* don't allow loading stages without `cmd`
This was loosened when introducing parametrization.
But, `foreach`..`do` should be considered first
before stage's regular structure. And, cmd is made
required (as it should have been)
Related: #5371, #5370, #5312
* fix schema
When I opened it, I got confused for a second and had to double-check if it was really empty.
The empty file gives a false assurance. We do need to add one, which I'll add later, but for now,
we better remove it.
This happens because KeyError quotes the error message.
We'd still like to keep StageNotFound being a KeyError
semantics, but workaround the __str__ implementation.
Mypy was having some problem with the previous `__str__`
override.
Looking at the log is confusing, as the version is stuck in `1.11.0` whereas the recent one is `1.11.10`.
As the updater is also dependent on that version, it prints an ugly update message while running too.
https://github.com/iterative/dvc/issues/5023#issuecomment-758450519
* scm: implement gitignore; use `pygit2` backend
* Add is_ignored implementation on `pygit2` and `gitpython` backends.
* Also implement `._reset()` on `repo.scm` which is most of the changes.
* fix tests
* use _backend_func
* minor fixups
* Moved exception handling imports wherever they occur
* Move dvc.tree and dvc.external_repo imports to it's
nearest use, as it is expensive for `--version`.
* Move analytics import to when it's used
* Move format_link to place where error occurs
A great benefit to this is that, if the dvc.yaml file
is never interpolated, the parser is never built, which
reduces the gap even further between 1.0 and 2.0 files
I had to keep providing default value before when profiling
simple functions as an API. I added a default now, and
also made `debugtools` customizable so that you could
edit the code to, for example, have `pdb` enabled by default.
* Add pyinstrument: a statistical profiler
It is easier than cprofile, in that it can show stacktrace
of where things are slow.
Also refactored/moved debugging related utils to a separate
module
* use ipdb first, then fallback to pdb
* use ipdb first, then fallback to pdb
* Cleanup log messages when loading stages
1. Simplified callback stage debug message
2. Reduced no lock entry message to trace
If lockfile does not have some entry, it will simply complain
that lockfile is not uptodate once.
If there is no lockfile or empty lockfile, it will complain that
lockfile not found.
3. Annoying parametrization related wdir-interpolated debug message
removed.
* Update dvc/stage/__init__.py
* black
* Skip paramdependency for vars
Previously, `configs` which was a list of PathInfo was being used
to check if the `vars` should be skipped or not.
But, the file is stored as a string that made that check redundant
which might have caused #5117 as they were being stored twice,
once as a expanded paramdependency and next as a flattened dict
from vars, which later during `flatten`, it might have panicked
from that duplication.
* Use defaultdict
* change add_vars to collect_vars
minor refactors, code comments
Rename interim to definitions
provide location of entry during init for definition
small fixups to set_temporarily
Refactor loop to use dict comprehension
Provide same interface for definitions
Make it work for composite lists
Check for sequence rather than a list
refactor tests
Fixes#4955Fixes#4957Fixes#4963
We try to optimize `tree.exists` calls and probably few others
in that they either look directly into the workspace or,
to the cache without running graph checks. It does not seem
to be possible just to run graph checks on `find_outs_by_path`
due to those optimizations.
So, that's why, the `collect` also does a graph check for this
reason.
Fixes#5027Fixes#4010
* run/repro: show command on dry-run, add ">" to script output
This rollbacks the earlier commit to not show command during
the dry-run, and also prepends ">" in the command script
instead of "$" sign.
* fix tests
* repro: support regex/foreach-group to run at once
Fixes#4912Fixes#4886Fixes#4958
* Use `tree.isdir` rather than `os.path.isdir`
* Use glob rather than regex
* Update dvc/command/repro.py
* s/regex/glob
* disable glob on `collect_granular`
There's no need for a glob here
* add tests for `collect` and `collect_granular`
* s3: use mocked s3 instead of the real one
Not removing the fixtures just yet
* Update moto
Tests was failing, because of a change in `responses` package, which
`moto` was mocking. This was fixed on
https://github.com/spulec/moto/pull/3466 which is only available on
`dev122` release.
* Use mocked s3 in checkout tests
`Azure` tests are flaky for some reason. When I investigated one
of the job [1], something seemed wrong with the Azurite or
the client lib.
On `AzureTree.blob_service`, we do the following to check if the
container exists:
```python
try:
container_client.get_container_properties()
except ResourceNotFoundError:
container_client.create_container()
```
When looking at the test log, I found out that the
`get_container_properties` was failing with
`ErrorCode:ContainerNotFound` whereas `create_container`
successively was failing with `ErrorCode:ContainerAlreadyExists`.
`create_container` is correct, but `get_container_properties` is
incorrect, as many of the Azure tests have already run successfully
before.
So, either the Azurite has/had bug or the client lib has a bug.
Anyway, we'll find out after updating the Azurite.
[1]: https://github.com/iterative/dvc/runs/1448864706#step:7:4029
* parametrization: prevent dumping parametrized stages
This is a bit hackish, but does the job done.
It feels like we need some unified way to be able to load
and dump, but not sure how. Till then, this dataclass
will be helpful in communicating between those.
* s/Meta/RawData
* Update dvc/stage/loader.py
Some of the tests are run by default, other requires
`--enable-{remote}` to run. The remotes that are run by default
also have a `--disable-{remote}` flag. All of the remotes can be
enabled using `--all` flag.
To use this functionality in the fixture,
`test_config` fixture has to be used, and requires to call
`test_config.requires(remote)`.
In an individual test, you can simply use `pytest.mark.remote`
to make them skip or run.
Via pyparsing. The error message is not that great, as it's greedy.
But, it does point out where the mistakes are, correctly most of the
time. :)
Also raises appropriate error messages for ParseError
The schema is changed of `vars` to accomodate the both use of
`imports` and setting variables locally.
Now:
```
vars:
- params.yaml
- foo: foo
bar: bar
```
Also, same distinction occurs for the stages.
Saugat Pachhai