Both record construction and record layouting have very similar
constructions. These each had their own layouter with slightly different
variations. Variations here lead to subtley different bugs in layout for
nearly identicle syntactic forms.
The record update logic is more advanced and respects `IndentPolicyLeft`.
Instead of keeping these layouters distinct we can consolidate
construction logic into the update logic. This results in a smaller
volume of code and more uniform layouting of syntax for these simlilar
forms.
Record constructors with fields and wildcards are not included in this
consolidation. A TODO has been left to handle this consolidation later.
This fixes the code-path for the commandline interface
(unfortunately not covered by our testsuite).
The other code-path should be covered by the changes in
https://github.com/alanz/ghc-exactprint/pull/68.
Previously, we could only import a type operator with no subsequent
list, i.e.
import Foo ( (:.) )
was fine, but
import Foo ( (:.)(..) )
import Foo ( (:.)((:.) )
import Foo ( (:.)(A, b) )
would all break. Brittany would attempt to output them as
import Foo ( :.(..) )
import Foo ( :.((:.) )
import Foo ( :.(A, b) )
I believe the problem was that although `ieName <$> lie` was returning
an `IEWrappedName` with the same contents as used in `layoutWrapped`,
it had different location annotations; and the parentheses are
apparently saved in the location annotations.
During alt-transform, when gather spacings, previously
we tracked different non-bottom spacings separately even
though they would be treated in the same way during any
future transformations (apart from certain exceptions that
don't practically give better results). Instead we now
merge such spacings into one, giving more space for other
spacings when pruning to the spacings limit.
Sergey Vinokurov
Lennart Spitzner
Evan Rutledge Borden
Alan Zimmerman
Doug Beardsley
Phil Hazelden
Robert
Christian Kjær
Adam Bergmark
Leon Schoorl
Bryan Richter