This commit avoids running into an incomplete record update warning
in the Hoogle backend of Haddock.

This was only noticed now, because incomplete record updates were broken
in GHC 9.6. Now that they are fixed, we have to avoid running into them!

Bartłomiej Cieślar authored 1 year ago and Ben Gamari committed 1 year ago

MR !10283 changes the alternatives for WarningTxt pass. This MR reflects those changes in the haddock codebase.

Signed-off-by: Bartłomiej Cieślar <bcieslar2001@gmail.com>

Associated data families were being given their own export DocStructureItems,
which resulted in them being documented separately from their classes in
haddocks. This commit fixes it.

Finley McIlwaine authored 1 year ago and Ben Gamari committed 1 year ago

Due to ghc!10469.

Finley McIlwaine authored 1 year ago and Ben Gamari committed 1 year ago

The NFData SourceText instance is now available in GHC

Handle mod_iface mi_warns now being IfaceWarnings

MR !10283 changes the alternatives for WarningTxt pass. This MR reflects those changes in the haddock codebase.

Signed-off-by: Bartłomiej Cieślar <bcieslar2001@gmail.com>

With hi-haddock as is, there is an awkward silence during the load operation.
This commit makes haddock use the default `batchMsg` Messager for progress
reporting, and makes the default GHC verbosity level 1, so the user can see
what GHC is doing.

hi-haddock for ghc 9.6

This section is a bit of a WIP due to the unstable nature of hi-haddock
and the lack of tooling supporting it, but its a good start.

Vladislav Zavialov authored 2 years ago and Matthew Pickering committed 1 year ago

Consider this type signature:

	kindOf :: forall {k} (a :: k). Proxy a -> Proxy k

Prior to this fix, the LaTeX backend rendered it like this:

	kindOf :: forall k a. Proxy a -> Proxy k

Now we preserve explicit specificity and kind annotations.

Vladislav Zavialov authored 2 years ago and Matthew Pickering committed 1 year ago

Prior to this change, atSign was defined as follows:

	atSign unicode = text (if unicode then "@" else "@")

Yes, this is the same symbol '\64' and not your font playing
tricks on you. Now we define:

	atSign = char '@'

Both the LaTeX and the Xhtml backend are updated accordingly.

Neither of these MRs are ready to land yet which causes issues with
other MRs which are ready to land and need haddock changes.

This section is a bit of a WIP due to the unstable nature of hi-haddock
and the lack of tooling supporting it, but its a good start.

* Add --trace-args flag for tracing arguments received to standard output
* Avoid recompiling due to changes in optimization flags
* Update users guide and changes.md

Link to Data.Tuple instead of GHC.Tuple.Prim

All diffs now boil down to the expected differences resulting from
declarations being reified from TyThings in hi-haddock. Surface
syntax now has much less control over the syntax used in the
documentation.

`lookupOccEnv` was used to resolve `OccNames` with warnings attached, but
it doesn't look in the record field namespace. Thus, if a record field
had a warning attached, it would not resolve and the warning map would
not include it. This commit fixes by using `lookupOccEnv_WithFields`
instead.

- Stop using kind `*` in html test sources
- Add TypeOperators where necessary to avoid warnings and future errors
- Rename some test modules to match their module names

Changes to GHC pretty printing code have resulted in some
differences to Haddock's LaTeX output.

- Type variables are printed explicitly quantified in the
  LinearTypes test
- Wildcard types in type family equations are now printed numbered,
  e.g. _1 _2, in the TypeFamilies3 test
- Combined signatures in DefaultSignatures test are now documented
  as separate signatures

Accept several changes in Hoogle tests:

Pretty printing logic no longer prints the `(Proxy (Proxy (...))`
chain in Bug806 with parentheses. Since this test was only meant
to verify that line breaks do not occur, accept the change.

`tyThingToLHsDecl` is called for class and data declarations, which ends
up "synifying" the type variables and giving unlifted types kind
signatures. As a result, type variables of kind `Type -> Type`
are now printed with kind signatures in Hoogle output. This could
be changed by manually drop kind signatures from class variables
in the Hoogle backend if the behavior is deemed unacceptable.

Sometimes subordinate declarations are exported separate from their
parent declarations (e.g. record selectors). In this case, a type
signature is cobbled together for the export item in `extractDecl`.
Since this type signature is very manually constructed, it may lack
kind signatures of decls constructed from `tyThingToLHsDecl`. An
example of this is the `type-sigs` Hoogle test.

Change `*` to `Type` in Hoogle test refs. I don't think this will
break Hoogle behavior, since it appears to not consider type
signatures in search. I have not fully verified this.