#!/usr/bin/env python3
# -*- coding: utf-8 -*-
'''
Script to run one simulation.
Example command to run an ensemble using 'jobrun' via the runner module:
jobrun ./runcx -r -- -o output/run -p ctl.nyears=5000
'''
import subprocess as subp 
import sys, os, socket, argparse, shutil, glob, datetime, json
from distutils.dir_util import copy_tree
from pandas import DataFrame

# Check whether runner library is installed or not:
try:
    from runner.ext.namelist import param_summary, param_write_to_files
    runner_is_installed = True
except:
    runner_is_installed = False

class DictAction(argparse.Action):
    '''Parse a list of parameters ['key=val',...] into a dict.
    Convert parameter values to appropriate type (str,float,int)
    Adapted from: https://sumit-ghosh.com/articles/parsing-dictionary-key-value-pairs-kwargs-argparse-python/
    '''
    def __call__(self, parser, namespace, values, option_string=None):
        setattr(namespace, self.dest, dict())
        for value in values:
            key, valstr = value.split('=')
            if ',' in valstr:
                raise Exception('Only one value allowed for each parameter using -p: {}={}'.format(key,valstr))
            try:
                value = float(valstr)
                if value % 1 == 0 and not '.' in valstr: value = int(value)
            except:
                value = valstr 
            getattr(namespace, self.dest)[key] = value

def main():
    '''Main subroutine to run one simulation.'''

    # Define name of standard info json configuration file 
    info_file = os.path.basename(__file__) + ".js"
    
    if os.path.isfile(info_file):
        # If file is found, load it up
        info = json.load(open(info_file))
    else: 
        # Copy file from config directory, inform user, 
        # and then load it. 
        # error_msg = ("Required json file '{}' containing run options not found. \n".format(info_file) +
        #              "This is probably the first time you are running this script. \n" +
        #              "The default file will be generated now. Check it is correct, \n" +
        #              "or copy your own file, and then run this script again.")
        # print(error_msg)

        info_file_default = os.path.join("config",info_file)
        shutil.copy(info_file_default,info_file)
        
        # Load info file 
        info = json.load(open(info_file))

        error_msg = ("Missing required json configuration file.\n" +
                     "Copied {} to {}.\n".format(info_file_default,info_file) +
                     "Confirm defaults are correct and try again.")
        raise Exception(error_msg)

    ## Get defaults for arguments ##
    default_jobname  = info['defaults']['jobname']
    default_email    = info['defaults']['email']
    default_group    = info['defaults']['group']
    default_omp      = info['defaults']['omp']
    default_wall     = info['defaults']['wall']
    default_qos      = info['defaults']['qos']
    default_part     = info['defaults']['partition'] 
    default_template = info['defaults']['job_template']

    # Substitute USER for actual username in email if found:
    default_email   = default_email.replace('USER',os.environ.get('USER'))
    
    # Make exe_aliases string for printing help
    exe_aliases_str = "; ".join(["{}={}".format(key,val) 
                        for key,val in info["exe_aliases"].items()])

    # Additional info 
    queues = info['job_queues'] 

    ### Manage command-line arguments ############################

    # Initialize argument parser
    parser = argparse.ArgumentParser()

    # Add options
    parser.add_argument('-e','--exe',type=str,default='climber',
        help="Define the executable file to use here. " \
             "Shortcuts: " + exe_aliases_str)
    parser.add_argument('-r','--run',action="store_true",
        help='Run the executable after preparing the job?')
    parser.add_argument('-s','--submit',action="store_true",
        help='Run the executable after preparing the job by submitting to the queue?')
    parser.add_argument('-q','--qos',type=str, default=default_qos,
        help='Name of the qos the job should be submitted to (priority,short,medium,long)')
    parser.add_argument('-w','--wall', type=int, default=default_wall,
        help='Maximum wall time [hrs] to allow for job (only for jobs submitted to queue)')
    parser.add_argument('--part',type=str, metavar="PARTITION", default=default_part,
        help='Name of the partition the job should be submitted to (e.g., haswell, broadwell)')
    parser.add_argument('--omp', type=int, default=default_omp,
        help='Specify number of threads for omp job submission (default = 0 implies no omp specification)')
    parser.add_argument('--email', type=str, default=default_email,
        help='Email address to send job notifications from cluster')
    parser.add_argument('--group', type=str, default=default_group,
        help='Email address to send job notifications from cluster')
    parser.add_argument('-x',action="store_true",
        help='This argument is used when the script called by jobrun')
    parser.add_argument('-v',action="store_true",help='Verbose script output?')

    parser.add_argument("-p",metavar="KEY=VALUE",nargs='+',action=DictAction,
                        help="Set a number of key-value pairs "
                             "(do not put spaces before or after the = sign). "
                             "If a value contains spaces, you should define "
                             "it with double quotes: "
                             'foo="this is a sentence". Note that '
                             "values are always treated as strings.")

    # Required argument:
    # Note: Given the `-p` argument above with multiple parameters possible,
    # `rundir` cannot be a positional argument. Thus it is specified as 
    # required with the `-o` prefix:
    requiredNamed = parser.add_argument_group('required named arguments')
    requiredNamed.add_argument('-o',dest='rundir',metavar='RUNDIR',type=str,required=True,
         help='Path where simulation will run and store output.')

    # Parse the arguments
    args = parser.parse_args()

    ### Manage user options and arguments ############################
    
    # Options
    exe_path    = args.exe       # Path relative to current working directory (cwd)
    run         = args.run 
    submit      = args.submit 
    qos         = args.qos  
    wall        = args.wall 
    partition   = args.part
    omp         = args.omp 
    email       = args.email 
    group       = args.group
    with_runner = args.x   
    verbose     = args.v 
    par         = args.p 
    rundir      = args.rundir 

    # Define some variables based on default values (not parameters for now)
    jobname                 = default_jobname
    path_jobscript_template = default_template 

    # Could maybe add this as a parameter choice:
    with_profiler = False 
    
    # Additional options, consistency checks

    if with_runner and not runner_is_installed:
        error_msg = "The Python module 'runner' is not installed, or not installed properly. Do not use option -x. " \
                    "This option should only be used by the runner::jobrun script internally. " \
                    "To install runner, see: https://github.com/alex-robinson/runner"
        raise Exception(error_msg)

    if not par is None and not runner_is_installed:
        error_msg = "The Python module 'runner' is not installed, or not installed properly. Do not use option -p. " \
                    "To install runner, see: https://github.com/alex-robinson/runner"
        raise Exception(error_msg)

    # Copy the executable file to the output directory, or
    # call it from its compiled location?    
    copy_exec = True 

    # If argument 'submit' is true, then ensure 'run' is also true
    if submit: run = True 

    # Expand executable path alias if defined, otherwise exe_path remains unchanged.
    if exe_path in info["exe_aliases"]:
        exe_path = info["exe_aliases"].get(exe_path)
    
    # Also extract executable filename 
    exe_fname = os.path.basename(exe_path)

    # Executable assumed to take output directory as command line argument
    # Define argument depending on whether running from rundir (copy_exec==True)
    # or from central directory. 
    if copy_exec: 
        exe_args = "./" 
    else:
        exe_args = rundir 

    # Define prefix part of command for running with a profiler (eg vtune)
    if with_profiler:
        # Use prefix for vtune profiler 
        if copy_exec:
            profiler_prefix = "amplxe-cl -c hotspots -r {} -- ".format("./")
        else:
            profiler_prefix = "amplxe-cl -c hotspots -r {} -- ".format(rundir)
    else: 
        profiler_prefix = ""
    
    # Make sure input file(s) exist 
    if not os.path.isfile(exe_path):
        print("Input file does not exist: {}".format(exe_path))
        sys.exit() 
    

    ### Start the script to make the job, and then run it ############################


    ##########################################
    # 1. Make the job (output directory, 
    #    parameter files, additional 
    #    data files/links)
    ##########################################

    # Using `jobrun` (runner), the rundir has already been created,
    # if not, it needs to be created here (removing existing nc and nml files): 
    if not with_runner: 
        makedirs(rundir,remove=True)

    # Copy files as needed 
    copy_files(info["files"],rundir) 

    # Copy special dir that need a different destination 
    for src, dst in info["dir-special"].items():
        copy_dir(src,rundir,dst)

    ## Generate symbolic links to input data folders
    for link in info["links"]:
        make_link(link,rundir)

    # Copy exe file to rundir (even if running from cwd)
    shutil.copy(exe_path,rundir)

    ### PARAMETER FILES ###

    # Get list of parameter files to manage
    par_paths = list(info["par_paths"].values())

    # Copy the default parameter files to the rundir 
    copy_files(par_paths,rundir)
    
    # Get list of new parameter file destinations 
    par_paths_rundir = []
    for par_path in par_paths:
        par_path_now = os.path.join(rundir,os.path.basename(par_path))
        par_paths_rundir.append(par_path_now) 
      
    # First modify parameters according to command-line values
    # "-p key=val key=val ..." argument.
    if not par is None:
        
        # Write parameters to files (in rundir!)
        param_write_to_files(par,par_paths_rundir,par_paths_rundir,info["grp_aliases"])
        
    # Next modify parameters according to command-line values specified
    # by runner (via `jobrun` "-p key=val [key=val ...]" argument)
    if with_runner:

        # Read param file runner.json always written by runner to rundir
        f  = os.path.join(rundir, 'runner.json')
        js = json.load(open(f))
        par_runner = js['params'] 

        # Summarize what is to be done
        param_summary(par_runner,rundir,verbose) 

        # Write parameters to files (in rundir!)
        param_write_to_files(par_runner,par_paths_rundir,par_paths_rundir,info["grp_aliases"])
        
    ### DONE WITH PARAMETER FILES ###

    ## Write summary file to rundir 

    run_info = {}
    run_info['Command'] = " ".join(sys.argv)
    run_info['Called via runner'] = with_runner 
    
    # Write the current git revision information to output directory 
    if os.path.isdir(".git"):
        run_info['git hash'] = get_git_revision_hash()
    else:
        run_info['git hash'] = "Not under git version control."
    
    with open(os.path.join(rundir,"run_info.txt"), 'w') as file:
        for key, val in run_info.items():
            file.write("{} : {}\n".format(key,val))

    ##########################################
    # 2. Run the job
    ##########################################

    # Generate the appropriate executable command to run job
    if copy_exec:
        # Assume executable is running from rundir
        executable = "{}./{} {}".format(profiler_prefix,exe_fname,exe_args)

    else:
        # Assume executable will run from current working directory 
        cwd = os.getcwd()
        executable = "{}{}/{} {}".format(profiler_prefix,cwd,exe_path,exe_args)

    # Run the job if desired 
    if run:

        if submit:
            # Submit job to queue 
            pid = submitjob(path_jobscript_template,rundir,executable,qos,
                            wall,partition,group,email,omp,jobname,queues) 

        else:
            # Run job in background 
            pid = runjob(rundir,executable,omp)

    return 


######### Helper functions ############### 


def runjob(rundir,cmd,omp):
    '''Run a job generated with makejob.'''

    if omp > 0:
        env_cmd = "&& export OMP_NUM_THREADS={}".format(omp) # && export OMP_THREAD_LIMIT=2"
    else:
        env_cmd = ""
    
    cmd_job = "cd {} {} && exec {} > {} &".format(rundir,env_cmd,cmd,"out.out")
    #cmd_job = "cd {} {} && mpiexec -n 2 {} > {} &".format(rundir,env_cmd,executable,"out.out")
    
    print("Running job in background: {}".format(cmd_job))

    # Run the command (ie, change to output directory and submit job)
    # Note: the argument `shell=True` can be a security hazard, but should
    # be ok in this context, see https://docs.python.org/2/library/subprocess.html#frequently-used-arguments
    #jobstatus = subp.check_call(cmd_job,shell=True)
    try:
        jobstatus = subp.check_output(cmd_job,shell=True,stdin=None,stderr=None)
    except subp.CalledProcessError as error:
        print(error)
        sys.exit() 
    
    return jobstatus

def submitjob(path_template,rundir,cmd,qos,wall,partition,group,email,omp,jobname,queues=None):
    '''Submit a job to a HPC queue (qsub,sbatch)'''

    # Get info about current system
    username  = os.environ.get('USER')
    #hostname  = os.environ.get('HOSTNAME')
    hostname  = socket.getfqdn()             # Returns full domain name

    # Create the jobscript using current info
    nm_jobscript   = 'job.submit'
    path_jobscript = "{}/{}".format(rundir,nm_jobscript)
    
    if "brigit" in hostname:
        # Host is the UCM brigit cluster, use the following submit script
        script  = generate_jobscript(path_template,cmd,jobname,group,email,qos,wall,partition,omp,queues)
        jobfile = open(path_jobscript,'w').write(script)
        cmd_job = "cd {} && sbatch {}".format(rundir,nm_jobscript)
    
    else:
        # Host is the PIK 2015 cluster, use the following submit script
        script  = generate_jobscript(path_template,cmd,jobname,group,email,qos,wall,partition,omp,queues)
        jobfile = open(path_jobscript,'w').write(script)
        cmd_job = "cd {} && sbatch {}".format(rundir,nm_jobscript)
    
    # Run the command (ie, change to output directory and submit job)
    # Note: the argument `shell=True` can be a security hazard, but should
    # be ok in this context, see https://docs.python.org/2/library/subprocess.html#frequently-used-arguments
    try:
        out = subp.check_output(cmd_job,shell=True,stderr=subp.STDOUT)
        jobstatus = out.decode("utf-8").strip() 
        print(jobstatus)
    except subp.CalledProcessError as error:
        print(error)
        sys.exit() 

    return jobstatus

def make_link(srcname,rundir,target=None):
    '''Make a link in the output dir.'''

    if target is None: target = srcname 

    # Define destination in rundir
    dstname = os.path.join(rundir,target)
    if os.path.islink(dstname): os.unlink(dstname)

    # Make link from srcname to dstname
    if os.path.islink(srcname):
        linkto = os.readlink(srcname)
        os.symlink(linkto, dstname)
    elif os.path.isdir(srcname):
        srcpath = os.path.abspath(srcname)
        os.symlink(srcpath,dstname)
    else:
        print("Warning: path does not exist {}".format(srcname))

    return

def copy_dir(path,rundir,target):
    '''Copy file(s) to run directory.'''

    dst = os.path.join(rundir,target)
    #shutil.copytree(path,dst,dirs_exist_ok=True)  # Only v3.8+
    copy_tree(path, dst)

    return 

def copy_file(path,rundir,target):
    '''Copy file(s) to run directory.'''

    dst = os.path.join(rundir,target)
    shutil.copy(path,dst)

    return 

def copy_files(paths,rundir):
    '''Bulk copy file(s) to run directory.'''

    for pnow in paths:
        shutil.copy(pnow,rundir)

    return 

def makedirs(dirname,remove):
    '''
    Make a directory (including sub-directories),
    but first ensuring that path doesn't already exist
    or some other error prevents the creation.
    '''

    try:
        os.makedirs(dirname)
        print('Directory created: {}'.format(dirname))
    except OSError:
        if os.path.isdir(dirname):
            print('Directory already exists: {}'.format(dirname))
            if remove:
                for f in glob.glob("{}/*.nc".format(dirname)): 
                    os.remove(f)
                for f in glob.glob("{}/*.nml".format(dirname)):
                    os.remove(f)
                #print('Removed *.nml and *.nc files.')
            pass
        else:
            # There was an error on creation, so make sure we know about it
            raise

    return

def autofolder(params,outfldr0):
    '''Given a list of parameters,
       generate an appropriate folder name.
    '''

    parts = []

    for p in params:
        parts.append( p.short() )

    # Join the parts together, combine with the base output dir
    autofldr = '.'.join(parts)
    outfldr  = outfldr0 + autofldr + '/'

    return outfldr

def get_git_revision_hash():
    #githash = subp.check_output(['git', 'describe', '--always', '--long', 'HEAD']).strip()
    githash = subp.check_output(['git', 'rev-parse', 'HEAD']).strip()
    return githash.decode("ascii") 

def generate_jobscript(template,cmd,jobname,group,email,qos,wall,partition,omp,queues=None):
    '''Definition of the job script based on a template file and 
    fields in < > that can be substituted.'''

# Extra parameter options to consider:
##SBATCH --partition=ram_gpu
##SBATCH --mem=50000 
    
    # Check queue information if available
    if not queues == None: 

        # First ensure desired queue is available 
        if not qos in queues:
            error_msg = "Error generating jobscript. Desired queue not in available list. \n" \
                        "{} not in {}".format(qos,"-".join(queues))
            raise Exception(error_msg)

        # Check that wall time was set properly
        if wall > queues[qos]['wall']:
            error_msg = "Error generating jobscript. Wall time too large for desired queue. \n" \
                        "wall={} greater than limit of {} for queue '{}'".format(wall,queues[qos]['wall'],qos)
            raise Exception(error_msg)

    # If omp has been set, generate a jobscript string with appropriate settings
    if omp > 0:
        # Use openmp settings 
        # Notes:
        #    --cpus-per-task=32 on broadwell, 16 on haswell
        
        # Read in omp section template
        omp_script = open(template+"_omp",'r').read()

        # Now make substitutions to match argument choices 
        omp_script = omp_script.replace('<OMP>',"{}".format(omp))
        omp_script = omp_script.replace('<PARTITION>',partition)
    
    else: 
        # No openmp 
        omp_script = "" 


    # Read in jobscript template 
    job_script = open(template,'r').read()

    # Substitute in omp section as needed to make full job script
    job_script = job_script.replace('<OMPSECTION>',omp_script)
    
    # Make additional substitutions to match argument choices 
    job_script = job_script.replace('<PARTITION>',partition)
    job_script = job_script.replace('<QOS>',qos)
    job_script = job_script.replace('<WALL>',"{}".format(wall))
    job_script = job_script.replace('<JOBNAME>',jobname)
    job_script = job_script.replace('<GROUP>',group)
    job_script = job_script.replace('<EMAIL>',email)
    job_script = job_script.replace('<CMD>',cmd)
    
    return job_script



if __name__ == "__main__": 

    # Call main function...
    main()



