This new command allows to reapply a template, keeping old answers but discarding subproject evolution.
It is useful when there are bugs replaying an old version of the template, or when the subproject has drifted too much from the template and you need to reset it.
BREAKING CHANGE: All CLI calls to Copier must now include the subcommand as the 1st argument. For example, `copier` must become now `copier update`; also `copier ./tpl ./dst` must become `copier copy ./tpl ./dst`.
BREAKING CHANGE: All flags must go after the subcommand now. For example, `copier -r HEAD update ./dst` must now become `copier update -r HEAD ./dst` or `copier update ./dst -r HEAD`.
BREAKING CHANGE: Automatic mode removed. Since now subcommands are required, the automatic mode is removed.
BREAKING CHANGE: Deprecated `copier.copy` function is removed. Use `copier.run_copy`, `copier.run_update` or `copier.run_recopy` explicitly as needed.
Fix https://github.com/copier-org/copier/issues/1081
Close https://github.com/copier-org/copier/issues/1082
I've indirectly fixed the validation of answers to questions for which the skip condition is met. Before, a question with a truthy skip condition was not properly skipped such that (a) its default value (if present) was still validated and could cause a validation error or (b) an error was raised that the question was required although it should have been skipped. As a side effect of the previous implementation (which didn't properly skip questions), answers of skipped questions were recorded in the answers file which IMO makes no sense.
I've changed the way skipped questions are handled such that the skip condition is evaluated early and the question is immediately skipped when the skip condition is truthy. To prevent records of such answers in the answers file, e.g. when they originate from previous runs, when they are provided by the user via the `--data` flag, etc., any answers in the internal answers map are removed (or rather marked as removed). As a result, I needed to adapt a few tests and even removed one that specifically tested for the presence of those records in the answers file.
This new behavior should be backwards compatible and is more logical in my opinion. As a side effect, some slightly confusing code block could be removed which handled some special case of using a templated default answer.
Fixes#1125.
* refactor(tests): add type hints and clean up
* fix: import `Protocol` from `typing-extensions` when using Python 3.7
* fix: import `Literal` from `typing-extensions` when using Python 3.7
* refactor(tests): create template and subproject directory using `tmp_path_factory` fixture
* refactor(tests): use OS-agnostic file path separator
* refactor(tests): use `/` separator for `skip_if_exists` paths
* style: fix formatting error
* refactor(tests): expect POSIX path for local repo URL
* docs: add note to docstring
* feat: print spawn logs on failure, and fix failure in test
The test was failing because it turns out that `tui` won't print letters that must not change, when running an interactive TUI.
I changed the error message for something that will force all letters to change. This way it works as expected.
Co-authored-by: Jairo Llopis <yajo.sk8@gmail.com>
Before this patch, an answers file could remember answers to questions that were not present in the questionary.
STR:
1. In one project, apply one template that has a `.copier-answers.yml` file.
2. Now apply another template that has a different default value in `_answers_file`.
3. The second answers file would contain answers to the 1st template too.
@moduon MT-616
A skipped question was not saved by Questionary, so I need to check if the answer is exactly like the question and, in that case, re-render it again.
Docs clarified to avoid false expectations on what should be the answer.
Fixes https://github.com/copier-org/copier/issues/529
* Refactor
* Fix#110.
* Rewrite test_config_exclude, test_config_exclude_overridden and test_config_include. These tests were badly designed, using a monkeypatch that would never happen in the real world, and actually producing false positives. I moved them to test_exclude.py and rewritten to test the what and not the how.
* Fix#214 by removing skip option. Relevant tests use the better skip_if_exists=["**"].
* Remove subdirectory flag from API/CLI. It was confusing and could lead to bad maintenance situations. Fixes#315.
* Remove extra_paths and fix#321
* Remember that you cannot use _copier_conf.src_path as a path now
* use dataclasses
* Create errors module, simplify some tests, fix many others
* Fix some tests, complete EnvOps removal
* Fix#214 and some tests related to it
* Reorder code
* Update docs and imports
* Modularize test_complex_questions
* Interlink worker and questionary a bit better
* Removal of Questionary class, which only had 1 meaningful method that is now merged into Worker to avoid circular dependencies.
* Fix#280 in a simple way: only user answers are type-casted inside API, and CLI transforms all `--data` using YAML always. Predictable.
* Use prereleases correctly.
* Reorder AnswersMap to have a more significative repr.
* Simpler cache for old `Question.get_choices()` method (renamed now).
* fix wrong test
* Fix test_subdirectory
* Fix test_tasks (and remove tests/demo_tasks)
* Fix path filter tests, and move it to test_exclude, where it belongs
* Make test_config pass
* Fix more wrongly designed tests
* Use cached_property backport if needed
* xfail test known to fail on mac
* force posix paths on windows
* Add typing_extensions for python < 3.8
* Sort dependencies in pyproject.toml
* Support python 3.6 str-to-datetime conversion
* Workaround https://bugs.python.org/issue43095
* xfail test_path_filter on windows
* Upgrade mkdocs and other dependencies to fix https://github.com/pawamoy/mkdocstrings/issues/222
* Add missing reference docs.
* Add workaround for https://github.com/pawamoy/mkdocstrings/pull/209
* Docs.
* Remove validators module
* Add workaround for https://github.com/pawamoy/mkdocstrings/issues/225
* Restore docs autorefs as explained in https://github.com/pawamoy/mkdocstrings/issues/226#issuecomment-775413562.
* Workaround https://github.com/pawamoy/pytkdocs/issues/86
* use a new toolkit for user prompting: questionary (patched).
* multiline questions.
* conditional questions.
* new toolkit for ui tests: pexpect.
* fix lots of tests.
* fix windows builds with newer poetry-dynamic-versioning.
* linters and mypy are now tests, to have faster ci.
* update deprecated ci commands.
* removed dependencies from old times.
* Remove toml 0.5+ syntax (dotted keys)
* Use powershell where syntax is compatible
* Change skip to xfail
* xfail pexpect tests on windows. I'm tired of trying to make it work
* more docs
* possibly something more.
- `ConfigData` objects now know where do answers come from.
- Their `.data` attribute is now a computed property that merges all those answer sources in priority order.
- To merge dicts, `ChainMap` is used extensively.
- `query_user_data` gets both `last_answer_data` and `forced_answers_data` instead of the single `answers_data` argument it got before.
- By knowing where does an answer come from, now Copier avoids asking something if the user already answered that from other source. For example, if you use `copier -d some_question=some_answer copy $src $dst`, then Copier will not ask you about `some_question` again, because you already answered that. 🎉
- All answer sources are deep-copied to avoid modifying mutable dicts and affecting further copier executions.
- A minimal amount of tests got updated as they were now supposed to fail with new behavior.
- Ignore flake8 `E501 line too long` errors. We use black, that's just unnecessarily annoying.
@Tecnativa TT23705
When data came from `--data` flag, it was always a string.
If some data was expected to be of another type, it could make the template fail.
The type casting system must be applied also to user answers, so here's the patch, docs and tests.
@Tecnativa TT20357
When using the `--data` CLI option, all incoming data are strings. This is the expected behavior because at this point we still don't know the expected types of all answers (we don't even know the questions).
Now the type casting is done smartly when strings come in any form.
There are 2 special cases handled here:
1. When the expected type is bool and the incoming data is string. Handled by YAML because strings like "false", "0" and "no" would evaluate to `True` otherwise.
2. When the expected type is JSON or YAML but the incoming data is not a string. This can only happen when data comes in from API calls (not CLI) or is loaded from default answers in `copier.yml`. In these cases, we don't need to do any type casting, because the input is already a scalar.
@Tecnativa TT20357
Fix#89 and fix#89 by adding a new extended syntax to make prompts way more usable.
Leverages plumbum's helpers, so prompt code is removed from Copier codebase, making it smaller BTW.
See README and the new test to understand the new features.
@Tecnativa TT20357
