ driver.error(driver.loc_,
+ "Overflow escape in " + std::string(yytext),
+ raw.size() + 1);
+}
+
+\"{JSONStringCharacter}*\\u[0-9A-Fa-f]{0,3}\" {
+ /* Bad string with an open unicode escape at the end */
+ std::string raw(yytext+1);
+ size_t pos = raw.size() - 1;
+ for (; pos > 0; --pos) {
+ char c = raw[pos];
+ if (c == 'u') {
+ break;
+ }
+ }
+ driver.error(driver.loc_,
+ "Overflow unicode escape in " + std::string(yytext),
+ pos + 1);
+}
+
+"[" { return NetconfParser::make_LSQUARE_BRACKET(driver.loc_); }
+"]" { return NetconfParser::make_RSQUARE_BRACKET(driver.loc_); }
+"{" { return NetconfParser::make_LCURLY_BRACKET(driver.loc_); }
+"}" { return NetconfParser::make_RCURLY_BRACKET(driver.loc_); }
+"," { return NetconfParser::make_COMMA(driver.loc_); }
+":" { return NetconfParser::make_COLON(driver.loc_); }
+
+{int} {
+ /* An integer was found. */
+ std::string tmp(yytext);
+ int64_t integer = 0;
+ try {
+ /* In substring we want to use negative values (e.g. -1).
+ In enterprise-id we need to use values up to 0xffffffff.
+ To cover both of those use cases, we need at least
+ int64_t. */
+ integer = boost::lexical_cast<int64_t>(tmp);
+ } catch (const boost::bad_lexical_cast &) {
+ driver.error(driver.loc_, "Failed to convert " + tmp + " to an integer.");
+ }
+
+ /* The parser needs the string form as double conversion is no lossless */
+ return NetconfParser::make_INTEGER(integer, driver.loc_);
+}
+
+[-+]?[0-9]*\.?[0-9]*([eE][-+]?[0-9]+)? {
+ /* A floating point was found. */
+ std::string tmp(yytext);
+ double fp = 0.0;
+ try {
+ fp = boost::lexical_cast<double>(tmp);
+ } catch (const boost::bad_lexical_cast &) {
+ driver.error(driver.loc_, "Failed to convert " + tmp + " to a floating point.");
+ }
+
+ return NetconfParser::make_FLOAT(fp, driver.loc_);
+}
+
+true|false {
+ string tmp(yytext);
+ return NetconfParser::make_BOOLEAN(tmp == "true", driver.loc_);
+}
+
+null {
+ return NetconfParser::make_NULL_TYPE(driver.loc_);
+}
+
+(?i:true) driver.error (driver.loc_, "JSON true reserved keyword is lower case only");
+
+(?i:false) driver.error (driver.loc_, "JSON false reserved keyword is lower case only");
+
+(?i:null) driver.error (driver.loc_, "JSON null reserved keyword is lower case only");
+
+<*>. driver.error (driver.loc_, "Invalid character: " + std::string(yytext));
+
+<<EOF>> {
+ if (driver.states_.empty()) {
+ return NetconfParser::make_END(driver.loc_);
+ }
+ driver.loc_ = driver.locs_.back();
+ driver.locs_.pop_back();
+ driver.file_ = driver.files_.back();
+ driver.files_.pop_back();
+ if (driver.sfile_) {
+ fclose(driver.sfile_);
+ driver.sfile_ = 0;
+ }
+ if (!driver.sfiles_.empty()) {
+ driver.sfile_ = driver.sfiles_.back();
+ driver.sfiles_.pop_back();
+ }
+ netconf__delete_buffer(YY_CURRENT_BUFFER);
+ netconf__switch_to_buffer(driver.states_.back());
+ driver.states_.pop_back();
+
+ BEGIN(DIR_EXIT);
+}
+
+%%
+
+using namespace isc::dhcp;
+
+void
+ParserContext::scanStringBegin(const std::string& str, ParserType parser_type)
+{
+ start_token_flag = true;
+ start_token_value = parser_type;
+
+ file_ = "<string>";
+ sfile_ = 0;
+ loc_.initialize(&file_);
+ yy_flex_debug = trace_scanning_;
+ YY_BUFFER_STATE buffer;
+ buffer = netconf__scan_bytes(str.c_str(), str.size());
+ if (!buffer) {
+ fatal("cannot scan string");
+ /* fatal() throws an exception so this can't be reached */
+ }
+}
+
+void
+ParserContext::scanFileBegin(FILE * f,
+ const std::string& filename,
+ ParserType parser_type)
+{
+ start_token_flag = true;
+ start_token_value = parser_type;
+
+ file_ = filename;
+ sfile_ = f;
+ loc_.initialize(&file_);
+ yy_flex_debug = trace_scanning_;
+ YY_BUFFER_STATE buffer;
+
+ /* See netconf_lexer.cc header for available definitions */
+ buffer = netconf__create_buffer(f, 65536 /*buffer size*/);
+ if (!buffer) {
+ fatal("cannot scan file " + filename);
+ }
+ netconf__switch_to_buffer(buffer);
+}
+
+void
+ParserContext::scanEnd() {
+ if (sfile_)
+ fclose(sfile_);
+ sfile_ = 0;
+ static_cast<void>(netconf_lex_destroy());
+ /* Close files */
+ while (!sfiles_.empty()) {
+ FILE* f = sfiles_.back();
+ if (f) {
+ fclose(f);
+ }
+ sfiles_.pop_back();
+ }
+ /* Delete states */
+ while (!states_.empty()) {
+ netconf__delete_buffer(states_.back());
+ states_.pop_back();
+ }
+}
+
+void
+ParserContext::includeFile(const std::string& filename) {
+ if (states_.size() > 10) {
+ fatal("Too many nested include.");
+ }
+
+ FILE* f = fopen(filename.c_str(), "r");
+ if (!f) {
+ fatal("Can't open include file " + filename);
+ }
+ if (sfile_) {
+ sfiles_.push_back(sfile_);
+ }
+ sfile_ = f;
+ states_.push_back(YY_CURRENT_BUFFER);
+ YY_BUFFER_STATE buffer;
+ buffer = netconf__create_buffer(f, 65536 /*buffer size*/);
+ if (!buffer) {
+ fatal( "Can't scan include file " + filename);
+ }
+ netconf__switch_to_buffer(buffer);
+ files_.push_back(file_);
+ file_ = filename;
+ locs_.push_back(loc_);
+ loc_.initialize(&file_);
+
+ BEGIN(INITIAL);
+}
+
+namespace {
+/** To avoid unused function error */
+class Dummy {
+ /* cppcheck-suppress unusedPrivateFunction */
+ void dummy() { yy_fatal_error("Fix me: how to disable its definition?"); }
+};
+}
+#endif /* !__clang_analyzer__ */