populate_roi.py

#!/usr/bin/env python
# -*- coding: utf-8 -*-
"""
...
"""

#
#  Copyright (C) 2009 University of Dundee. All rights reserved.
#
#
#  This program is free software; you can redistribute it and/or modify
#  it under the terms of the GNU General Public License as published by
#  the Free Software Foundation; either version 2 of the License, or
#  (at your option) any later version.
#  This program is distributed in the hope that it will be useful,
#  but WITHOUT ANY WARRANTY; without even the implied warranty of
#  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
#  GNU General Public License for more details.
#
#  You should have received a copy of the GNU General Public License along
#  with this program; if not, write to the Free Software Foundation, Inc.,
#  51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
#


import tempfile
import logging
import time
import sys
import csv
import re
from threading import Thread
from getpass import getpass
from getopt import getopt, GetoptError
from queue import Queue

from omero.rtypes import rdouble, rstring, rint, unwrap
from omero.model import OriginalFileI, PlateI, PlateAnnotationLinkI, ImageI, \
    FileAnnotationI, RoiI, EllipseI, PointI
from omero.grid import ImageColumn, WellColumn, RoiColumn, LongColumn, \
    DoubleColumn

from omero.sys import ParametersI
from omero.util.temp_files import create_path
from omero import client

from xml.etree.cElementTree import ElementTree, iterparse

log = logging.getLogger("omero.util.populate_roi")


def usage(error):
    """Prints usage so that we don't have to. :)"""
    cmd = sys.argv[0]
    print("""%s
Usage: %s [-s hostname] [-u username | -k session_key] <-p port> [plate_id]
Runs measurement population code for a given plate.

Options:
  -s    OMERO hostname to use
  -p    OMERO port to use [defaults to 4064]
  -u    OMERO username to use
  -k    OMERO session key to use
  -m    Measurement index to populate
  -i    Dump measurement information and exit (no population)
  -d    Print debug statements
  -t    Number of threads to use when populating [defaults to 1]

Examples:
  %s -s localhost -p 4063 -u bob 27

Report bugs to ome-devel@lists.openmicroscopy.org.uk""" % (error, cmd, cmd))
    sys.exit(2)

###
# Worker and ThreadPool from...
# http://code.activestate.com/recipes/577187-python-thread-pool/
###


class Worker(Thread):

    """Thread executing tasks from a given tasks queue"""

    def __init__(self, tasks):
        Thread.__init__(self)
        self.tasks = tasks
        self.daemon = True
        self.start()

    def run(self):
        while True:
            func, args, kargs = self.tasks.get()
            try:
                func(*args, **kargs)
            except Exception as e:
                log.exception(e)
            self.tasks.task_done()


class ThreadPool(object):

    """Pool of threads consuming tasks from a queue"""

    def __init__(self, num_threads):
        self.tasks = Queue(num_threads)
        for _ in range(num_threads):
            Worker(self.tasks)

    def add_task(self, func, *args, **kargs):
        """Add a task to the queue"""
        self.tasks.put((func, args, kargs))

    def wait_completion(self):
        """Wait for completion of all the tasks in the queue"""
        self.tasks.join()

# Global thread pool for use by ROI workers
thread_pool = None


def get_thread_pool():
    global thread_pool
    if thread_pool is None:
        thread_pool = ThreadPool(1)
    return thread_pool


class MeasurementError(Exception):

    """
    Raised by the analysis or measurement context when an error condition
    is reached.
    """
    pass


class DownloadingOriginalFileProvider(object):

    """
    Provides original file data by downloading it from an OMERO raw file store.
    """

    # Default raw file store buffer size
    BUFFER_SIZE = 1024 * 1024  # 1MB

    def __init__(self, service_factory):
        self.service_factory = service_factory
        self.raw_file_store = self.service_factory.createRawFileStore()
        self.dir = create_path("populate_roi", "dir", folder=True)

    def get_original_file_data(self, original_file):
        """
        Downloads an original file to a temporary file and returns an open
        file handle to that temporary file seeked to zero. The caller is
        responsible for closing the temporary file.
        """
        log.info("Downloading original file: %d" % original_file.id.val)
        self.raw_file_store.setFileId(original_file.id.val)
        temporary_file = tempfile.NamedTemporaryFile(mode='rt+',
                                                     dir=str(self.dir))
        size = original_file.size.val
        for i in range((size // self.BUFFER_SIZE) + 1):
            index = i * self.BUFFER_SIZE
            data = self.raw_file_store.read(index, self.BUFFER_SIZE)
            temporary_file.write(data.decode("utf-8"))
        temporary_file.seek(0)
        temporary_file.truncate(size)
        return temporary_file

    def __delete__(self):
        self.raw_file_store.close()


class AbstractPlateAnalysisCtx(object):

    """
    Abstract class which aggregates and represents all measurement runs made on
    a given Plate.
    """

    DEFAULT_ORIGINAL_FILE_PROVIDER = DownloadingOriginalFileProvider

    def __init__(self, images, original_files, original_file_image_map,
                 plate_id, service_factory):
        super(AbstractPlateAnalysisCtx, self).__init__()
        self.images = images
        self.numcols, self.numrows = self.guess_geometry(self.images)
        self.original_files = original_files
        self.original_file_image_map = original_file_image_map
        self.plate_id = plate_id
        self.service_factory = service_factory
        self.log_files = dict()
        self.detail_files = dict()
        self.measurements = dict()

    def guess_geometry(self, images):
        max_col = 0
        max_row = 0
        for image in images:
            # Using only first well sample link
            ws = image.copyWellSamples()[0]
            well = ws.well
            max_col = max(max_col, well.column.val)
            max_row = max(max_row, well.row.val)
        return (max_col + 1, max_row + 1)

    def colrow_from_wellnumber(self, width, wellnumber):
        x = wellnumber - 1
        col = x % width
        row = x // width
        return (col, row)

    def image_from_wellnumber(self, wellnumber):
        col, row = self.colrow_from_wellnumber(self.numcols, wellnumber)
        log.debug("Finding image for %s (%s,%s)..." % (wellnumber, col, row))
        for image in self.images:
            well = image.copyWellSamples()[0].well
            if well.column.val == col and well.row.val == row:
                return image
        raise Exception(
            "Could not find image for (col,row)==(%s,%s)" % (col, row))

    ###
    # Abstract methods
    ###

    def is_this_type(klass):
        """
        Concrete implementations are to return True if the class pertinent
        for the original files associated with the plate.
        """
        raise Exception("To be implemented by concrete implementations.")
    is_this_type = classmethod(is_this_type)

    def get_measurement_count(self):
        """Returns the number of recognized measurement runs."""
        raise Exception("To be implemented by concrete implementations.")

    def get_measurement_ctx(self, index):
        """Returns the measurement context for a given index."""
        raise Exception("To be implemented by concrete implementations.")

    def get_result_file_count(self, measurement_index):
        """
        Return the number of result files associated with a measurement run.
        """
        raise Exception("To be implemented by concrete implementations.")


class MIASPlateAnalysisCtx(AbstractPlateAnalysisCtx):

    """
    MIAS dataset concrete class implementation of an analysis context. MIAS
    measurements are aggregated based on a single "log" file. A result
    file is present for each stitched (of multiple fields) mosaic and
    contains the actual measured results and ROI.
    """

    # Python datetime format string of the log filename completion date/time
    datetime_format = '%Y-%m-%d-%Hh%Mm%Ss'

    # Regular expression matching a log filename
    log_regex = re.compile(r'.*log(\d+-\d+-\d+-\d+h\d+m\d+s).txt$')

    # Regular expression matching a result filename
    detail_regex = re.compile(
        r'^Well(\d+)_(.*)_detail_(\d+-\d+-\d+-\d+h\d+m\d+s).txt$')

    # Companion file format
    companion_format = 'Companion/MIAS'

    def __init__(self, images, original_files, original_file_image_map,
                 plate_id, service_factory):
        super(MIASPlateAnalysisCtx, self).__init__(
            images, original_files, original_file_image_map, plate_id,
            service_factory)
        self._populate_log_and_detail_files()
        self._populate_measurements()

    def _populate_log_and_detail_files(self):
        """
        Strips out erroneous files and collects the log and result original
        files based on regular expression matching.
        """
        for original_file in self.original_files:
            if original_file.mimetype and \
                    original_file.mimetype.val != self.companion_format:
                # In OMERO5, the mimetype will not be set.
                continue
            name = original_file.name.val
            match = self.log_regex.match(name)
            if match:
                d = time.strptime(match.group(1), self.datetime_format)
                self.log_files[d] = original_file
                continue
            match = self.detail_regex.match(name)
            if match:
                d = time.strptime(match.group(3), self.datetime_format)
                self.detail_files[d] = original_file
                continue

    def _populate_measurements(self):
        """
        Result original files are only recognizable as part of a given
        measurement (declared by a log file) based upon their parsed
        date/time of completion as encoded in the filename. This method
        collects result original files and groups them by collective
        parsed date/time of completion.
        """
        log_timestamps = list(self.log_files.keys())
        log_timestamps.sort()
        detail_timestamps = list(self.detail_files.keys())
        detail_timestamps.sort()
        for log_timestamp in log_timestamps:
            self.measurements[log_timestamp] = list()
        for detail_timestamp in detail_timestamps:
            for log_timestamp in log_timestamps:
                if detail_timestamp < log_timestamp:
                    self.measurements[log_timestamp].append(
                        self.detail_files[detail_timestamp])
                    break

    ###
    # Abstract method implementations
    ###

    def is_this_type(klass, original_files):
        for original_file in original_files:
            format = unwrap(original_file.mimetype)
            # In OMERO5, "Companion/*" is unused.
            if (format is None or format == klass.companion_format) \
               and klass.log_regex.match(original_file.name.val):
                return True
    is_this_type = classmethod(is_this_type)

    def get_measurement_count(self):
        return len(list(self.measurements.keys()))

    def get_measurement_ctx(self, index):
        key = list(self.log_files.keys())[index]
        sf = self.service_factory
        original_file = self.log_files[key]
        result_files = self.measurements[key]
        provider = self.DEFAULT_ORIGINAL_FILE_PROVIDER(sf)
        return MIASMeasurementCtx(self, sf, provider, original_file,
                                  result_files)

    def get_result_file_count(self, measurement_index):
        key = list(self.log_files.keys())[measurement_index]
        return len(self.measurements[key])


class FlexPlateAnalysisCtx(AbstractPlateAnalysisCtx):

    """
    Flex dataset concrete class implementation of an analysis context. Flex
    measurements are aggregated in a single ".res" XML file and contain no
    ROI.
    """

    # Companion file format
    companion_format = 'Companion/Flex'

    def __init__(self, images, original_files, original_file_image_map,
                 plate_id, service_factory):
        super(FlexPlateAnalysisCtx, self).__init__(
            images, original_files, original_file_image_map, plate_id,
            service_factory)
        path_original_file_map = dict()
        for original_file in original_files:
            path = original_file.path.val
            name = original_file.name.val
            format = original_file.mimetype.val
            if format == self.companion_format and name.endswith('.res'):
                path_original_file_map[path] = original_file
        self.measurements = list(path_original_file_map.values())

    ###
    # Abstract method implementations
    ###

    def is_this_type(klass, original_files):
        for original_file in original_files:
            format = unwrap(original_file.mimetype)
            name = original_file.name.val
            if format == klass.companion_format and name.endswith('.res'):
                return True
        return False
    is_this_type = classmethod(is_this_type)

    def get_measurement_count(self):
        return len(self.measurements)

    def get_measurement_ctx(self, index):
        sf = self.service_factory
        original_file = self.measurements[index]
        result_files = []
        provider = self.DEFAULT_ORIGINAL_FILE_PROVIDER(sf)
        return FlexMeasurementCtx(self, sf, provider, original_file,
                                  result_files)

    def get_result_file_count(self, measurement_index):
        return 1


class InCellPlateAnalysisCtx(AbstractPlateAnalysisCtx):

    """
    InCell dataset concrete class implementation of an analysis context.
    InCell measurements are from InCell Analyzer and are aggregated in a
    single gargantuan (often larger than 100MB per plate) XML file.
    """

    # Companion file format
    companion_format = 'Companion/InCell'

    def __init__(self, images, original_files, original_file_image_map,
                 plate_id, service_factory):
        super(InCellPlateAnalysisCtx, self).__init__(
            images, original_files, original_file_image_map, plate_id,
            service_factory)
        path_original_file_map = dict()
        for original_file in original_files:
            path = original_file.path.val
            name = original_file.name.val
            format = original_file.mimetype.val
            if format == self.companion_format and name.endswith('.xml'):
                path_original_file_map[path] = original_file
        self.measurements = list(path_original_file_map.values())

    ###
    # Abstract method implementations
    ###

    def is_this_type(klass, original_files):
        for original_file in original_files:
            format = unwrap(original_file.mimetype)
            name = original_file.name.val
            if format == klass.companion_format and name.endswith('.xml'):
                return True
        return False
    is_this_type = classmethod(is_this_type)

    def get_measurement_count(self):
        return len(self.measurements)

    def get_measurement_ctx(self, index):
        sf = self.service_factory
        original_file = self.measurements[index]
        result_files = []
        provider = self.DEFAULT_ORIGINAL_FILE_PROVIDER(sf)
        return InCellMeasurementCtx(self, sf, provider, original_file,
                                    result_files)

    def get_result_file_count(self, measurement_index):
        return 1


class PlateAnalysisCtxFactory(object):

    """
    The plate analysis context factory is responsible for detecting and
    returning a plate analysis context instance for a given plate.
    """

    implementations = (FlexPlateAnalysisCtx, MIASPlateAnalysisCtx,
                       InCellPlateAnalysisCtx)

    def __init__(self, service_factory):
        self.service_factory = service_factory
        self.query_service = self.service_factory.getQueryService()

    def find_images_for_plate(self, plate_id):
        """
        Retrieves all the images associated with a given plate. Fetched
        are the Image's WellSample, the WellSample's Well, the annotation
        stack associated with the Image and each annotation's linked
        original file.
        """
        # The query that follows is doublely linked:
        #  * Image --> WellSample --> Well
        #  * Well --> WellSample --> Image
        # This is to facilitate later "ordered" access of fields/well
        # samples required by certain measurement contexts (notably InCell).
        log.debug("Loading image...")
        images = self.query_service.findAllByQuery(
            'select img from Image as img '
            'join fetch img.wellSamples as ws '
            'join fetch ws.well as w '
            'join fetch w.wellSamples as ws2 '
            'join w.plate as p '
            'left outer join fetch img.annotationLinks as ia_links '
            'left outer join fetch ia_links.child as ia '
            'left outer join fetch ia.file as i_o_file '
            'where p.id = %d' % plate_id, None)
        log.debug("Loading plate...")
        plate = self.query_service.findByQuery(
            'select p from Plate p '
            'left outer join fetch p.annotationLinks as pa_links '
            'left outer join fetch pa_links.child as pa '
            'left outer join fetch pa.file as p_o_file '
            'where p.id = %d' % plate_id, None)
        log.debug("Linking plate and images...")
        for image in images:
            for ws in image.copyWellSamples():
                ws.well.plate = plate
        return images, plate

    def gather_original_files(self, obj, original_files,
                              original_file_obj_map):
        for annotation_link in obj.copyAnnotationLinks():
            annotation = annotation_link.child
            if isinstance(annotation, FileAnnotationI):
                f = annotation.file
                original_files.add(f)
                if original_file_obj_map is not None:
                    original_file_obj_map[f.id.val] = obj

    # OMERO5 support

    def find_filesets_for_plate(self, plateid):
        """
        OMERO5 support. See #12235
        """
        return self.query_service.findAllByQuery((
            'select ofile from Fileset f '
            'join f.usedFiles as fse '
            'join fse.originalFile as ofile '
            'join f.images as i '
            'join i.wellSamples ws '
            'join ws.well w '
            'join w.plate p '
            'where p.id = :id'),
            ParametersI().addId(plateid))

    def get_analysis_ctx(self, plate_id):
        """Retrieves a plate analysis context for a given plate."""
        # Using a set since 1) no one was using the image.id key and 2)
        # we are now also collecting original files from plates (MIAS)
        # for which there's no clear key. Since all the files are loaded
        # in a single shot, double linking should not cause a problem.
        original_files = set()
        original_file_image_map = dict()
        images, plate = self.find_images_for_plate(plate_id)
        fileset = self.find_filesets_for_plate(plate_id)
        if fileset:
            original_files.update(fileset)
            self.gather_original_files(plate, original_files, None)
        else:
            plates = set()
            for image in images:
                for ws in image.copyWellSamples():
                    plate = ws.well.plate
                    if plate not in plates:
                        plates.add(plate)
                        self.gather_original_files(plate, original_files, None)
                self.gather_original_files(
                    image, original_files, original_file_image_map)
        for klass in self.implementations:
            if klass.is_this_type(original_files):
                return klass(images, original_files,
                             original_file_image_map,
                             plate_id, self.service_factory)
        raise MeasurementError(
            "Unable to find suitable analysis context for plate: %d" %
            plate_id)


class MeasurementParsingResult(object):

    """
    Holds the results of a measurement parsing event.
    """

    def __init__(self, sets_of_columns=None):
        if sets_of_columns is None:
            self.sets_of_columns = list()
        else:
            self.sets_of_columns = sets_of_columns

    def append_columns(self, columns):
        """Adds a set of columns to the parsing result."""
        self.sets_of_columns.append(columns)


class AbstractMeasurementCtx(object):

    """
    Abstract class which aggregates and represents all the results produced
    from a given measurement run. It also provides a scaffold for interacting
    with the OmeroTables infrastructure.
    """

    # The number of ROI to have parsed before streaming them to the server
    ROI_UPDATE_LIMIT = 1000

    def __init__(self, analysis_ctx, service_factory, original_file_provider,
                 original_file, result_files):
        super(AbstractMeasurementCtx, self).__init__()
        self.thread_pool = get_thread_pool()
        self.analysis_ctx = analysis_ctx
        self.service_factory = service_factory
        self.original_file_provider = original_file_provider
        self.query_service = self.service_factory.getQueryService()
        self.update_service = self.service_factory.getUpdateService()
        self.original_file = original_file
        self.result_files = result_files

        # Establish the rest of our initial state
        self.wellimages = dict()
        for image in self.analysis_ctx.images:
            for well_sample in image.copyWellSamples():
                well = well_sample.well
                idx = well.copyWellSamples().index(well_sample)
                row = well.row.val
                column = well.column.val
                if row not in self.wellimages:
                    self.wellimages[row] = dict()
                if column not in self.wellimages[row]:
                    self.wellimages[row][column] = []
                # Now we save the image at it's proper index
                l = self.wellimages[row][column]
                for x in range(idx - len(l) + 1):
                    l.append(None)
                l[idx] = image

    def get_well_images(self, row, col):
        """
        Takes a row and a col index and returns a tuple
        of Well and image. Either might be None. Uses the
        first image found to find the Well and therefore
        must be loaded (image->wellSample->well)
        """
        try:
            images = self.wellimages[row][col]
            if not images:
                return (None, None)
            image = images[0]
            well = image.copyWellSamples()[0].well
            return (well, images)
        except KeyError:
            # This has the potential to happen alot with the
            # datasets we have given the split machine acquisition
            # ".flex" file storage.
            log.warn("WARNING: Missing data for row %d column %d" %
                     (row, col))
            return (None, None)

    def update_table(self, columns):
        """Updates the OmeroTables instance backing our results."""
        # Create a new OMERO table to store our measurement results
        sr = self.service_factory.sharedResources()
        name = self.get_name()
        self.table = sr.newTable(1, '/%s.r5' % name)
        if self.table is None:
            raise MeasurementError(
                "Unable to create table: %s" % name)

        # Retrieve the original file corresponding to the table for the
        # measurement, link it to the file annotation representing the
        # umbrella measurement run, link the annotation to the plate from
        # which it belongs and save the file annotation.
        table_original_file = self.table.getOriginalFile()
        table_original_file_id = table_original_file.id.val
        log.info("Created new table: %d" % table_original_file_id)
        unloaded_o_file = OriginalFileI(table_original_file_id, False)
        self.file_annotation.file = unloaded_o_file
        unloaded_plate = PlateI(self.analysis_ctx.plate_id, False)
        plate_annotation_link = PlateAnnotationLinkI()
        plate_annotation_link.parent = unloaded_plate
        plate_annotation_link.child = self.file_annotation
        plate_annotation_link = \
            self.update_service.saveAndReturnObject(plate_annotation_link)
        self.file_annotation = plate_annotation_link.child

        t0 = int(time.time() * 1000)
        self.table.initialize(columns)
        log.debug("Table init took %sms" % (int(time.time() * 1000) - t0))
        t0 = int(time.time() * 1000)
        column_report = dict()
        for column in columns:
            column_report[column.name] = len(column.values)
        log.debug("Column report: %r" % column_report)
        self.table.addData(columns)
        self.table.close()
        log.info("Table update took %sms" % (int(time.time() * 1000) - t0))

    def create_file_annotation(self, set_of_columns):
        """
        Creates a file annotation to represent a set of columns from our
        measurment.
        """
        self.file_annotation = FileAnnotationI()
        self.file_annotation.ns = \
            rstring('openmicroscopy.org/omero/measurement')
        name = self.get_name(set_of_columns)
        self.file_annotation.description = rstring(name)

    def update_rois(self, rois, batches, batch_no):
        """
        Updates a set of ROI for a given batch updating the batches
        dictionary with the saved IDs.
        """
        log.debug("Saving %d ROI for batch %d" % (len(rois), batch_no))
        t0 = int(time.time() * 1000)
        roi_ids = self.update_service.saveAndReturnIds(rois)
        log.info("Batch %d ROI update took %sms" %
                 (batch_no, int(time.time() * 1000) - t0))
        batches[batch_no] = roi_ids

    def image_from_original_file(self, original_file):
        """Returns the image from which an original file has originated."""
        m = self.analysis_ctx.original_file_image_map
        return m[original_file.id.val]

    def parse_and_populate(self):
        """
        Calls parse and populate, updating the OmeroTables instance backing
        our results and the OMERO database itself.
        """
        result = self.parse()
        if result is None:
            return
        for i, columns in enumerate(result.sets_of_columns):
            self.create_file_annotation(i)
            self.parse_and_populate_roi(columns)
            self.populate(columns)

    ###
    # Abstract methods
    ###

    def get_name(self, set_of_columns=None):
        """Returns the name of the measurement, and a set of columns."""
        raise Exception("To be implemented by concrete implementations.")

    def parse(self):
        """Parses result files, returning a MeasurementParsingResult."""
        raise Exception("To be implemented by concrete implementations.")

    def parse_and_populate_roi(self, columns):
        """
        Parses and populates ROI from column data in the OMERO database.
        """
        raise Exception("To be implemented by concrete implementations.")

    def populate(self, columns):
        """
        Populates an OmeroTables instance backing our results and ROI
        linkages.
        """
        raise Exception("To be implemented by concrete implementations.")


class MIASMeasurementCtx(AbstractMeasurementCtx):

    """
    MIAS measurements are a set of tab delimited text files per well. Each
    TSV file's content is prefixed by the analysis parameters.
    """

    # The OmeroTable ImageColumn index
    IMAGE_COL = 0

    # The OmeroTable RoiColumn index
    ROI_COL = 1

    # Expected columns in NEO datasets
    NEO_EXPECTED = ('Image', 'ROI', 'Label', 'Row', 'Col', 'Nucleus Area',
                    'Cell Diam.', 'Cell Type', 'Mean Nucleus Intens.')

    # Expected columns in MNU datasets
    MNU_EXPECTED = ('Image', 'ROI', 'row', 'col', 'type')

    def __init__(self, analysis_ctx, service_factory, original_file_provider,
                 original_file, result_files):
        super(MIASMeasurementCtx, self).__init__(
            analysis_ctx, service_factory, original_file_provider,
            original_file, result_files)

    def get_empty_columns(self, n_columns):
        """
        Retrieves a set of empty OmeroTables columns for the analysis results
        prefixed by an ImageColumn and RoiColumn to handle these linked
        object indexes.
        """
        columns = [ImageColumn('Image', '', list()),
                   RoiColumn('ROI', '', list())]
        for i in range(n_columns):
            columns.append(DoubleColumn('', '', list()))
        return columns

    ###
    # Overriding abstract implementation
    ###

    def image_from_original_file(self, original_file):
        """
        Overriding the abstract implementation since the companion
        files are no longer attached to the images, but only to the plate
        for MIAS. Instead, we use the filename itself to find the image.
        """
        name = original_file.name.val
        # Copy: '^Well(\d+)_(.*)_detail_(\d+-\d+-\d+-\d+h\d+m\d+s).txt$'
        match = MIASPlateAnalysisCtx.detail_regex.match(name)
        if match:
            well_num = int(match.group(1))
            return self.analysis_ctx.image_from_wellnumber(well_num)
        else:
            raise Exception("Not a detail file")

    ###
    # Abstract method implementations
    ###

    def get_name(self, set_of_columns=None):
        return self.original_file.name.val[:-4]

    def parse(self):
        columns = None
        for result_file in self.result_files:
            log.info("Parsing: %s" % result_file.name.val)
            image = self.image_from_original_file(result_file)
            provider = self.original_file_provider
            data = provider.get_original_file_data(result_file)
            try:
                rows = list(csv.reader(data, delimiter='\t'))
            finally:
                data.close()
            rows.reverse()
            if columns is None:
                columns = self.get_empty_columns(len(rows[0]))
            for row in rows:
                try:
                    for i, value in enumerate(row):
                        value = float(value)
                        columns[i + 2].values.append(value)
                    columns[self.IMAGE_COL].values.append(image.id.val)
                except ValueError:
                    for i, value in enumerate(row):
                        columns[i + 2].name = value
                    break
        log.debug("Returning %d columns" % len(columns))
        return MeasurementParsingResult([columns])

    def _parse_neo_roi(self, columns):
        """Parses out ROI from OmeroTables columns for 'NEO' datasets."""
        log.debug("Parsing %s NEO ROIs..." % (len(columns[0].values)))
        image_ids = columns[self.IMAGE_COL].values
        rois = list()
        # Save our file annotation to the database so we can use an unloaded
        # annotation for the saveAndReturnIds that will be triggered below.
        self.file_annotation = \
            self.update_service.saveAndReturnObject(self.file_annotation)
        unloaded_file_annotation = \
            FileAnnotationI(self.file_annotation.id.val, False)
        batch_no = 1
        batches = dict()
        for i, image_id in enumerate(image_ids):
            unloaded_image = ImageI(image_id, False)
            roi = RoiI()
            shape = EllipseI()
            values = columns[6].values
            diameter = rdouble(float(values[i]))
            shape.theZ = rint(0)
            shape.theT = rint(0)
            values = columns[4].values
            shape.x = rdouble(float(values[i]))
            values = columns[3].values
            shape.y = rdouble(float(values[i]))
            shape.radiusX = diameter
            shape.radiusY = diameter
            roi.addShape(shape)
            roi.image = unloaded_image
            roi.linkAnnotation(unloaded_file_annotation)
            rois.append(roi)
            if len(rois) == self.ROI_UPDATE_LIMIT:
                self.thread_pool.add_task(
                    self.update_rois, rois, batches, batch_no)
                rois = list()
                batch_no += 1
        self.thread_pool.add_task(self.update_rois, rois, batches, batch_no)
        self.thread_pool.wait_completion()
        batch_keys = list(batches.keys())
        batch_keys.sort()
        for k in batch_keys:
            columns[self.ROI_COL].values += batches[k]

    def _parse_mnu_roi(self, columns):
        """Parses out ROI from OmeroTables columns for 'MNU' datasets."""
        log.debug("Parsing %s MNU ROIs..." % (len(columns[0].values)))
        image_ids = columns[self.IMAGE_COL].values
        rois = list()
        # Save our file annotation to the database so we can use an unloaded
        # annotation for the saveAndReturnIds that will be triggered below.
        self.file_annotation = \
            self.update_service.saveAndReturnObject(self.file_annotation)
        unloaded_file_annotation = \
            FileAnnotationI(self.file_annotation.id.val, False)
        batch_no = 1
        batches = dict()
        for i, image_id in enumerate(image_ids):
            unloaded_image = ImageI(image_id, False)
            roi = RoiI()
            shape = PointI()
            shape.theZ = rint(0)
            shape.theT = rint(0)
            values = columns[3].values
            shape.x = rdouble(float(values[i]))
            values = columns[2].values
            shape.y = rdouble(float(values[i]))
            roi.addShape(shape)
            roi.image = unloaded_image
            roi.linkAnnotation(unloaded_file_annotation)
            rois.append(roi)
            if len(rois) == self.ROI_UPDATE_LIMIT:
                self.thread_pool.add_task(
                    self.update_rois, rois, batches, batch_no)
                rois = list()
                batch_no += 1
        self.thread_pool.add_task(self.update_rois, rois, batches, batch_no)
        self.thread_pool.wait_completion()
        batch_keys = list(batches.keys())
        batch_keys.sort()
        for k in batch_keys:
            columns[self.ROI_COL].values += batches[k]

    def parse_and_populate_roi(self, columns):
        names = [column.name for column in columns]
        neo = [name in self.NEO_EXPECTED for name in names]
        mnu = [name in self.MNU_EXPECTED for name in names]
        for name in names:
            log.debug("Column: %s" % name)
        if len(columns) == 9 and False not in neo:
            self._parse_neo_roi(columns)
        elif len(columns) == 5 and False not in mnu:
            self._parse_mnu_roi(columns)
        else:
            log.warn("Unknown ROI type for MIAS dataset: %r" % names)

    def populate(self, columns):
        """
        Query performed::

                first_roi = columns[self.ROI_COL].values[0]
                first_roi = self.query_service.findByQuery(
                        'select roi from Roi as roi ' \
                        'join fetch roi.annotationLinks as link ' \
                        'join fetch link.child ' \
                        'where roi.id = %d' % first_roi, None)
                self.file_annotation = first_roi.copyAnnotationLinks()[0].child
        """
        self.update_table(columns)


class FlexMeasurementCtx(AbstractMeasurementCtx):

    """
    Flex measurements are located deep within a ".res" XML file container
    and contain no ROI.
    """

    # The XPath to the <Area> which aggregate an acquisition
    AREA_XPATH = './/Areas/Area'

    # The XPath to the an analysis <Parameter>; will become a column header
    # and is below AREA_XPATH
    PARAMETER_XPATH = './/Wells/ResultParameters/Parameter'

    # The XPath to a <Well> which has had at least one acquisition event
    # within and is below AREA_XPATH
    WELL_XPATH = './/Wells/Well'

    # The XPath to a <Result> for a given well and is below WELL_XPATH
    RESULT_XPATH = './/Result'

    def __init__(self, analysis_ctx, service_factory, original_file_provider,
                 original_file, result_files):
        super(FlexMeasurementCtx, self).__init__(
            analysis_ctx, service_factory, original_file_provider,
            original_file, result_files)

    def get_empty_columns(self, headers):
        """
        Retrieves a set of empty OmeroTables columns for the analysis results
        prefixed by a WellColumn to handle linked object indexes.
        """
        columns = {'Well': WellColumn('Well', '', list())}
        for header in headers:
            columns[header] = DoubleColumn(header, '', list())
        return columns

    ###
    # Abstract method implementations
    ###

    def get_name(self, set_of_columns=None):
        return self.original_file.name.val[:-4]

    def parse(self):
        log.info("Parsing: %s" % self.original_file.name.val)
        provider = self.original_file_provider
        data = provider.get_original_file_data(self.original_file)
        try:
            et = ElementTree(file=data)
        finally:
            data.close()
        root = et.getroot()
        areas = root.findall(self.AREA_XPATH)
        log.debug("Area count: %d" % len(areas))
        for i, area in enumerate(areas):
            result_parameters = area.findall(self.PARAMETER_XPATH)
            log.debug("Area %d result children: %d" %
                      (i, len(result_parameters)))
            if len(result_parameters) == 0:
                log.warn("%s contains no analysis data." % self.get_name())
                return
            headers = list()
            for result_parameter in result_parameters:
                headers.append(result_parameter.text)
            columns = self.get_empty_columns(headers)
            wells = area.findall(self.WELL_XPATH)
            for well in wells:
                # Rows and columns are 1-indexed, OMERO wells are 0-indexed
                row = int(well.get('row')) - 1
                column = int(well.get('col')) - 1
                try:
                    v = columns['Well'].values
                    wellobj, images = self.get_well_images(row, column)
                    if not wellobj:
                        continue
                    v.append(wellobj.id.val)
                except:
                    log.exception("ERROR: Failed to get well images")
                    continue
                results = well.findall(self.RESULT_XPATH)
                for result in results:
                    name = result.get('name')
                    columns[name].values.append(float(result.text))
        return MeasurementParsingResult([list(columns.values())])

    def parse_and_populate_roi(self, columns):
        pass

    def populate(self, columns):
        self.update_table(columns)


class InCellMeasurementCtx(AbstractMeasurementCtx):

    """
    InCell Analyzer measurements are located deep within an XML file container.
    """

    # Cells expected centre of gravity columns
    CELLS_CG_EXPECTED = ['Cell: cgX', 'Cell: cgY']

    # Nulcei expected centre of gravity columns
    NUCLEI_CG_EXPECTED = ['Nucleus: cgX', 'Nucleus: cgY']

    # Expected source attribute value for cell data
    CELLS_SOURCE = 'Cells'

    # Expected source attribute value for nuclei data
    NUCLEI_SOURCE = 'Nuclei'

    # Expected source attribute value for organelle data
    ORGANELLES_SOURCE = 'Organelles'

    def __init__(self, analysis_ctx, service_factory, original_file_provider,
                 original_file, result_files):
        super(InCellMeasurementCtx, self).__init__(
            analysis_ctx, service_factory, original_file_provider,
            original_file, result_files)

    def check_sparse_data(self, columns):
        """
        Checks a set of columns for sparse data (one column shorter than
        the rest) and adds -1 where appropriate.
        """
        length = None
        for i, column in enumerate(columns):
            if column.name == 'ROI':
                # ROI are processed late so we don't care if this column
                # is sparse or not.
                continue
            current_length = len(column.values)
            if length is not None:
                if current_length > length:
                    log.debug("%s length %d > %d modding previous column" %
                              (column.name, current_length, length))
                    columns[i - 1].values.append(-1.0)
                if current_length < length:
                    log.debug("%s length %d < %d modding current column" %
                              (column.name, current_length, length))
                    column.values.append(-1.0)
            length = len(column.values)

    ###
    # Abstract method implementations
    ###

    def get_name(self, set_of_columns=None):
        if set_of_columns is None:
            return self.original_file.name.val[:-4]
        if set_of_columns == 0:
            return self.original_file.name.val[:-4] + ' Cells'
        if set_of_columns == 1:
            return self.original_file.name.val[:-4] + ' Nuclei'
        if set_of_columns == 2:
            return self.original_file.name.val[:-4] + ' Organelles'

    def parse(self):
        log.info("Parsing: %s" % self.original_file.name.val)
        provider = self.original_file_provider
        data = provider.get_original_file_data(self.original_file)
        try:
            events = ('start', 'end')
            well_data = None
            n_roi = 0
            n_measurements = 0
            cells_columns = {'Image': ImageColumn('Image', '', list()),
                             'Cell': LongColumn('Cell', '', list()),
                             'ROI': RoiColumn('ROI', '', list())
                             }
            organelles_columns = {'Image': ImageColumn('Image', '', list()),
                                  'Cell': LongColumn('Cell', '', list()),
                                  }
            nuclei_columns = {'Image': ImageColumn('Image', '', list()),
                              'Cell': LongColumn('Cell', '', list()),
                              'ROI': RoiColumn('ROI', '', list())
                              }
            for event, element in iterparse(data, events=events):
                if event == 'start' and element.tag == 'WellData' \
                   and element.get('cell') != 'Summary':
                    row = int(element.get('row')) - 1
                    col = int(element.get('col')) - 1
                    i = int(element.get('field')) - 1
                    try:
                        well, images = self.get_well_images(row, col)
                        if not images:
                            continue
                        image = images[i]
                    except:
                        log.exception("ERROR: Failed to get well images")
                        continue
                    self.check_sparse_data(list(cells_columns.values()))
                    self.check_sparse_data(list(nuclei_columns.values()))
                    self.check_sparse_data(list(organelles_columns.values()))
                    cell = int(element.get('cell'))
                    cells_columns['Cell'].values.append(cell)
                    nuclei_columns['Cell'].values.append(cell)
                    organelles_columns['Cell'].values.append(cell)
                    well_data = element
                    cells_columns['Image'].values.append(image.id.val)
                    nuclei_columns['Image'].values.append(image.id.val)
                    organelles_columns['Image'].values.append(image.id.val)
                elif well_data is not None and event == 'start' \
                        and element.tag == 'Measure':
                    source = element.get('source')
                    key = element.get('key')
                    value = float(element.get('value'))
                    if source == self.CELLS_SOURCE:
                        columns_list = [cells_columns]
                    elif source == self.NUCLEI_SOURCE:
                        columns_list = [nuclei_columns]
                    elif source == self.ORGANELLES_SOURCE:
                        columns_list = [organelles_columns]
                    else:
                        columns_list = [cells_columns, nuclei_columns,
                                        organelles_columns]
                    for columns in columns_list:
                        if key not in columns:
                            columns[key] = DoubleColumn(key, '', list())
                        columns[key].values.append(value)
                    n_measurements += 1
                elif event == 'end' and element.tag == 'WellData':
                    if well_data is not None:
                        n_roi += 1
                        well_data.clear()
                        well_data = None
                else:
                    element.clear()
            # Final row sparseness check
            self.check_sparse_data(list(cells_columns.values()))
            self.check_sparse_data(list(nuclei_columns.values()))
            self.check_sparse_data(list(organelles_columns.values()))
            log.info("Total ROI: %d" % n_roi)
            log.info("Total measurements: %d" % n_measurements)
            sets_of_columns = [list(cells_columns.values()), list(nuclei_columns.values()),
                               list(organelles_columns.values())]
            return MeasurementParsingResult(sets_of_columns)
        finally:
            data.close()

    def parse_and_populate_roi(self, columns_as_list):
        # First sanity check our provided columns
        names = [column.name for column in columns_as_list]
        log.debug('Parsing columns: %r' % names)
        cells_expected = [name in names for name in self.CELLS_CG_EXPECTED]
        nuclei_expected = [name in names for name in self.NUCLEI_CG_EXPECTED]
        if (False in cells_expected) and (False in nuclei_expected):
            log.warn("Missing CGs for InCell dataset: %r" % names)
            log.warn('Removing resultant empty ROI column.')
            for column in columns_as_list:
                if RoiColumn == column.__class__:
                    columns_as_list.remove(column)
            return
        # Reconstruct a column name to column map
        columns = dict()
        for column in columns_as_list:
            columns[column.name] = column
        image_ids = columns['Image'].values
        rois = list()
        # Save our file annotation to the database so we can use an unloaded
        # annotation for the saveAndReturnIds that will be triggered below.
        self.file_annotation = \
            self.update_service.saveAndReturnObject(self.file_annotation)
        unloaded_file_annotation = \
            FileAnnotationI(self.file_annotation.id.val, False)
        # Parse and append ROI
        batch_no = 1
        batches = dict()
        for i, image_id in enumerate(image_ids):
            unloaded_image = ImageI(image_id, False)
            if False in nuclei_expected:
                # Cell centre of gravity
                roi = RoiI()
                shape = PointI()
                shape.theZ = rint(0)
                shape.theT = rint(0)
                shape.x = rdouble(float(columns['Cell: cgX'].values[i]))
                shape.y = rdouble(float(columns['Cell: cgY'].values[i]))
                roi.addShape(shape)
                roi.image = unloaded_image
                roi.linkAnnotation(unloaded_file_annotation)
                rois.append(roi)
            elif False in cells_expected:
                # Nucleus centre of gravity
                roi = RoiI()
                shape = PointI()
                shape.theZ = rint(0)
                shape.theT = rint(0)
                shape.x = rdouble(float(columns['Nucleus: cgX'].values[i]))
                shape.y = rdouble(float(columns['Nucleus: cgY'].values[i]))
                roi.addShape(shape)
                roi.image = unloaded_image
                roi.linkAnnotation(unloaded_file_annotation)
                rois.append(roi)
            else:
                raise MeasurementError('Not a nucleus or cell ROI')
            if len(rois) == self.ROI_UPDATE_LIMIT:
                self.thread_pool.add_task(
                    self.update_rois, rois, batches, batch_no)
                rois = list()
                batch_no += 1
        self.thread_pool.add_task(self.update_rois, rois, batches, batch_no)
        self.thread_pool.wait_completion()
        batch_keys = list(batches.keys())
        batch_keys.sort()
        for k in batch_keys:
            columns['ROI'].values += batches[k]

    def populate(self, columns):
        self.update_table(columns)

if __name__ == "__main__":
    try:
        options, args = getopt(sys.argv[1:], "s:p:u:m:k:t:id")
    except GetoptError as xxx_todo_changeme:
        (msg, opt) = xxx_todo_changeme.args
        usage(msg)

    try:
        plate_id, = args
        plate_id = int(plate_id)
    except ValueError:
        usage("Plate ID must be a specified and a number!")

    username = None
    hostname = None
    port = 4064  # SSL
    measurement = None
    info = False
    session_key = None
    logging_level = logging.INFO
    thread_count = 1
    for option, argument in options:
        if option == "-u":
            username = argument
        if option == "-s":
            hostname = argument
        if option == "-p":
            port = int(argument)
        if option == "-m":
            measurement = int(argument)
        if option == "-i":
            info = True
        if option == "-k":
            session_key = argument
        if option == "-d":
            logging_level = logging.DEBUG
        if option == "-t":
            thread_count = int(argument)
    if session_key is None and username is None:
        usage("Username must be specified!")
    if session_key is None and hostname is None:
        usage("Host name must be specified!")
    if session_key is None:
        password = getpass()

    logging.basicConfig(level=logging_level)
    c = client(hostname, port)
    c.setAgent("OMERO.populate_roi")
    c.enableKeepAlive(60)
    try:
        if session_key is not None:
            service_factory = c.joinSession(session_key)
        else:
            service_factory = c.createSession(username, password)

        log.debug('Creating pool of %d threads' % thread_count)
        thread_pool = ThreadPool(thread_count)
        factory = PlateAnalysisCtxFactory(service_factory)
        analysis_ctx = factory.get_analysis_ctx(plate_id)
        n_measurements = analysis_ctx.get_measurement_count()
        if measurement is not None and measurement >= n_measurements:
            usage("measurement %d not a valid index!")
        if info:
            for i in range(n_measurements):
                n_result_files = analysis_ctx.get_result_file_count(i)
                print("Measurement %d has %d result files." % \
                    (i, n_result_files))
            sys.exit(0)
        if measurement is not None:
            measurement_ctx = analysis_ctx.get_measurement_ctx(measurement)
            measurement_ctx.parse_and_populate()
        else:
            for i in range(n_measurements):
                measurement_ctx = analysis_ctx.get_measurement_ctx(i)
                measurement_ctx.parse_and_populate()
    finally:
        c.closeSession()