Problem was regexes treated `[,(` and `],)` equally, considering eg
`[..)` balanced.
Considering `()` is also used within titles without signifying tags, it
might be worth investigating only matching that kind of bracketed tag at
the end of the title.
Closes: #783
* main: Replace return by explicit calls to exit()
In anticipation of breaking out these sections in their own functions
* Make doc-getting flags shortcut
This saves on pointless validation/setup work when only getting help
information. Moreover, these were the only actions that were in the
middle of the parse/validate/setup/run core logic of main(), moving them
out clears the way to cleanly breaking it up.
Removes -v alias for --version. If this is undesirable, a trick similar
to that for --sites-list can be used to shortcut it as well.
* Move up flag implication logic, var renaming
These are "virtual flags", should be set up asap after actual flags are
setup. Ideally, these would be set up in arg parsing, but this is
sometimes impossible/impractical.
Future improvement: use callbacks to say eg --updatealways sets
options.update, options.updatealways
* Move up validation
Fail fast if the arguments are invalid
* Internalize list_only into options
Helps keep related state together
* Pack up configs, printers for easier passing
* Break up main() into phases
* Remove unnecessary semicolon
* Unbundle configs, printers
This reverts commit 5dd44bbfc3.
Revertion reasons:
1) Initial commit was broken -- it reordered parameters in invocations
to `get_config()`. This happened because python complained about
invocations of the form `f(x,**d,z)` -- positional parameters may not
appear after kwarg expansion. I mistakenly believed kwarg expansion
would consume the slots of those parameters, and so this code would
be equivalent to `f(x,z,**d)`. Instead, this passes `z` to the second
positional parameter, which luckily enough had a key contained in `d`
so it only caused a TypeError over the multiple values for that
parameter.
2) To maintain the vision of the original commit would require multiple
pessimizations *over* the previous state. Specifically:
1) Using our running example of invocations of the form `f(x,**d,z)`,
we'd need to turn `z` into a keyword argument. Since Python has no
way of writing "`z` is a keyword argument whose value is taken
from the current scope", that forces writing `f(x,**d,z=z)`.
(Even if a proposal like <https://lwn.net/Articles/818129/> is
accepted, we wouldn't be able to use it since we need to support
Python 2)
2) `dispatch()` uses `fail` internally. So we have one of two
options:
* Bundle `warn, fail` in `dispatch`'s arguments, and add a line
`fail=printers['fail']` to the top of `dispatch`
* Don't bundle `warn, fail` in `dispatch`'s arguments, and have
`dispatch` bundle them instead
Neither of these is palatable, especially over
* Don't bundle `warn, fail` anywhere
* Restore -v alias for version
As 0847fc9 suggested might be desired
