Warning
This document is for an in-development version 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.visualization.data_providers.registry
from typing import (
Dict,
Optional,
Type,
Union,
)
from typing_extensions import Literal
from galaxy.datatypes.data import (
Data,
Newick,
Nexus,
)
from galaxy.datatypes.interval import (
Bed,
ChromatinInteractions,
ENCODEPeak,
Gff,
Gtf,
Interval,
)
from galaxy.datatypes.tabular import (
Tabular,
Vcf,
)
from galaxy.datatypes.xml import Phyloxml
from galaxy.model import NoConverterException
from galaxy.visualization.data_providers import genome
from galaxy.visualization.data_providers.basic import (
BaseDataProvider,
ColumnDataProvider,
)
from galaxy.visualization.data_providers.phyloviz import PhylovizDataProvider
# a dict keyed on datatype with a 'default' string key.
PROVIDER_BY_DATATYPE_CLASS_DICT = Dict[Union[Literal["default"], Type[Data]], Type[BaseDataProvider]]
DATA_PROVIDER_BY_TYPE_NAME_DICT = Dict[str, Union[Type[BaseDataProvider], PROVIDER_BY_DATATYPE_CLASS_DICT]]
[docs]class DataProviderRegistry:
"""
Registry for data providers that enables listing and lookup.
"""
[docs] def __init__(self):
# Mapping from dataset type name to a class that can fetch data from a file of that
# type. First key is converted dataset type; if result is another dict, second key
# is original dataset type.
self.dataset_type_name_to_data_provider: DATA_PROVIDER_BY_TYPE_NAME_DICT = {
"tabix": {
Vcf: genome.VcfTabixDataProvider,
Bed: genome.BedTabixDataProvider,
Gtf: genome.GtfTabixDataProvider,
ENCODEPeak: genome.ENCODEPeakTabixDataProvider,
Interval: genome.IntervalTabixDataProvider,
ChromatinInteractions: genome.ChromatinInteractionsTabixDataProvider,
"default": genome.TabixDataProvider,
},
"interval_index": genome.IntervalIndexDataProvider,
"bai": genome.BamDataProvider,
"bam": genome.SamDataProvider,
"bigwig": genome.BigWigDataProvider,
"bigbed": genome.BigBedDataProvider,
"column_with_stats": ColumnDataProvider,
}
[docs] def get_data_provider(self, trans, name=None, source="data", raw=False, original_dataset=None):
"""
Returns data provider matching parameter values. For standalone data
sources, source parameter is ignored.
"""
data_provider: Optional[BaseDataProvider]
data_provider_class: Type[BaseDataProvider]
# any datatype class that is a subclass of another needs to be
# checked before the parent in this conditional.
if raw:
# Working with raw data.
if isinstance(original_dataset.datatype, Gff):
data_provider_class = genome.RawGFFDataProvider
elif isinstance(original_dataset.datatype, Bed):
data_provider_class = genome.RawBedDataProvider
elif isinstance(original_dataset.datatype, Vcf):
data_provider_class = genome.RawVcfDataProvider
elif isinstance(original_dataset.datatype, Tabular):
data_provider_class = ColumnDataProvider
elif isinstance(original_dataset.datatype, (Nexus, Newick, Phyloxml)):
data_provider_class = PhylovizDataProvider
data_provider = data_provider_class(original_dataset=original_dataset)
else:
# Working with converted or standalone dataset.
if name:
# Provider requested by name; get from mappings.
value = self.dataset_type_name_to_data_provider[name]
if isinstance(value, dict):
# value is a PROVIDER_BY_DATATYPE_CLASS_DICT
# Get converter by dataset extension; if there is no data provider,
# get the default.
default_type = value.get("default")
assert default_type
data_provider_class = value.get(original_dataset.datatype.__class__, default_type)
else:
data_provider_class = value
# If name is the same as original dataset's type, dataset is standalone.
# Otherwise, a converted dataset is being used.
if name == original_dataset.ext:
data_provider = data_provider_class(original_dataset=original_dataset)
else:
converted_dataset = original_dataset.get_converted_dataset(trans, name)
deps = original_dataset.get_converted_dataset_deps(trans, name)
data_provider = data_provider_class(
original_dataset=original_dataset, converted_dataset=converted_dataset, dependencies=deps
)
elif original_dataset:
# No name, so look up a provider name from datatype's information.
# Dataset must have data sources to get data.
if not original_dataset.datatype.data_sources:
return None
# Get data provider mapping and data provider.
data_provider_mapping = original_dataset.datatype.data_sources
if "data_standalone" in data_provider_mapping:
data_provider = self.get_data_provider(
trans, name=data_provider_mapping["data_standalone"], original_dataset=original_dataset
)
else:
source_list = data_provider_mapping[source]
if isinstance(source_list, str):
source_list = [source_list]
# Find a valid data provider in the source list.
for source in source_list:
try:
data_provider = self.get_data_provider(
trans, name=source, original_dataset=original_dataset
)
break
except NoConverterException:
pass
return data_provider