Roadmap for `v1.0.0`
I reviewed all the configuration options which are used in the code and found quite a few inconsistencies. I have listed proposed fixes for them, separated into breaking and non-breaking changes. We can add warning for the breaking changes now, and then change in v1.0.0.
Let's discuss these lists. Are there any items which should be added or removed? For the breaking changes, how long of a warning period should we have? For any of them, should we only start warning in v1.0.0 and then change in v1.1.0?
Review of configuration options
| Option | kwarg | CLI | ini | default |
|---|---|---|---|---|
| Enable testing | N/A | --mpl |
N/A | False |
| Enable baseline image generation, at specified directory path (relative to where pytest was run) | N/A | --mpl-generate-path |
N/A | None |
| Enable baseline hash generation, at specified file path (relative to where pytest was run) | N/A | --mpl-generate-hash-library |
N/A | None |
| Directory containing baseline images (relative to the test file) |
baseline_dir |
--mpl-baseline-path (relative to where pytest was run) |
baseline/ |
|
Whether --mpl-baseline-path should also be relative to the test file |
N/A | --mpl-baseline-relative |
N/A | False |
| Filename of the baseline image | filename |
N/A | N/A | test name |
| Whether to include the module name in the baseline image filename | ? | mpl-use-full-test-name |
False |
|
| File containing baseline hashes (relative to the test file) |
hash_library |
--mpl-hash-library |
no hash comparison | |
| RMS tolerance | tolerance |
2 |
||
| Whether to standardise metadata | deterministic |
True (PNG: False) |
||
| kwargs to pass to savefig | savefig_kwargs |
N/A | N/A | {} |
| Matplotlib style | style |
classic |
||
| Whether to remove axis tick labels | remove_text |
False |
||
| Matplotlib backend | backend |
agg |
||
| Directory to write testing artifacts to (relative to where pytest was run) |
N/A | --mpl-results-path |
mpl-results-path |
temp dir |
| Whether to save result images for passing tests | N/A | --mpl-results-always |
mpl-results-always |
False (True if generating a HTML summary) |
| Which test summaries to generate, if any | N/A | --mpl-generate-summary={html,json,basic-html} |
None |
Notes
- N/A are options which we don't think should exist.
- Empty boxes are options which we do think should exist.
- ? are options which could exist but would be of limited use.
- Due to how paths are computed by Python, the baseline directory and hash library paths can be absolute. In this case
--mpl-baseline-relativeand what directory the paths are interpreted to be relative to does not have any effect. - Path specified in kwargs should be relative to the test file, and paths specified in CLI and ini options should be relative to where pytest was run.
Non-breaking changes
- [x] Document that backend can be specified. #199
- [x] Document the option to use the full test name. #199
- [x] Document that
mpl-use-full-test-nameini overrides thefilenamekwarg. #199 - [x] Add
mpl-baseline-pathini option (relative to where pytest was run). #181 - [x] Expand
mpl-use-full-test-nameini option to CLI (and maybe kwarg). #181 - [x] Add
mpl-hash-libraryini option (relative to where pytest was run). #181 - [x] Add
--mpl-default-toleranceCLI andmpl-default-toleranceini options. #181 - [ ] Add CLI and ini options for
deterministic. #197 - [x] Add
--mpl-default-styleCLI andmpl-default-styleini options. #181 - [ ] Add CLI and ini options for
remove_text. - [x] Add CLI and ini options for
backend. #181 - [x] Add
generate-summaryini option for generating test summaries. #181
Breaking changes
- [ ]
--mpl-hash-libraryshould be relative to where pytest was run (to match--mpl-baseline-path). - [ ] Local
hash_librarykwarg should take precedence over global--mpl-hash-libraryCLI option. #154 - [ ] Deprecate the use of multiple hash libraries in the same pytest run (not necessary because files are small and include the full module path anyway). #154
- [ ] Deprecate the use of multiple testing modes in the same pytest run (e.g. warn if one test only has hash comparison while the others only have image comparison)
- [ ] Use Matplotlib's default style instead of
'classic'. - [ ] Change the default RMS tolerance to
0. - [ ] Enable deterministic PNG files by default.
- [ ] Deprecate
remove_textkwarg and replace with better namedremove_ticks_and_titles - [ ] (???) Add a
remove_textkwarg that actually removes all text. - [ ] (???) Rename
baseline_dir/--mpl-baseline-pathto have consistent name.
@ConorMacBride This is brilliant! Thanks for putting this together.
I'm spinning back up on pytest-mpl. I've cleared the decks and pretty much secured some free time to dedicate to finally getting #150 across the line.
In the process of getting my head back in the space I'm going to cross-reference your table above with the codebase and also any observations that I previously made to make sure that we've captured everything up for discussion in the v1.0.0 roadmap 👍
@ConorMacBride The only additionally changes that I've noticed not captured above are:
- The
baseline_dirkwarg takes precedence over the--mpl-baseline-path. The order elsewhere is CLI -> INI -> KWARG. Although, this may have changed in #181, I've not checked yet. Was this intentional by design? - #150 will introduce a hashing tolerance, so we need to be able to identify the existing
toleranceas the RMS tolerance e.g.,--mpl-rms-tolerance(CLI),mpl-rms-tolerance(INI) andrms_tolerance(kwargs) or just stick with the changes from #181 i.e.,--mpl-default-tolerance(CLI),mpl-default-tolerance(INI) andtolerance?