Simulations.py

    def create(self):
        """
        Main loop for creating the pbs scripts, based on cases, which
        contains the case name as key and tag dictionairy as value
        """

        # dynamically set walltime based on the number of time steps
        # for thyra, make a list so we base the walltime on the slowest case
        self.nr_time_steps = []
        self.duration = []
        self.t0 = []
        # '[time_stop]' '[dt_sim]'

        # REMARK: this i not realy consistent with how the result and log file
        # dirs are allowed to change for each individual case...
        # first check if the pbs_out_dir exists, this dir is considered to be
        # the same for all cases present in cases
        # self.tags['[run_dir]']
        case0 = list(self.cases.keys())[0]
        path = self.cases[case0]['[run_dir]'] + self.pbs_out_dir
        if not os.path.exists(path):
            os.makedirs(path)

        # create pbs_in base dir
        path = self.cases[case0]['[run_dir]'] + self.pbs_in_dir
        if not os.path.exists(path):
            os.makedirs(path)

        # number the pbs jobs:
        count2 = self.pbs_start_number
        # initial cpu count is zero
        count1 = 1
        # scan through all the cases
        i, i_tot = 1, len(self.cases)
        ended = True

        for case in self.cases:

            # get a shorter version for the current cases tag_dict:
            tag_dict = self.cases[case]

            # group all values loaded from the tag_dict here, to keep overview
            # the directories to SAVE the results/logs/turb files
            # load all relevant dir settings: the result/logfile/turbulence/zip
            # they are now also available for starting() and ending() parts
            hawc2_exe = tag_dict['[hawc2_exe]']
            self.case = case.replace('.htc', '')
            self.sim_id = tag_dict['[sim_id]']
            self.results_dir = tag_dict['[res_dir]']
            self.eigenfreq_dir = tag_dict['[eigenfreq_dir]']
            self.logs_dir = tag_dict['[log_dir]']
            self.animation_dir = tag_dict['[animation_dir]']
            self.TurbDirName = tag_dict['[turb_dir]']
            self.TurbDb = tag_dict['[turb_db_dir]']
            self.wakeDb = tag_dict['[wake_db_dir]']
            self.meandDb = tag_dict['[meand_db_dir]']
            self.WakeDirName = tag_dict['[wake_dir]']
            self.MeanderDirName = tag_dict['[meander_dir]']
            self.ModelZipFile = tag_dict['[model_zip]']
            self.htc_dir = tag_dict['[htc_dir]']
            self.hydro_dir = tag_dict['[hydro_dir]']
            self.mooring_dir = tag_dict['[mooring_dir]']
            self.model_path = tag_dict['[run_dir]']
            self.turb_base_name = tag_dict['[turb_base_name]']
            self.wake_base_name = tag_dict['[wake_base_name]']
            self.meand_base_name = tag_dict['[meand_base_name]']
            self.pbs_queue_command = tag_dict['[pbs_queue_command]']
            self.walltime = tag_dict['[walltime]']
            self.dyn_walltime = tag_dict['[auto_walltime]']
            self.case_duration = tag_dict['[duration]']

            # create the pbs_out_dir if necesary
            try:
                path = tag_dict['[run_dir]'] + tag_dict['[pbs_out_dir]']
                if not os.path.exists(path):
                    os.makedirs(path)
                self.pbs_out_dir = tag_dict['[pbs_out_dir]']
            except:
                pass

            # create pbs_in subdirectories if necessary
            try:
                path = tag_dict['[run_dir]'] + tag_dict['[pbs_in_dir]']
                if not os.path.exists(path):
                    os.makedirs(path)
                self.pbs_in_dir = tag_dict['[pbs_in_dir]']
            except:
                pass

            try:
                self.copyback_files = tag_dict['[copyback_files]']
                self.copyback_frename = tag_dict['[copyback_frename]']
            except KeyError:
                pass

            try:
                self.copyto_generic = tag_dict['[copyto_generic]']
                self.copyto_files = tag_dict['[copyto_files]']
            except KeyError:
                pass

            # related to the dynamically setting the walltime
            duration = float(tag_dict['[time_stop]'])
            dt = float(tag_dict['[dt_sim]'])
            self.nr_time_steps.append(duration/dt)
            self.duration.append(float(tag_dict['[duration]']))
            self.t0.append(float(tag_dict['[t0]']))

            if self.verbose:
                print('htc_dir in pbs.create:')
                print(self.htc_dir)
                print(self.model_path)

            # we only start a new case, if we have something that ended before
            # the very first case has to start with starting
            if ended:
                count1 = 1

#                # when jobs depend on other jobs (constant node loading)
#                if self.que_jobdeps:
#                    jobid = self.pref + str(self.jobid_list[i-1])
#                    jobid_dep = self.pref + str(self.jobid_deps[i-1])
#                else:
#                    jobid = self.pref + str(count2)
#                    jobid_dep = None
                if self.short_job_names:
                    jobid = self.pref + str(count2)
                else:
                    jobid = tag_dict['[case_id]']
                if self.pbs_fname_appendix and self.short_job_names:
                    # define the path for the new pbs script
                    pbs_in_fname = '%s_%s.p' % (tag_dict['[case_id]'], jobid)
                else:
                    pbs_in_fname = '%s.p' % (tag_dict['[case_id]'])
                pbs_path = self.model_path + self.pbs_in_dir + pbs_in_fname
                # Start a new pbs script, we only need the tag_dict here
                self.starting(tag_dict, jobid)
                ended = False

            # -----------------------------------------------------------------
            # WRITING THE ACTUAL JOB PARAMETERS

            # browse to the curren scratch directory
            self.pbs += "\n\n"
            self.pbs += '# ' + '-'*60 + '\n'
            # evaluates to true if LAUNCH_PBS_MODE is NOT set
            self.pbs += "# evaluates to true if LAUNCH_PBS_MODE is NOT set\n"
            self.pbs += "if [ -z ${LAUNCH_PBS_MODE+x} ] ; then\n"
            self.pbs += "  echo \n"
            self.pbs += "  echo 'Execute commands on scratch nodes'\n"
            self.pbs += "  cd %s/$USER/$PBS_JOBID\n" % self.node_run_root
            self.pbs += "  # create unique dir for each CPU\n"
            self.pbs += '  mkdir "%i"; cd "%i"\n' % (count1, count1)

            # output the current scratch directory
            self.pbs += "  pwd\n"
            # zip file has been copied to the node before (in start_pbs())
            # unzip now in the CPU scratch directory (zip file is one level up)
            self.pbs += "  /usr/bin/unzip ../" + self.ModelZipFile + '\n'
            # create all directories, especially relevant if there are case
            # dependent sub directories that are not present in the ZIP file
            self.pbs += "  mkdir -p " + self.htc_dir + '\n'
            self.pbs += "  mkdir -p " + self.results_dir + '\n'
            self.pbs += "  mkdir -p " + self.logs_dir + '\n'
            if self.TurbDirName is not None or self.TurbDirName != 'None':
                self.pbs += "  mkdir -p " + self.TurbDirName + '\n'
            if self.WakeDirName and self.WakeDirName != self.TurbDirName:
                self.pbs += "  mkdir -p " + self.WakeDirName + '\n'
            if self.MeanderDirName and self.MeanderDirName != self.TurbDirName:
                self.pbs += "  mkdir -p " + self.MeanderDirName + '\n'
            if self.hydro_dir:
                self.pbs += "  mkdir -p " + self.hydro_dir + '\n'
            # create the eigen analysis dir just in case that is necessary
            if self.eigenfreq_dir:
                self.pbs += '  mkdir -p %s \n' % self.eigenfreq_dir

            # and copy the htc file to the node
            self.pbs += "  cp -R $PBS_O_WORKDIR/" + self.htc_dir \
                + case +" ./" + self.htc_dir + '\n'

            # if there is a turbulence file data base dir, copy from there
            if self.TurbDb:
                turb_dir_src = os.path.join('$PBS_O_WORKDIR', self.TurbDb)
            else:
                turb_dir_src = os.path.join('$PBS_O_WORKDIR', self.TurbDirName)

            # the original behaviour makes assumptions on the turbulence box
            # names: turb_base_name_xxx_u.bin, turb_base_name_xxx_v.bin
            if self.turb_base_name is not None:
                turb_src = os.path.join(turb_dir_src, self.turb_base_name)
                self.pbs += "  cp -R %s*.bin %s \n" % (turb_src, self.TurbDirName)
            # more generally, literally define the names of the boxes for u,v,w
            # components
            elif '[turb_fname_u]' in tag_dict:
                turb_u = os.path.join(turb_dir_src, tag_dict['[turb_fname_u]'])
                turb_v = os.path.join(turb_dir_src, tag_dict['[turb_fname_v]'])
                turb_w = os.path.join(turb_dir_src, tag_dict['[turb_fname_w]'])
                self.pbs += "  cp %s %s \n" % (turb_u, self.TurbDirName)
                self.pbs += "  cp %s %s \n" % (turb_v, self.TurbDirName)
                self.pbs += "  cp %s %s \n" % (turb_w, self.TurbDirName)

            # if there is a turbulence file data base dir, copy from there
            if self.wakeDb and self.WakeDirName:
                wake_dir_src = os.path.join('$PBS_O_WORKDIR', self.wakeDb)
            elif self.WakeDirName:
                wake_dir_src = os.path.join('$PBS_O_WORKDIR', self.WakeDirName)
            if self.wake_base_name is not None:
                wake_src = os.path.join(wake_dir_src, self.wake_base_name)
                self.pbs += "  cp -R %s*.bin %s \n" % (wake_src, self.WakeDirName)

            # if there is a turbulence file data base dir, copy from there
            if self.meandDb and self.MeanderDirName:
                meand_dir_src = os.path.join('$PBS_O_WORKDIR', self.meandDb)
            elif self.MeanderDirName:
                meand_dir_src = os.path.join('$PBS_O_WORKDIR', self.MeanderDirName)
            if self.meand_base_name is not None:
                meand_src = os.path.join(meand_dir_src, self.meand_base_name)
                self.pbs += "  cp -R %s*.bin %s \n" % (meand_src, self.MeanderDirName)

            # copy and rename input files with given versioned name to the
            # required non unique generic version
            for fname, fgen in zip(self.copyto_files, self.copyto_generic):
                self.pbs += "  cp -R $PBS_O_WORKDIR/%s ./%s \n" % (fname, fgen)

            # only apply the wine fix in PBS mode
            self.pbs += '  ' + self.winefix
            # TODO: activate python env, calculate post-processing
#            self.pbs += 'echo `python -c "import wetb; print(wetb.__version__)"`\n'

            # end of the file copying in PBS mode
            self.pbs += '# ' + '-'*60 + '\n'
            self.pbs += "else\n"
            # when in find+xargs mode, browse to the relevant CPU
            self.pbs += '  # with find+xargs we first browse to CPU folder\n'
            self.pbs += '  cd "$CPU_NR"\n'
            self.pbs += "fi\n"
            self.pbs += '# ' + '-'*60 + '\n'

            self.pbs += 'echo ""\n'
            self.pbs += '# evaluates to true if LAUNCH_PBS_MODE is NOT set\n'
            self.pbs += "if [ -z ${LAUNCH_PBS_MODE+x} ] ; then\n"
            # the hawc2 execution commands via wine, in PBS mode fork and wait
            param = (self.wine, hawc2_exe, self.htc_dir+case, self.wine_appendix)
            self.pbs += '  echo "execute HAWC2, fork to background"\n'
            self.pbs += "  %s %s ./%s %s &\n" % param
            # FIXME: running post-processing will only work when 1 HAWC2 job
            # per PBS file, otherwise you have to wait for each job to finish
            # first and then run the post-processing for all those cases
            if self.maxcpu == 1:
                self.pbs += '  wait\n'
                if self.pyenv is not None:
                    self.pbs += '  echo "POST-PROCESSING"\n'
                    self.pbs += '  source activate %s\n' % self.pyenv
                    self.pbs += "  "
                    self.checklogs()
                    self.pbs += "  "
                    self.postprocessing()
                    self.pbs += '  source deactivate\n'
            self.pbs += "else\n"
            param = (self.wine, hawc2_exe, self.htc_dir+case, self.wine_appendix)
            self.pbs += '  echo "execute HAWC2, do not fork and wait"\n'
            self.pbs += "  %s %s ./%s %s \n" % param
            self.pbs += '  echo "POST-PROCESSING"\n'
            self.pbs += "  "
            self.checklogs()
            self.pbs += "  "
            self.postprocessing()
            self.pbs += "fi\n"

            #self.pbs += "wine get_mac_adresses" + '\n'
            # self.pbs += "cp -R ./*.mac  $PBS_O_WORKDIR/." + '\n'
            # -----------------------------------------------------------------

            # and we end when the cpu's per node are full
            if int(count1/self.maxcpu) == 1:
                # write the end part of the pbs script
                self.ending(pbs_path)
                ended = True
                # print progress:
                replace = ((i/self.maxcpu), (i_tot/self.maxcpu), self.walltime)
                if not self.silent:
                    print('pbs script %3i/%i walltime=%s' % replace)

            count2 += 1
            i += 1
            # the next cpu
            count1 += 1

        # it could be that the last node was not fully loaded. In that case
        # we do not have had a succesfull ending, and we still need to finish
        if not ended:
            # write the end part of the pbs script
            self.ending(pbs_path)
            # progress printing
            replace = ( (i/self.maxcpu), (i_tot/self.maxcpu), self.walltime )
            if not self.silent:
                print('pbs script %3i/%i walltime=%s, partially loaded' % replace)
#            print 'pbs progress, script '+format(i/self.maxcpu,'2.0f')\
#                + '/' + format(i_tot/self.maxcpu, '2.0f') \
#                + ' partially loaded...'

    def starting(self, tag_dict, jobid):
        """
        First part of the pbs script
        """

        # a new clean pbs script!
        self.pbs = ''
        self.pbs += "### Standard Output" + ' \n'

        case_id = tag_dict['[case_id]']

        # PBS job name
        self.pbs += "#PBS -N %s \n" % (jobid)
        self.pbs += "#PBS -o ./" + self.pbs_out_dir + case_id + ".out" + '\n'
        # self.pbs += "#PBS -o ./pbs_out/" + jobid + ".out" + '\n'
        self.pbs += "### Standard Error" + ' \n'
        self.pbs += "#PBS -e ./" + self.pbs_out_dir + case_id + ".err" + '\n'
        # self.pbs += "#PBS -e ./pbs_out/" + jobid + ".err" + '\n'
        self.pbs += '#PBS -W umask=0003\n'
        self.pbs += "### Maximum wallclock time format HOURS:MINUTES:SECONDS\n"
#        self.pbs += "#PBS -l walltime=" + self.walltime + '\n'
        self.pbs += "#PBS -l walltime=[walltime]\n"
        if self.qsub == 'time':
            self.pbs += "#PBS -a [start_time]" + '\n'
        elif self.qsub == 'depend':
            # set job dependencies, job_id refers to PBS job_id, which is only
            # assigned to a job at the moment it gets qsubbed into the que
            self.pbs += "[nodeps]PBS -W depend=afterany:[job_id]\n"

#        if self.que_jobdeps:
#            self.pbs += "#PBS -W depend=afterany:%s\n" % jobid_dep
#        else:
#            self.pbs += "#PBS -a [start_time]" + '\n'

        # in case of gorm, we need to make it work correctly. Now each job
        # has a different scratch dir. If we set maxcpu to 12 they all have
        # the same scratch dir. In that case there should be done something
        # differently

        # specify the number of nodes and cpu's per node required
        if self.maxcpu > 1:
            # Number of nodes and cpus per node (ppn)
            lnodes = int(math.ceil(len(self.cases)/float(self.maxcpu)))
            lnodes = 1
            self.pbs += "#PBS -l nodes=%i:ppn=%i\n" % (lnodes, self.maxcpu)
        else:
            self.pbs += "#PBS -l nodes=1:ppn=1\n"
            # Number of nodes and cpus per node (ppn)

        self.pbs += "### Queue name" + '\n'
        # queue names for Thyra are as follows:
        # short walltime queue (shorter than an hour): '#PBS -q xpresq'
        # or otherwise for longer jobs: '#PBS -q workq'
        self.pbs += self.pbs_queue_command + '\n'
        self.pbs += '\n' + '# ' + '='*78 + '\n'

        # ignore all the file copying when running in xargs mode:
        # when varibale LAUNCH_PBS_MODE is set, file copying is ignored
        # and will have to be done elsewhere
        # we do this so the same launch script can be used either with the node
        # scheduler and the PBS system (for example when re-running cases)
        # evaluates to true if LAUNCH_PBS_MODE is NOT set
        self.pbs += "if [ -z ${LAUNCH_PBS_MODE+x} ] ; then\n"

        self.pbs += "  ### Create scratch directory and copy data to it \n"
        # output the current directory
        self.pbs += "  cd $PBS_O_WORKDIR" + '\n'
        self.pbs += '  echo "current working dir (pwd):"\n'
        self.pbs += "  pwd \n"
        # The batch system on Gorm allows more than one job per node.
        # Because of this the scratch directory name includes both the
        # user name and the job ID, that is /scratch/$USER/$PBS_JOBID
        # if not scratch, make the dir
        if self.node_run_root != '/scratch':
            self.pbs += '  mkdir -p %s/$USER\n' % self.node_run_root
            self.pbs += '  mkdir -p %s/$USER/$PBS_JOBID\n' % self.node_run_root

        # copy the zip files to the scratch dir on the node
        self.pbs += "  cp -R ./" + self.ModelZipFile + \
            ' %s/$USER/$PBS_JOBID\n' % (self.node_run_root)
        self.pbs += "fi\n"
        self.pbs += '# ' + '-'*78 + '\n'

    def ending(self, pbs_path):
        """
        Last part of the pbs script, including command to write script to disc
        COPY BACK: from node to
        """
        self.pbs += "\n\n"
        self.pbs += '# ' + "="*78 + "\n"
        self.pbs += "### Epilogue\n"
        # evaluates to true if LAUNCH_PBS_MODE is NOT set
        self.pbs += '# evaluates to true if LAUNCH_PBS_MODE is NOT set\n'
        self.pbs += "if [ -z ${LAUNCH_PBS_MODE+x} ] ; then\n"
        self.pbs += "  ### wait for jobs to finish \n"
        self.pbs += "  wait\n"
        self.pbs += '  echo ""\n'
        self.pbs += '# ' + '-'*78 + '\n'
        self.pbs += '  echo "Copy back from scratch directory" \n'
        for i in range(1, self.maxcpu+1, 1):

            # navigate to the cpu dir on the node
            # The batch system on Gorm allows more than one job per node.
            # Because of this the scratch directory name includes both the
            # user name and the job ID, that is /scratch/$USER/$PBS_JOBID
            self.copyback_all_files("pbs_mode", i)
            # find+xargs mode only makes sense when maxcpu==1, cpu handling
            # for this mode is handled elsewhere
            if self.maxcpu == 1:
                self.pbs += 'else\n'
                self.copyback_all_files("find+xargs", None)

#            # and delete it all (but that is not allowed)
#            self.pbs += 'cd ..\n'
#            self.pbs += 'ls -lah\n'
#            self.pbs += 'echo $PBS_JOBID\n'
#            self.pbs += 'rm -r $PBS_JOBID \n'

            # Delete the batch file at the end. However, is this possible since
            # the batch file is still open at this point????
            # self.pbs += "rm "

        # end of PBS/find+xargs mode switching if/else
        self.pbs += 'fi\n'

        # base walltime on the longest simulation in the batch
        nr_time_steps = max(self.nr_time_steps)
        # TODO: take into acccount the difference between time steps with
        # and without output. This penelaty also depends on the number of
        # channels outputted. So from 0 until t0 we have no penalty,
        # from t0 until t0+duration we have the output penalty.

        # always a predifined lead time to account for startup losses
        tmax = int(nr_time_steps*self.secperiter*self.iterperstep + self.tlead)
        if self.dyn_walltime:
            dt_seconds = datetime.datetime.fromtimestamp(tmax)
            self.walltime = dt_seconds.strftime('%H:%M:%S')
            self.pbs = self.pbs.replace('[walltime]', self.walltime)
        else:
            self.pbs = self.pbs.replace('[walltime]', self.walltime)
        # and reset the nr_time_steps list for the next pbs job file
        self.nr_time_steps = []
        self.t0 = []
        self.duration = []

        # TODO: add logfile checking support directly here. In that way each
        # node will do the logfile checking and statistics calculations right
        # after the simulation. Figure out a way how to merge the data from
        # all the different cases afterwards

        self.pbs += "exit\n"

        if self.verbose:
            print('writing pbs script to path: ' + pbs_path)

        # and write the script to a file:
        write_file(pbs_path, self.pbs, 'w')
        # make the string empty again, for memory
        self.pbs = ''

    def copyback_all_files(self, mode, cpu_nr):
        """Copy back all the files from either scratch to run_dir (PBS mode),
        or from CPU sub-directory back to main directory in find+xargs mode.
        """
        if mode=="find+xargs":
            foper = "rsync -a --remove-source-files" # move files instead of copy
            dst = os.path.join('..', self.sim_id, '')
            dst_db = '../'
            cd2model = "  cd %s\n" % os.path.join(self.node_run_root, '$USER',
                                                  '$PBS_JOBID', '$CPU_NR', '')
            pbs_mode = False
        else:
            foper = "cp -R"
            dst = "$PBS_O_WORKDIR/"
            dst_db = dst
            pbs_mode = True
            cd2model = "  cd %s\n" % os.path.join(self.node_run_root, '$USER',
                                                  '$PBS_JOBID', '%i' % cpu_nr, '')

        # navigate to the cpu dir on the node
        # The batch system on Gorm/Jess allows more than one job per node.
        # Because of this the scratch directory name includes both the
        # user name and the job ID, that is /scratch/$USER/$PBS_JOBID/CPU_NR
        self.pbs += cd2model

        # create the log, res etc dirs in case they do not exist. Only relevant
        # for pbs_mode, they are created in advance in find+xargs
        if pbs_mode:
            mk = '  mkdir -p'
            self.pbs += "%s %s\n" % (mk, os.path.join(dst, self.results_dir))
            self.pbs += "%s %s\n" % (mk, os.path.join(dst, self.logs_dir))
            if self.animation_dir:
                self.pbs += "%s %s\n" % (mk, os.path.join(dst, self.animation_dir))
            if self.copyback_turb and self.TurbDb:
                self.pbs += "%s %s\n" % (mk, os.path.join(dst, self.TurbDb))
            elif self.copyback_turb:
                self.pbs += "%s %s\n" % (mk, os.path.join(dst, self.TurbDirName))
            if self.copyback_turb and self.wakeDb:
                self.pbs += "%s %s\n" % (mk, os.path.join(dst, self.wakeDb))
            elif self.WakeDirName and self.WakeDirName != self.TurbDirName:
                self.pbs += "%s %s\n" % (mk, os.path.join(dst, self.WakeDirName))
            if self.copyback_turb and self.meandDb:
                self.pbs += "%s %s\n" % (mk, os.path.join(dst, self.meandDb))
            elif self.MeanderDirName and self.MeanderDirName != self.TurbDirName:
                self.pbs += "%s %s\n" % (mk, os.path.join(dst, self.MeanderDirName))

        # and copy the results and log files frome the scratch to dst
        res_dst = os.path.join(dst, self.results_dir, ".")
        self.pbs += "  %s %s. %s\n" % (foper, self.results_dir, res_dst)
        log_dst = os.path.join(dst, self.logs_dir, ".")
        self.pbs += "  %s %s. %s\n" % (foper, self.logs_dir, log_dst)
        if self.animation_dir:
            ani_dst = os.path.join(dst, self.animation_dir, ".")
            self.pbs += "  %s %s. %s\n" % (foper, self.animation_dir, ani_dst)

        if self.eigenfreq_dir:
            # just in case the eig dir has subdirs for the results, only
            # select the base path and cp -r will take care of the rest
            p1 = self.eigenfreq_dir.split('/')[0]
            p2 = os.path.join(dst, p1, ".")
            self.pbs += "  cp -R %s/. %s \n" % (p1, p2)
            # for eigen analysis with floater, modes are in root
            eig_dir_sys = os.path.join(dst, self.eigenfreq_dir, 'system/', '.')
            self.pbs += '  mkdir -p %s \n' % eig_dir_sys
            self.pbs += "  cp -R mode* %s \n" % eig_dir_sys
            self.pbs += "  %s mode* %s \n" % (foper, eig_dir_sys)

        # only copy the turbulence files back if they do not exist
        # for all *.bin files on the node
        cmd = '  for i in `ls *.bin`; do  if [ -e %s$i ]; '
        cmd += 'then echo "$i exists no copyback"; else echo "$i copyback"; '
        cmd += 'cp $i %s; fi; done\n'
        # copy back turbulence file?
        # browse to the node turb dir
        self.pbs += '\n  echo ""\n'
        self.pbs += '  echo "COPY BACK TURB IF APPLICABLE"\n'
        if self.copyback_turb and self.TurbDb:
            self.pbs += '  cd %s\n' % self.TurbDirName
            tmp = (os.path.join(dst_db, self.TurbDb, ''),)*2
            self.pbs += cmd % tmp
            # and back to normal model root
            self.pbs += cd2model
        elif self.copyback_turb:
            self.pbs += '  cd %s\n' % self.TurbDirName
            tmp = (os.path.join(dst, self.TurbDirName, ''),)*2
            self.pbs += cmd % tmp
            # and back to normal model root
            self.pbs += cd2model

        if self.copyback_turb and self.wakeDb:
            self.pbs += '  cd %s\n' % self.WakeDirName
            tmp = (os.path.join(dst_db, self.wakeDb, ''),)*2
            self.pbs += cmd % tmp
            # and back to normal model root
            self.pbs += cd2model
        elif self.copyback_turb and self.WakeDirName:
            self.pbs += '  cd %s\n' % self.WakeDirName
            tmp = (os.path.join(dst, self.WakeDirName, ''),)*2
            self.pbs += cmd % tmp
            # and back to normal model root
            self.pbs += cd2model

        if self.copyback_turb and self.meandDb:
            self.pbs += '  cd %s\n' % self.MeanderDirName
            tmp = (os.path.join(dst_db, self.meandDb, ''),)*2
            self.pbs += cmd % tmp
            # and back to normal model root
            self.pbs += cd2model
        elif self.copyback_turb and self.MeanderDirName:
            self.pbs += '  cd %s\n' % self.MeanderDirName
            tmp = (os.path.join(dst, self.MeanderDirName, ''),)*2
            self.pbs += cmd % tmp
            # and back to normal model root
            self.pbs += cd2model

        self.pbs += '  echo "END COPY BACK TURB"\n'
        self.pbs += '  echo ""\n\n'

        # copy back any other kind of files, as specified in copyback_files
        self.pbs += '  echo "COPYBACK [copyback_files]/[copyback_frename]"\n'
        if len(self.copyback_frename) == 0:
            self.copyback_frename = self.copyback_files
        for fname, fnew in zip(self.copyback_files, self.copyback_frename):
            dst_fnew = os.path.join(dst, fnew)
            self.pbs += "  %s %s %s \n" % (foper, fname, dst_fnew)
        self.pbs += '  echo "END COPYBACK"\n'
        self.pbs += '  echo ""\n\n'

        if pbs_mode:
            # check what is left
            self.pbs += '  echo ""\n'
            self.pbs += '  echo "following files are on '
            self.pbs += 'node/cpu %i (find .):"\n' % cpu_nr
            self.pbs += '  find .\n'
        self.pbs += '# ' + '-'*78 + '\n'

    def checklogs(self):
        """
        """
        self.pbs += 'python -c "from wetb.prepost import statsdel; '
        rpl = (os.path.join(self.logs_dir, self.case+'.log'))
        self.pbs += 'statsdel.logcheck(\'%s\')"\n' % rpl

    def postprocessing(self):
        """Run post-processing just after HAWC2 has ran
        """
        self.pbs += 'python -c "from wetb.prepost import statsdel; '
        fsrc = os.path.join(self.results_dir, self.case)
        rpl = (fsrc, str(self.case_duration), '.csv')
        self.pbs += ('statsdel.calc(\'%s\', no_bins=46, m=[3, 4, 6, 8, 10, 12], '
                     'neq=%s, i0=0, i1=None, ftype=\'%s\')"\n' % rpl)

    def check_results(self, cases):
        """
        Cross-check if all simulations on the list have returned a simulation.
        Combine with ErrorLogs to identify which errors occur where.
        """

        cases_fail = {}

        if not self.silent:
            print('checking if all log and result files are present...', end='')

        # check for each case if we have results and a log file
        for cname, case in cases.items():
            run_dir = case['[run_dir]']
            res_dir = case['[res_dir]']
            log_dir = case['[log_dir]']
            cname_ = cname.replace('.htc', '')
            f_log = os.path.join(run_dir, log_dir, cname_)
            f_res = os.path.join(run_dir, res_dir, cname_)
            if not os.path.exists(f_log + '.log'):
                cases_fail[cname] = copy.copy(cases[cname])
                continue
            try:
                size_sel = os.stat(f_res + '.sel').st_size
                size_dat = os.stat(f_res + '.dat').st_size
            except OSError:
                size_sel = 0
                size_dat = 0
            if size_sel < 5 or size_dat < 5:
                cases_fail[cname] = copy.copy(cases[cname])

        if not self.silent:
            print('done!')

        # length will be zero if there are no failures
        return cases_fail


# TODO: rewrite the error log analysis to something better. Take different
# approach: start from the case and see if the results are present. Than we
# also have the tags_dict available when log-checking a certain case
class ErrorLogs(windIO.LogFile):
    """
    Analyse all HAWC2 log files in any given directory
    ==================================================

    Usage:
    logs = ErrorLogs()
    logs.MsgList    : list with the to be checked messages. Add more if required
    logs.ResultFile : name of the result file (default is ErrorLog.csv)
    logs.PathToLogs : specify the directory where the logsfile reside,
                        the ResultFile will be saved in the same directory.
                        It is also possible to give the path of a specific
                        file, the logfile will not be saved in this case. Save
                        when all required messages are analysed with save()
    logs.check() to analyse all the logfiles and create the ResultFile
    logs.save() to save after single file analysis

    logs.MsgListLog : [ [case, line nr, error1, line nr, error2, ....], [], ...]
    holding the error messages, empty if no err msg found
    will survive as long as the logs object exists. Keep in
    mind that when processing many messages with many error types (as defined)
    in MsgList might lead to an increase in memory usage.

    logs.MsgListLog2 : dict(key=case, value=[found_error, exit_correct]
        where found_error and exit_correct are booleans. Found error will just
        indicate whether or not any error message has been found

    All files in the speficied folder (PathToLogs) will be evaluated.
    When Any item present in MsgList occurs, the line number of the first
    occurance will be displayed in the ResultFile.
    If more messages are required, add them to the MsgList
    """

    # TODO: move to the HAWC2 plugin for cases

    def __init__(self, silent=False, cases=None, resultfile='ErrorLog.csv'):

        # call init from base class
        super(ErrorLogs, self).__init__()

        self.PathToLogs = ''
        self.ResultFile = resultfile
        self.cases = cases
        self.silent = silent

    # TODO: save this not a csv text string but a df_dict, and save as excel
    # and DataFrame!
    def check(self, appendlog=False, save_iter=False):
        """Check all log files that are to be found in the directory
        ErrorLogs.PathToLogs, or check the specific log file if
        ErrorLogs.PathToLogs points to a specific log file.
        """

        # MsgListLog = []

        # load all the files in the given path
        FileList = []
        for files in os.walk(self.PathToLogs):
            FileList.append(files)

        # if the instead of a directory, a file path is given
        # the generated FileList will be empty!
        try:
            NrFiles = len(FileList[0][2])
        # input was a single file:
        except:
            NrFiles = 1
            # simulate one entry on FileList[0][2], give it the file name
            # and save the directory on in self.PathToLogs
            tmp = self.PathToLogs.split(os.path.sep)[-1]
            # cut out the file name from the directory
            self.PathToLogs = self.PathToLogs.replace(tmp, '')
            FileList.append([ [],[],[tmp] ])
            single_file = True
        i=1

        # walk trough the files present in the folder path
        for fname in FileList[0][2]:
            fname_lower = fname.lower()
            # progress indicator
            if NrFiles > 1:
                if not self.silent:
                    print('progress: ' + str(i) + '/' + str(NrFiles))

            # open the current log file
            f_log = os.path.join(self.PathToLogs, str(fname_lower))

            if self.cases is not None:
                case = self.cases[fname.replace('.log', '.htc')]
            else:
                case = None
            self.readlog(f_log, case=case, save_iter=save_iter)
            i += 1

#            # if no messages are found for the current file, than say so:
#            if len(MsgList2) == len(self.MsgList):
#                tempLog[-1] = 'NO MESSAGES FOUND'

        # if we have only one file, don't save the log file to disk. It is
        # expected that if we analyse many different single files, this will
        # cause a slower script
        if single_file:
            # now we make it available over the object to save and let it grow
            # over many analysis
            # self.MsgListLog = copy.copy(MsgListLog)
            pass
        else:
            self.save(appendlog=appendlog)

    def save(self, appendlog=False, suffix=None):

        contents = self._header()
        contents = self._msglistlog2csv(contents)

        # write csv file to disk, append to facilitate more logfile analysis
        if isinstance(suffix, str):
            tmp = self.ResultFile.replace('.csv', '_%s.csv' % suffix)
            fname = os.path.join(self.PathToLogs, tmp)
        else:
            fname = os.path.join(self.PathToLogs, str(self.ResultFile))
        if not self.silent:
            print('Error log analysis saved at:')
            print(fname)
        if appendlog:
            mode = 'a'
        else:
            mode = 'w'
        with open(fname, mode) as f:
            f.write(contents)


class ModelData(object):
    """
    Second generation ModelData function. The HawcPy version is crappy, buggy
    and not mutch of use in the optimisation context.
    """
    class st_headers(object):
        """
        Indices to the respective parameters in the HAWC2 st data file
        """
        r     = 0
        m     = 1
        x_cg  = 2
        y_cg  = 3
        ri_x  = 4
        ri_y  = 5
        x_sh  = 6
        y_sh  = 7
        E     = 8
        G     = 9
        Ixx   = 10
        Iyy   = 11
        I_p   = 12
        k_x   = 13
        k_y   = 14
        A     = 15
        pitch = 16
        x_e   = 17
        y_e   = 18

    def __init__(self, verbose=False, silent=False):
        self.verbose = verbose
        self.silent = silent
        # define the column width for printing
        self.col_width = 13
        # formatting and precision
        self.float_hi = 9999.9999
        self.float_lo =  0.01
        self.prec_float = ' 9.05f'
        self.prec_exp =   ' 8.04e'
        self.prec_loss = 0.01

        #0 1  2    3    4    5    6    7   8 9 10   11
        #r m x_cg y_cg ri_x ri_y x_sh y_sh E G I_x  I_y
        #12    13  14  15  16  17  18
        #I_p/K k_x k_y A pitch x_e y_e
        # 19 cols
        self.st_column_header_list = ['r', 'm', 'x_cg', 'y_cg', 'ri_x', \
            'ri_y', 'x_sh', 'y_sh', 'E', 'G', 'I_x', 'I_y', 'J', 'k_x', \
            'k_y', 'A', 'pitch', 'x_e', 'y_e']

        self.st_column_header_list_latex = ['r','m','x_{cg}','y_{cg}','ri_x',\
            'ri_y', 'x_{sh}','y_{sh}','E', 'G', 'I_x', 'I_y', 'J', 'k_x', \
            'k_y', 'A', 'pitch', 'x_e', 'y_e']

        # make the column header
        self.column_header_line = 19 * self.col_width * '=' + '\n'
        for k in self.st_column_header_list:
            self.column_header_line += k.rjust(self.col_width)
        self.column_header_line += '\n' + (19 * self.col_width * '=') + '\n'

    def fromline(self, line, separator=' '):
        # TODO: move this to the global function space (dav-general-module)
        """
        split a line, but ignore any blank spaces and return a list with only
        the values, not empty places
        """
        # remove all tabs, new lines, etc? (\t, \r, \n)
        line = line.replace('\t',' ').replace('\n','').replace('\r','')
        # trailing and ending spaces
        line = line.strip()
        line = line.split(separator)
        values = []
        for k in range(len(line)):
            if len(line[k]) > 0: #and k == item_nr:
                values.append(line[k])
                # break

        return values

    def load_st(self, file_path, file_name):
        """
        Now a better format: st_dict has following key/value pairs
            'nset'    : total number of sets in the file (int).
                        This should be autocalculated every time when writing
                        a new file.
            '007-000-0' : set number line in one peace
            '007-001-a' : comments for set-subset nr 07-01 (str)
            '007-001-b' : subset nr and number of data points, should be
                        autocalculate every time you generate a file
            '007-001-d' : data for set-subset nr 07-01 (ndarray(n,19))

        NOW WE ONLY CONSIDER SUBSET COMMENTS, SET COMMENTS, HOW ARE THEY
        TREADED NOW??

        st_dict is for easy remaking the same file. We need a different format
        for easy reading the comments as well. For that we have the st_comments
        """

        # TODO: store this in an HDF5 format! This is perfect for that.

        # read all the lines of the file into memory
        self.st_path, self.st_file = file_path, file_name
        FILE = open(os.path.join(file_path, file_name))
        lines = FILE.readlines()
        FILE.close()

        subset = False
        st_dict = dict()
        st_comments = dict()
        for i, line in enumerate(lines):

            # convert line to list space seperated list
            line_list = self.fromline(line)

            # see if the first character is marking something
            if i == 0:
                # it is possible that the NSET line is not defined
                parts = line.split(' ')
                try:
                    for k in range(10):
                        parts.remove(' ') # throws error when can't find
                except ValueError:
                    pass
                # we don't care what is on the nset line, just capture if
                # there are any comments lines
                set_nr = 0
                subset_nr = 0
                st_dict['000-000-0'] = line

            # marks the start of a set
            if line[0] == '#':
                #sett = True
                # first character is the #, the rest is the number
                set_nr = int(line_list[0][1:])
                st_dict['%03i-000-0' % set_nr] = line
                # and reset subset nr to zero now
                subset_nr = 0
                subset_nr_track = 0
                # and comments only format, back to one string
                st_comments['%03i-000-0' % set_nr] = ' '.join(line_list[1:])

            # marks the start of a subset
            elif line[0] == '$':
                subset_nr_track += 1
                subset = True
                subset_nr = int(line_list[0][1:])
                # and comments only format, back to one string
                setid = '%03i-%03i-b' % (set_nr, subset_nr)
                st_comments[setid] = ' '.join(line_list[2:])

                # check if the number read corresponds to tracking
                if subset_nr is not subset_nr_track:
                    msg = 'subset_nr and subset_nr_track do not match'
                    raise UserWarning(msg)

                nr_points = int(line_list[1])
                st_dict[setid] = line
                # prepare read data points
                sub_set_arr = scipy.zeros((nr_points,19), dtype=np.float64)
                # keep track of where we are on the data array, initialize
                # to 0 for starters
                point = 0

            # in case we are not in subset mode, we only have comments left
            elif not subset:
                # FIXME: how are we dealing with set comments now?
                # subset comments are coming before the actual subset
                # so we account them to one set later than we are now
                #if subset_nr > 0 :
                key = '%03i-%03i-a' % (set_nr, subset_nr+1)
                # in case it is not the first comment line
                if key in st_dict: st_dict[key] += line
                else: st_dict[key]  = line
                ## otherwise we have the set comments
                #else:
                    #key = '%03i-%03i-a' % (set_nr, subset_nr)
                    ## in case it is not the first comment line
                    #if st_dict.has_key(key): st_dict[key] += line
                    #else: st_dict[key]  = line

            # in case we have the data points, make sure there are enough
            # data poinst present, raise an error if it doesn't
            elif len(line_list)==19 and subset:
                # we can store it in the array
                sub_set_arr[point,:] = line_list
                # on the last entry:
                if point == nr_points-1:
                    # save to the dict:
                    st_dict['%03i-%03i-d' % (set_nr, subset_nr)]= sub_set_arr
                    # and indicate we're done subsetting, next we can have
                    # either set or subset comments
                    subset = False
                point += 1

            #else:
                #msg='error in st format: don't know where to put current line'
                #raise UserWarning, msg

        self.st_dict = st_dict
        self.st_comments = st_comments

    def _format_nr(self, number):
        """
        Automatic format the number

        prec_loss : float, default=0.01
            acceptible precision loss expressed in %

        """

        # the formatting of the number
        numabs = abs(number)
        # just a float precision defined in self.prec_float
        if (numabs < self.float_hi and numabs > self.float_lo):
            numfor = format(number, self.prec_float)