Snel/src/sequencer.cpp

/*!
 * \file
 *   Sequencer.cpp
 * \brief   A basic sequence interpreter for snel
 *
 *  Created on: Feb, 2019
 *      Author: Christos Choutouridis AEM: 8997
 *      email : <cchoutou@ece.auth.gr>
 */
#include "sequencer.h"

namespace snel {

   //! A type-safe memcpy
   template <typename T>
   void memcpy (T* to, const T* from, size_t n) {
      for (size_t i=0 ; i<n ; ++i)
         *to++ = *from++;
   }

   /*!
    * Split a string to tokens and store them in a container, using a delimiter
    * character.
    * \note
    *    Requires: container with push_back functionality
    * \param str     The input string to split
    * \param cont    The container to push the tokens (MUST have .push_back(T) member)
    * \param delim   The delimiter of type \p T to use
    */
   template <typename T, typename Container>
   void split(const std::basic_string<T>& str, Container& cont, T delim) {
      std::basic_stringstream<T> ss(str);
      std::basic_string<T> token;
      while (std::getline(ss, token, delim)) {
          cont.push_back(token);
      }
   }

   /*!
    * A very very simple filtering for leading ' ', comments and adjustments for
    * the '|' character.
    * \note
    *    The current snel implementation requires '|' to separated
    *    from left and right from the rest of the text, so we adjust the input to fit
    *    this expectation.
    * \param in   Input string
    * \return     Output string
    */
   std::string filter (const std::string in) {
      auto first = in.find_first_not_of(' ');
      std::string nonws = (first != std::string::npos) ?
                          in.substr(first) : "";
      std::string out{};
      char prev {0}, cur, next {0};

      // return empty if the first non-whitespace character is '#'
      if (nonws[0] == '#')
         return out;

      for (auto i = nonws.begin(); i != nonws.end() ; ++i) {
         cur = *i;
         if (cur == '|') {
            if (prev != ' ')  out += ' '; // prev| insert a space before
            next = *++i;
            if (next == '|' || next == ' ') {
               out += cur; cur = next;    // syntax ok
            }
            else {
               out += cur; out += ' ';    // |next, insert a space after
               cur = next;
            }
         }
         prev = cur;
         out += cur;
      }
      return out;
   }


   /*
    * ========== ArgList ==========
    */
   /*!
    * destructor to free up all the allocations
    * \note
    *    Using RAII for ArgList we ensure "no memory leak"
    */
   ArgList::~ArgList() {
      for (vtype p : args_)
         if (p != nullptr) delete[] p;
   }

   /*!
    * A push_back wrapper to our underling vector. This ensures NULL terminating
    * argument list for exec() sys-call.
    * \param item    The argument to push
    * \return        Reference to ArgList for chaining
    */
   ArgList& ArgList::push_back(const std::string& item) {
      vtype it = new type[item.length()+1];

      memcpy (it, item.c_str(), item.length()); // get data and null terminate them
      it[item.length()] = 0;

      args_.back() = it;   // update the argument vector
      args_.push_back(nullptr);
      return *this;
   }


   /*
    * ========== Pipe ===========
    */
   //! Destructor to free up all the resources
   Pipe::~Pipe() {
      // close any possibly leaked pipes
      if (fd[0] != -1)  close(fd[0]);
      if (fd[1] != -1)  close(fd[1]);
   }

   /*
    * ========== Child ==========
    */

   //! Destructor to free up all the resources
   Child::~Child () {
      for (int i=0 ; i<3 ; ++i)
         restore_std_if(i);
   }

   /*!
    * Redirect a \p std_fd file descriptor to a file
    * @param fn      The file name to used as std
    * @param std_fd  The file descriptor to redirect
    */
   void Child::redirect_std_if(std::string fn, fd_t std_fd) {
      if (fn != "") {
         if ((files[std_fd] = openat(AT_FDCWD, fn.c_str(), O_RDWR | O_CREAT, 0640)) == -1)
            throw Error ("Child: Can not open file");
         if ((sstd[std_fd] = dup (std_fd)) == -1)  // save STDxxx
            throw Error ("Child: Can not create file descriptor");
         if ((dup2(files[std_fd], std_fd)) == -1)  // use input as STDIN_
            throw Error ("Child: Can not redirect file descriptor");
      }
   }
   /*!
    * Restore and redirected std file descriptor to its original location
    * \param std_fd  The file descriptor to restore
    */
   void Child::restore_std_if(fd_t std_fd) {
      if (sstd[std_fd] != -1) {
         dup2(sstd[std_fd], std_fd);
         close (sstd[std_fd]);
         sstd[std_fd]  = -1;
      }
      if (files[std_fd] != -1) {
         close (files[std_fd]);
         files[std_fd] = -1;
      }

   }

   /*!
    * Parse the command of the current child to produce the actual arguments
    * \return  Reference to Child for chaining
    */
   Child& Child::make_arguments () {
      bool in{false}, out{false}, err{false};
      bool CanRedirectIn  = (!pipe_.from) ? true : false;
      bool CanRedirectOut = (!pipe_.to) ? true : false;

      for (auto& t: command) {
         if (t == "" || t== " " || t=="|") continue; // skip crap

         if (t == "&&")       logic = LogicOp::AND;
         else if (t == "||")  logic = LogicOp::OR;
         // one pass redirection parsing
         else if (CanRedirectIn && !t.compare(0, 1, "<"))
            filenames[STDIN_]  = t.substr(1);
         else if (CanRedirectOut&& !t.compare(0, 1, ">"))
            filenames[STDOUT_] = t.substr(1);
         else if (CanRedirectOut&& !t.compare(0, 2, "1>"))
            filenames[STDOUT_] = t.substr(2);
         else if (!t.compare(0, 2, "2>"))
            filenames[STDERR_] = t.substr(2);
         // two pass redirection parsing (if redirection came in 2 arguments)
         else if (t == "<")               in  = true;
         else if (t == "1>" || t == ">")  out = true;
         else if (t == "2>")              err = true;
         else if (in) {
            if (CanRedirectIn)
               filenames[STDIN_] = t;
            in = false;
         }
         else if (out) {
            if (CanRedirectOut)
               filenames[STDOUT_] = t;
            out = false;
         }
         else if (err) {
            filenames[STDERR_] = t; err = false;
         }
         else
            arguments.push_back(t);
      }
      return *this;
   }

   /*!
    * \brief   Execute the child in separate process.
    *
    * This is the hart of the snel. We use <tt>fork() - execvp()</tt> pair to execute
    * each child. In case of std redirection, the fd/dup2 handling is made in the
    * parent. In case of piping, the fd handling is made in parent and the
    * dup2() calls in the child process.
    *
    * A child is a member of a vector containing the command chain.
    * For ex:
    * \code ls && cat tralala || uname -a \endcode
    * is a chain with three childs (ls, cat, uname). \see Sequencer
    *
    * \param it      Iterator to the command chain execution
    * \param first   flag to indicate the first command in the chain
    * \return        True if the chain can stop.
    */
   bool Child::execute(std::vector<Child>::iterator it, bool first) {
      bool stop {false};
      Child& previous = (!first) ? *--it : *it; // get previous child safely

      // Check redirection requirements
      redirect_std_if (filenames[STDIN_],  STDIN_);
      redirect_std_if (filenames[STDOUT_], STDOUT_);
      redirect_std_if (filenames[STDERR_], STDERR_);

      // Check parent pipe control
      if (pipe_.to) {
         if((::pipe (pipe_.fd)) == -1)
            throw Error("Child: Can not create pipe");
      }
      pid_t pid = fork();
      if (pid < 0) {
         throw Error("Child: Can not create child process");
      }
      else if (pid == 0) {
         // =========== child ================

         // Some extra pipe checking while we still have our data
         if (pipe_.to) {         // transmitting child
            close(pipe_.fd[0]);  // close the read end (of the child copy)
            if ((dup2(pipe_.fd[1], STDOUT_)) == -1) // redirect output
               throw Error("Child pipe[to]: Can not redirect through pipe");
         }
         else if (!first && pipe_.from) {  // receiving child
            close (previous.pipe().fd[1]); // close the write end (of the child copy)
            if ((dup2(previous.pipe().fd[0], STDIN_)) == -1)  // redirect input
               throw Error("Child pipe[from]: Can not redirect through pipe");
         }

         execvp(arguments.front(), arguments.data());
         // error if we got here
         std::cout << "Can not invoke: " << arguments.front() << std::endl;
         throw Error("Child: Can not run child process");
      }
      else {
         // =========== parent ================
         if (pipe_.to)        close (pipe_.fd[1]); // Pipe fd control
         else if (!first && pipe_.from)
                              close (previous.pipe().fd[0]);

         int exit_status;
         waitpid(pid, &exit_status, 0);   // wait child process to finish
         restore_std_if (STDIN_);         // restore all redirections
         restore_std_if (STDOUT_);
         restore_std_if (STDERR_);

         // Check if we don't have to execute the rest of the chain
         if ( exit_status && logic == LogicOp::AND)   stop = true;
         if (!exit_status && logic == LogicOp::OR)    stop = true;
      }

      return stop;
   }




   /*
    * ======== Sequencer ===========
    */

   /*!
    * First parsing of each line. We split input in ';' to create
    * command chains and split each chain further to create child tokens.
    * These token will be processed further by the child's `make_arguments()`
    *
    * \param input   The input string to parse
    * \return        Reference to Sequencer for chaining
    */
   Sequencer& Sequencer::parse (const std::string& input) {
      std::vector<std::string> chains; // chains are vector of commands
      std::vector<std::string> tokens; // token is any ' ' separated string
      Child::command_t command;

      split (input, chains, ';');      // split input to command chains using ';'
      for (auto& chain : chains) {     // for each chain
         command.clear();              // clear child tokens
         tokens.clear();               // make a new token vector for command
         split (chain, tokens, ' ');   // split chain in to tokens using ' '

         seq_.emplace_back(std::vector<Child>{});     // make a new child for command
         // check tokens inside the chain
         bool from = false;
         for (auto& t: tokens) {
            command.push_back(t);      // get current token
            if (is_seperator(t)) {
               // construct a child with what we have so far and clear command
               seq_.back().emplace_back(command);
               command.clear();
               if (from) {
                  // set from flag to link with the previous child
                  seq_.back().back().pipe().from = from;
                  from = false;
               }

               if (is_pipe(t)) {
                  seq_.back().back().pipe().to = true;   // mark as transmitting child
                  from = true;   // mark to inform the next child as receiving
               }
            }
         }
         seq_.back().emplace_back(command);   // construct the child with tokens
         if (from)
            seq_.back().back().pipe().from = from;
      }
      return *this;
   }

   /*!
    * The main sequencer execution. We recurse for each command chain
    * and each child in the chain and we parse-execute the child
    * \return  Reference to Sequence for chaining
    */
   Sequencer& Sequencer::execute() {
      bool first;
      for (auto& chain : seq_) { // loop in all of the command chains
         first = true;
         for (auto it = chain.begin() ; it != chain.end(); ++it) {
            // loop all of the commands in the chain
            Child& command = *it;
            if (command.make_arguments().execute(it, first))
               break;   // break the chain if child's logic operator says so.
            first = false;
         }
      }
      seq_.clear();  // clear the sequencer's data for the next line (call all the destructors)
      return *this;
   }

}