datalad/datalad: ## 0.14.0 (February 02, 2021)

Git versions below v2.19.1 are no longer supported. #4650

The minimum git-annex version is still 7.20190503, but, if you're on Windows (or use adjusted branches in general), please upgrade to at least 8.20200330 but ideally 8.20210127 to get subdataset-related fixes. #4292 #5290

The minimum supported version of Python is now 3.6. #4879

publish is now deprecated in favor of push. It will be removed in the 0.15.0 release at the earliest.

A new command runner was added in v0.13. Functionality related to the old runner has now been removed: Runner, GitRunner, and run_gitcommand_on_file_list_chunks from the datalad.cmd module along with the datalad.tests.protocolremote, datalad.cmd.protocol, and datalad.cmd.protocol.prefix configuration options. #5229

The --no-storage-sibling switch of create-sibling-ria is deprecated in favor of --storage-sibling=off and will be removed in a later release. #5090

The get_git_dir static method of GitRepo is deprecated and will be removed in a later release. Use the dot_git attribute of an instance instead. #4597

The ProcessAnnexProgressIndicators helper from datalad.support.annexrepo has been removed. #5259

The save argument of install, a noop since v0.6.0, has been dropped. #5278

The get_URLS method of AnnexCustomRemote is deprecated and will be removed in a later release. #4955

ConfigManager.get now returns a single value rather than a tuple when there are multiple values for the same key, as very few callers correctly accounted for the possibility of a tuple return value. Callers can restore the old behavior by passing get_all=True. #4924

In 0.12.0, all of the assure_* functions in datalad.utils were renamed as ensure_*, keeping the old names around as compatibility aliases. The assure_* variants are now marked as deprecated and will be removed in a later release. #4908

The datalad.inteface.run module, which was deprecated in 0.12.0 and kept as a compatibility shim for datalad.core.local.run, has been removed. #4583

The saver argument of datalad.core.local.run.run_command, marked as obsolete in 0.12.0, has been removed. #4583

The dataset_only argument of the ConfigManager class was deprecated in 0.12 and has now been removed. #4828

The linux_distribution_name, linux_distribution_release, and on_debian_wheezy attributes in datalad.utils are no longer set at import time and will be removed in a later release. Use datalad.utils.get_linux_distribution instead. #4696

datalad.distribution.clone, which was marked as obsolete in v0.12 in favor of datalad.core.distributed.clone, has been removed. #4904

datalad.support.annexrepo.N_AUTO_JOBS, announced as deprecated in v0.12.6, has been removed. #4904

The compat parameter of GitRepo.get_submodules, added in v0.12 as a temporary compatibility layer, has been removed. #4904

The long-deprecated (and non-functional) url parameter of GitRepo.__init__ has been removed. #5342

Cloning onto a system that enters adjusted branches by default (as Windows does) did not properly record the clone URL. #5128

The RIA-specific handling after calling clone was correctly triggered by ria+http URLs but not ria+https URLs. #4977

If the registered commit wasn't found when cloning a subdataset, the failed attempt was left around. #5391

The remote calls to cp and chmod in create-sibling were not portable and failed on macOS. #5108

A more reliable check is now done to decide if configuration files need to be reloaded. #5276

The internal command runner's handling of the event loop has been improved to play nicer with outside applications and scripts that use asyncio. #5350 #5367

The subdataset handling for adjusted branches, which is particularly important on Windows where git-annex enters an adjusted branch by default, has been improved. A core piece of the new approach is registering the commit of the primary branch, not its checked out adjusted branch, in the superdataset. Note: This means that git status will always consider a subdataset on an adjusted branch as dirty while datalad status will look more closely and see if the tip of the primary branch matches the registered commit. #5241

The performance of the subdatasets command has been improved, with substantial speedups for recursive processing of many subdatasets. #4868 #5076

Adding new subdatasets via save has been sped up. #4793

get, save, and addurls gained support for parallel operations that can be enabled via the --jobs command-line option or the new datalad.runtime.max-jobs configuration option. #5022

addurls

• learned how to read data from standard input. #4669
• now supports tab-separated input. #4845
• now lets Python callers pass in a list of records rather than a file name. #5285
• gained a --drop-after switch that signals to drop a file's content after downloading and adding it to the annex. #5081
• is now able to construct a tree of files from known checksums without downloading content via its new --key option. #5184
• records the URL file in the commit message as provided by the caller rather than using the resolved absolute path. #5091
• is now speedier. #4867 #5022

create-sibling-github learned how to create private repositories (thanks to Nolan Nichols). #4769

create-sibling-ria gained a --storage-sibling option. When --storage-sibling=only is specified, the storage sibling is created without an accompanying Git sibling. This enables using hosts without Git installed for storage. #5090

The download machinery (and thus the datalad special remote) gained support for a new scheme, shub://, which follows the same format used by singularity run and friends. In contrast to the short-lived URLs obtained by querying Singularity Hub directly, shub:// URLs are suitable for registering with git-annex. #4816

A provider is now included for https://registry-1.docker.io URLs. This is useful for storing an image's blobs in a dataset and registering the URLs with git-annex. #5129

The add-readme command now links to the DataLad handbook rather than http://docs.datalad.org. #4991

New option datalad.locations.extra-procedures specifies an additional location that should be searched for procedures. #5156

The class for handling configuration values, ConfigManager, now takes a lock before writes to allow for multiple processes to modify the configuration of a dataset. #4829

clone now records the original, unresolved URL for a subdataset under submodule.<name>.datalad-url in the parent's .gitmodules, enabling later get calls to use the original URL. This is particularly useful for ria+ URLs. #5346

Installing a subdataset now uses custom handling rather than calling git submodule update --init. This avoids some locking issues when running get in parallel and enables more accurate source URLs to be recorded. #4853

GitRepo.get_content_info, a helper that gets triggered by many commands, got faster by tweaking its git ls-files call. #5067

wtf now includes credentials-related information (e.g. active backends) in the its output. #4982

The call_git* methods of GitRepo now have a read_only parameter. Callers can set this to True to promise that the provided command does not write to the repository, bypassing the cost of some checks and locking. #5070

New call_annex* methods in the AnnexRepo class provide an interface for running git-annex commands similar to that of the GitRepo.call_git* methods. #5163

It's now possible to register a custom metadata indexer that is discovered by search and used to generate an index. #4963

The ConfigManager methods get, getbool, getfloat, and getint now return a single value (with same precedence as git config --get) when there are multiple values for the same key (in the non-committed git configuration, if the key is present there, or in the dataset configuration). For get, the old behavior can be restored by specifying get_all=True. #4924

Command-line scripts are now defined via the entry_points argument of setuptools.setup instead of the scripts argument. #4695

Interactive use of --help on the command-line now invokes a pager on more systems and installation setups. #5344

The datalad special remote now tries to eliminate some unnecessary interactions with git-annex by being smarter about how it queries for URLs associated with a key. #4955

The GitRepo class now does a better job of handling bare repositories, a step towards bare repositories support in DataLad. #4911

More internal work to move the code base over to the new command runner. #4699 #4855 #4900 #4996 #5002 #5141 #5142 #5229