Warning
This document is for an old release of Galaxy. You can alternatively view this page in the latest release if it exists or view the top of the latest release's documentation.
Source code for galaxy.datatypes.xml
"""
XML format classes
"""
import logging
import re
from galaxy import util
from galaxy.datatypes.metadata import MetadataElement
from galaxy.datatypes.sniff import (
build_sniff_from_prefix,
disable_parent_class_sniffing,
FilePrefix,
)
from . import (
data,
dataproviders,
)
log = logging.getLogger(__name__)
OWL_MARKER = re.compile(r"\<owl:")
SBML_MARKER = re.compile(r"\<sbml")
[docs]@dataproviders.decorators.has_dataproviders
@build_sniff_from_prefix
class GenericXml(data.Text):
"""Base format class for any XML file."""
edam_format = "format_2332"
file_ext = "xml"
[docs] def set_peek(self, dataset):
"""Set the peek and blurb text"""
if not dataset.dataset.purged:
dataset.peek = data.get_file_peek(dataset.file_name)
dataset.blurb = "XML data"
else:
dataset.peek = "file does not exist"
dataset.blurb = "file purged from disk"
def _has_root_element_in_prefix(self, file_prefix: FilePrefix, root):
for line in file_prefix.line_iterator():
if not line.startswith("<?"):
break
# pattern match <root or <ns:root for any ns string
pattern = r"^<(\w*:)?%s" % root
return re.match(pattern, line) is not None
[docs] def sniff_prefix(self, file_prefix: FilePrefix):
"""
Determines whether the file is XML or not
>>> from galaxy.datatypes.sniff import get_test_fname
>>> fname = get_test_fname( 'megablast_xml_parser_test1.blastxml' )
>>> GenericXml().sniff( fname )
True
>>> fname = get_test_fname( 'interval.interval' )
>>> GenericXml().sniff( fname )
False
"""
return file_prefix.startswith("<?xml ")
[docs] @staticmethod
def merge(split_files, output_file):
"""Merging multiple XML files is non-trivial and must be done in subclasses."""
if len(split_files) > 1:
raise NotImplementedError(
"Merging multiple XML files is non-trivial and must be implemented for each XML type"
)
# For one file only, use base class method (move/copy)
data.Text.merge(split_files, output_file)
[docs] @dataproviders.decorators.dataprovider_factory("xml", dataproviders.hierarchy.XMLDataProvider.settings)
def xml_dataprovider(self, dataset, **settings):
dataset_source = dataproviders.dataset.DatasetDataProvider(dataset)
return dataproviders.hierarchy.XMLDataProvider(dataset_source, **settings)
[docs]@disable_parent_class_sniffing
class MEMEXml(GenericXml):
"""MEME XML Output data"""
file_ext = "memexml"
[docs] def set_peek(self, dataset):
"""Set the peek and blurb text"""
if not dataset.dataset.purged:
dataset.peek = data.get_file_peek(dataset.file_name)
dataset.blurb = "MEME XML data"
else:
dataset.peek = "file does not exist"
dataset.blurb = "file purged from disk"
[docs]@disable_parent_class_sniffing
class CisML(GenericXml):
"""CisML XML data""" # see: http://www.ncbi.nlm.nih.gov/pubmed/15001475
file_ext = "cisml"
[docs] def set_peek(self, dataset):
"""Set the peek and blurb text"""
if not dataset.dataset.purged:
dataset.peek = data.get_file_peek(dataset.file_name)
dataset.blurb = "CisML data"
else:
dataset.peek = "file does not exist"
dataset.blurb = "file purged from disk"
[docs]class Dzi(GenericXml):
"""
Deep zoom image format, see
https://github.com/openseadragon/openseadragon/wiki/The-DZI-File-Format
"""
# General elements.
MetadataElement(
name="base_name", desc="Base name for this dataset", default="DeepZoomImage", readonly=True, set_in_upload=True
)
MetadataElement(name="format", desc="File format of the tiles", default=None, readonly=True, visible=True)
MetadataElement(name="tile_size", desc="Size of tiles", default=None, readonly=True, visible=True)
# Collection elements.
MetadataElement(
name="max_level", desc="Max pyramid level", default=None, readonly=True, optional=True, visible=True
)
MetadataElement(name="quality", desc="Quality", default=None, readonly=True, optional=True, visible=True)
# Image elements.
MetadataElement(name="height", desc="Size height", default=None, readonly=True, optional=True, visible=True)
MetadataElement(
name="overlap",
desc="Overlap of all four sides of tiles",
default=None,
readonly=True,
optional=True,
visible=True,
)
MetadataElement(name="width", desc="Size width", default=None, readonly=True, optional=True, visible=True)
file_ext = "dzi"
[docs] def set_meta(self, dataset, **kwd):
tree = util.parse_xml(dataset.file_name)
root = tree.getroot()
dataset.metadata.format = root.get("Format")
dataset.metadata.tile_size = root.get("TileSize")
# DeepZoom image files can include
# xml namespace attributes.
if root.tag.find("Collection") >= 0:
dataset.metadata.max_level = root.get("MaxLevel")
dataset.metadata.quality = root.get("Quality")
elif root.tag.find("Image") >= 0:
dataset.metadata.overlap = root.get("Overlap")
for elem in root:
if elem.tag.find("Size") >= 0:
dataset.metadata.width = elem.get("Width")
dataset.metadata.height = elem.get("Height")
[docs] def get_visualizations(self, dataset):
"""Returns a list of visualizations for datatype"""
return ["openseadragon"]
[docs] def set_peek(self, dataset):
if not dataset.dataset.purged:
dataset.peek = data.get_file_peek(dataset.file_name)
dataset.blurb = "Deep Zoom Image"
else:
dataset.peek = "file does not exist"
dataset.blurb = "file purged from disc"
[docs] def sniff_prefix(self, file_prefix: FilePrefix):
"""
Checking for keyword - 'Collection' or 'Image' in the first 200 lines.
>>> from galaxy.datatypes.sniff import get_test_fname
>>> fname = get_test_fname('1.dzi')
>>> Dzi().sniff(fname)
True
>>> fname = get_test_fname('megablast_xml_parser_test1.blastxml')
>>> Dzi().sniff(fname)
False
"""
for line in file_prefix.line_iterator():
line = line.lower()
if line.find("<collection") >= 0 or line.find("<image") >= 0:
return True
return False
[docs]class Phyloxml(GenericXml):
"""Format for defining phyloxml data http://www.phyloxml.org/"""
edam_data = "data_0872"
edam_format = "format_3159"
file_ext = "phyloxml"
[docs] def set_peek(self, dataset):
"""Set the peek and blurb text"""
if not dataset.dataset.purged:
dataset.peek = data.get_file_peek(dataset.file_name)
dataset.blurb = "Phyloxml data"
else:
dataset.peek = "file does not exist"
dataset.blurb = "file purged from disk"
[docs] def sniff_prefix(self, file_prefix: FilePrefix):
""" "Checking for keyword - 'phyloxml' always in lowercase in the first few lines.
>>> from galaxy.datatypes.sniff import get_test_fname
>>> fname = get_test_fname( '1.phyloxml' )
>>> Phyloxml().sniff( fname )
True
>>> fname = get_test_fname( 'interval.interval' )
>>> Phyloxml().sniff( fname )
False
>>> fname = get_test_fname( 'megablast_xml_parser_test1.blastxml' )
>>> Phyloxml().sniff( fname )
False
"""
return self._has_root_element_in_prefix(file_prefix, "phyloxml")
[docs] def get_visualizations(self, dataset):
"""
Returns a list of visualizations for datatype.
"""
return ["phyloviz"]
[docs]class Owl(GenericXml):
"""
Web Ontology Language OWL format description
http://www.w3.org/TR/owl-ref/
"""
edam_format = "format_3262"
file_ext = "owl"
[docs] def set_peek(self, dataset):
if not dataset.dataset.purged:
dataset.peek = data.get_file_peek(dataset.file_name)
dataset.blurb = "Web Ontology Language OWL"
else:
dataset.peek = "file does not exist"
dataset.blurb = "file purged from disc"
[docs] def sniff_prefix(self, file_prefix: FilePrefix):
"""
Checking for keyword - '<owl' in the first 200 lines.
"""
return file_prefix.search(OWL_MARKER)
[docs]class Sbml(GenericXml):
"""
System Biology Markup Language
http://sbml.org
"""
file_ext = "sbml"
edam_data = "data_2024"
edam_format = "format_2585"
[docs] def set_peek(self, dataset):
if not dataset.dataset.purged:
dataset.peek = data.get_file_peek(dataset.file_name)
dataset.blurb = "System Biology Markup Language SBML"
else:
dataset.peek = "file does not exist"
dataset.blurb = "file purged from disc"
[docs] def sniff_prefix(self, file_prefix: FilePrefix):
"""
Checking for keyword - '<sbml' in the first 200 lines.
"""
return file_prefix.search(SBML_MARKER)