* [WIP] gdrive: add workaround for downloading empty files
Better to solve it on PyDrive's side:
https://github.com/iterative/dvc/issues/4286Fixes#4286
* deps: limit flatten_json
```
pylint...................................................................Failed
- hook id: pylint
- exit code: 2
Registered custom plugin. Some checks will be disabled for tests.
************* Module dvc.command.experiments
dvc/command/experiments.py:14:4: E0401: Unable to import 'flatten_json'
(import-error)
dvc/command/experiments.py:39:4: E0401: Unable to import 'flatten_json'
(import-error)
-----------------------------------
Your code has been rated at 9.99/10
Registered custom plugin. Some checks will be disabled for tests.
************* Module dvc.utils.diff
dvc/utils/diff.py:38:8: E0401: Unable to import 'flatten_json'
(import-error)
************* Module dvc.command.metrics
dvc/command/metrics.py:17:4: E0401: Unable to import 'flatten_json'
(import-error)
-----------------------------------
```
```
pylint...................................................................Failed
- hook id: pylint
- exit code: 2
Registered custom plugin. Some checks will be disabled for tests.
************* Module dvc.command.experiments
dvc/command/experiments.py:14:4: E0401: Unable to import 'flatten_json'
(import-error)
dvc/command/experiments.py:39:4: E0401: Unable to import 'flatten_json'
(import-error)
-----------------------------------
Your code has been rated at 9.99/10
Registered custom plugin. Some checks will be disabled for tests.
************* Module dvc.utils.diff
dvc/utils/diff.py:38:8: E0401: Unable to import 'flatten_json'
(import-error)
************* Module dvc.command.metrics
dvc/command/metrics.py:17:4: E0401: Unable to import 'flatten_json'
(import-error)
-----------------------------------
```
This allows us avoid collecting dvcignore for the whole repo if we only
care about particular paths. As a result, in a repo with 2 datasets
(2M + 0.5M files), creating a defunct stage takes ~4sec on 1.2.0, but
~1sec(most of it is actually dvc module initialization) with this PR.
This is also a pre-requisite for dynamic dvcignore and subrepo
collection (https://github.com/iterative/dvc/pull/4247) while walking
the tree.
Also, it is important to clarify that regular `dvc status`(without
arguments) has the same performance after this PR, because when we check
dataset for changes, we call things like `tree.exists()`, which call
dvcignore and make it collect dvcignore in the dataset itself, so we
still endup collecting dvcignore for the whole repo (including walking
into the datasets). This should be solved soon by telling dvcignore that
it shouldn't walk into the datasets searching for `.dvcignore`s.
We used `Popen` just to provide a general way of launching our daemons
on both Windows and *nix, but it is causing us issues when dvc is
launched from some exotic places (like nix package) or through `main()`
helper from a script that is not dvc.
This patch makes us just use `main()` directly in our double-forked
process.
Windows implementation has to use the `Popen` mechanism, because this is
the only way to daemonize a process on Windows. This means that dvc on
Windows is still susceptible to this problem, but there is nothing we
can do about it right now :(
Fixes#4206Fixes#4165
* dvc: get rid of is_walking_tree
No longer needed. Some applications of `isinstance()` will be refactored
in the future, as trees will just provide the needed functionality.
* dvc: move GitTree to dvc/tree/
* dvc: rename *RemoteTree to *Tree
* dvcignore: move file/dir/parent checks into DvcIgnoreFilter
* dvc: get rid of CleanTree
CleanTree is a very awkward wrapper that spreads the context across
the codebase and results in unexpected behaviour when using it.
This PR starts moving dvcignore-related logic into the trees themselves
(it makes a lot of sense, kinda like `state`) so they could deal with
it however they like.
There are at least two temporary ugly parts about this PR:
1) dvcignore is used by individual trees and not packed into
tree/base.py yet;
2) `dvcignore_root` argument. This one is caused by the dvcignore trying
to collect everything topdown starting from the certain root dir. What
it should do instead is for a certain path that is being checked look up
the tree through the parents until it finds repo root (.dvc dir) and
then stop. That would handle subrepos as well. At the same time we need
to leverage existing dvcignore trie structure to cache those results.
Working tree is really just a regular local tree and should be
used by outputs when trying to compute a hash for themselves.
We didn't use it previously because local tree was embeded into
the local remote class.
Related to #4050
Outputs and dependencies don't really care about the remotes, as they
don't do any pushing or pulling. What they really need are the trees,
so let's use them directly.
Part of #4050
Usually we need it to parametrize tests with, so there is no
reason to duplicate the code. I've left `local_remote` fixture
just because it is so convenient and often used in a lot of tests.
* [RFC] tests: make read_text() work with directories
This PR is just a showcase to see how we like it. I've converted a few
tests to it, but there will be much more to come.
```
assert (tmp_dir / "file") == "contents"
assert (tmp_dir / "dir").read_text() == {"subdir": {"file": "hello"}}
```
This approach is nicer because it allows us to use a declaratory
asserts instead of relying on fs files. The problem becomes much more
apparent once you get into remote workspaces, where we `trees_equal`
doesn't work.
* status: pass tree as a kwarg
Fixes#4150
Depends on https://github.com/iterative/dvc/pull/4155
This PR is just a showcase to see how we like it. I've converted a few
tests to it, but there will be much more to come.
```
assert (tmp_dir / "file") == "contents"
assert (tmp_dir / "dir").read_text() == {"subdir": {"file": "hello"}}
```
This approach is nicer because it allows us to use a declaratory
asserts instead of relying on fs files. The problem becomes much more
apparent once you get into remote workspaces, where we `trees_equal`
doesn't work.
flufl.lock released 4.0 in which there is no finalize method anymore, hence
```
dvc/lock.py:143:16: E1101: Instance of 'HardlinkLock' has no 'finalize' member (no-member)
```
Clouds that have a non-noop `makedirs` should implement it in their own
method. Hdfs somehow slipped under the radar and had `mkdir` used
explicitly in `copy` and `upload` methods.
* tree: walk: dynamically load directories as we go
This allows us to save time by not loading directories that we won't be
using anyway. This also allows us to list directories for which we don't
need the access to remote.
* list: add support for tracked directories
We are currently missing url and license metadata:
```
Name : dvc
Arch : x86_64
Version : 0.94.0
Release : 1
Size : 172 M
Repo : installed
From repo : dvc
Summary : Data Version Control - datasets, models, and experiments
versioning for ML or data science projects
URL : http://example.com/no-uri-given
License : unknown
Description : Data Version Control - datasets, models, and experiments
versioning for ML or data science projects
```
https://github.com/iterative/dvc/issues/4088#issuecomment-648270797
Currently `RepoTree.open` will throw FileNotFoundError if you have a
file that is tracked by dvc, but which has no md5 or a cache for which
is missing, even though you have the file itself in your workspace.
This corresponds very nicely to the `--no-commit` or `--no-exec`
scenario, where you might not want to commit your artifacts to cache
just yet.
Fixes#3974
Ruslan Kuprieiev
