BF: cmdline: Restore argcomplete functionality #4477
Conversation
The optimization in fb38c70 (OPT/RF: create only the subparser needed for the command, offer "closest" choices, 2018-05-17) sped up datalad, but it (would have [*]) broken the argcomplete-based command-line completion due to the spots in setup_parser() that raise an exception rather than returning the parser that argcomplete needs to inspect. To make completion work again, add a flag to setup_parser() that signals that datalad is being called by argcomplete. In this scenario, avoid failing early so that argcomplete can do its thing. Closes datalad#4473. [*]: When I tried from fb38c70^, completion still didn't work because an exception is raised in config.py due to cfg_kv_regex not matching when datalad is run under argcomplete (for reasons I haven't looked into). But, when I worked around that error, _then_ fb38c70 broke completion. Note that the cfg_kv_regex matching has since changed (b733b04, 2020-04-17) in a way that avoids the config.py exception.
Codecov Report
@@ Coverage Diff @@
## maint #4477 +/- ##
==========================================
- Coverage 89.68% 89.68% -0.01%
==========================================
Files 275 275
Lines 37102 37103 +1
==========================================
Hits 33275 33275
- Misses 3827 3828 +1
Continue to review full report at Codecov.
|
|
Awesome, thank you @kyleam! I wondered if it would be feasible to craft a test (not necessarily in this PR, but while the track is warm) to sure that argcomplete is happy with us? |
Sure, I'll make an issue and assign it to you so you don't forget ;) |
|
Ha ha, my track is cold! But I guess I could try by setting that env variable and setting what is spit out |
|
I hope you'll sort out who will provide the test first. Meanwhile I will merge this and autocomplete all day. Thx much! |
|
Resolved conflict with master, merged, and pushed. |
The optimization in fb38c70 (OPT/RF: create only the subparser
needed for the command, offer "closest" choices, 2018-05-17) sped up
datalad, but it (would have [*]) broken the argcomplete-based
command-line completion due to the spots in setup_parser() that raise
an exception rather than returning the parser that argcomplete needs
to inspect.
To make completion work again, add a flag to setup_parser() that
signals that datalad is being called by argcomplete. In this
scenario, avoid failing early so that argcomplete can do its thing.
Closes #4473.
[*]: When I tried from fb38c70^, completion still didn't work
because an exception is raised in config.py due to cfg_kv_regex
not matching when datalad is run under argcomplete (for reasons I
haven't looked into). But, when I worked around that error,
then fb38c70 broke completion. Note that the cfg_kv_regex
matching has since changed (b733b04, 2020-04-17) in a way that
avoids the config.py exception.