| Commit message (Collapse) | Author | Age | Files | Lines |
|
|
|
|
|
|
|
|
| |
Perf only change:
* avoid needlessly union-ing maps
* avoid synify-ing instances twice
Took this opportunity to add some docs too
|
|
|
|
|
|
|
|
|
|
|
|
| |
The old approach to fixing #469, while correct, consumes a lot of
memory. We ended up with a HUGE 'GblRdrEnv' in 'ic_rn_gbl_env'. However,
'getNameToInstancesIndex' takes that environment and compresses it down
to a much smaller 'ModuleSet'.
Now, we compute that 'ModuleSet' explicitly as we process modules. That
way we can just tell 'getNameToInstancesIndex' what modules to load
(instead of it trying to compute that information from the interactive
context).
|
|\ |
|
| | |
|
| |
| |
| |
| |
| |
| |
| |
| |
| | |
The GHC-side `getNameToInstancesIndex` filters out incorrectly some
instances because it is not aware of what modules are visible. On the
Haddock side, we need to pass in the modules we are processing.
On the GHC side, we need to check against _those_ modules when checking
if an instance is visible.
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
* Indicate source module of instances
Above instance, we now also display a link to the module where the
instance was defined. This is sometimes helpful in figuring out
what to import.
* Source module for type/data families too
* Remove parens
* Accept tests
|
| | |
|
|/
|
|
|
|
|
|
|
| |
The GHC-side `getNameToInstancesIndex` filters out incorrectly some
instances because it is not aware of what modules are visible. On the
Haddock side, we need to pass in the modules we are processing.
On the GHC side, we need to check against _those_ modules when checking
if an instance is visible.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
* Use Avails for export resolution
* Support reexported modules
* Factor out availExportItem
* Use avails for fullModuleExports
* Don't use subMap in attachInstances
* lookupDocs without subMap
* Completely remove subMap
* Only calculate unqualified modules when explicit export list is given
* Refactor
* Refine comment
* return
* Fix
* Refactoring
* Split avail if declaration is not exported itself
* Move avail splitting
|
|
|
|
| |
This reverts commit 3adf5bcb1a6c5326ab33dc77b4aa229a91d91ce9.
|
| |
|
|\ |
|
| |
| |
| |
| |
| | |
So we agree that reexported declarations which do not have subordinates (for example top-level functions) shouldn't have gotten fixities reexported according to the current logic. I wondered why for example Prelude.($) which is obviously reexported from GHC.Base has fixities attached (c.f. http://hackage.haskell.org/package/base-4.9.1.0/docs/Prelude.html#v:-36-).
The reason is this: In mkMaps we lookup all the subordinates of top-level declarations, of course top-level functions don't have subordinates so for them the resulting list is empty. In #644 I established the invariant that there won't be any empty lists in the subordinate map. Without the patch from #642 top-level functions now started to fail reexporting their fixities.
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
* Use new function getNameToInstancesIndex instead of tcRnGetInfo
There is some significant performance improvement in the ghc testsuite.
haddock.base: -23.3%
haddock.Cabal: -16.7%
haddock.compiler: -19.8%
* Remove unused imports
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
* Haddock support for bundled pattern synonyms
* Add fixities to bundled pattern synonyms
* Add bundled pattern synonyms to the synopsis
* Store bundled pattern fixities in expItemFixities
* Add test for bundled pattern synonyms
* Stop threading fixities
* Include bundled pattern synonyms for re-exported data types
Sadly, fixity information isn't found for re-exported data types
* Support for pattern synonyms
* Modify tests after #631
* Test some reexport variations
* Also lookup bundled pattern synonyms from `InstalledInterface`s
* Check isExported for bundled pattern synonyms
* Pattern synonym is exported check
* Always look for pattern synonyms in the current module
Another overlooked cornercase
* Account for types named twice in export lists
Also introduce a fast function for nubbing on a `Name` and use it
throughout the code base.
* correct fixities for reexported pattern synonyms
* Fuse concatMap and map
* Remove obsolete import
* Add pattern synonyms to visible exports
* Fix test
* Remove corner case
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
There is some performance improvement.
GHC compiler:
| version | bytes allocated | cpu_seconds
---------------------------------
| before | 56057108648 | 41.0
| after | 51592019560 | 35.1
base:
| version | bytes allocated | cpu_seconds
---------------------------------
| before | 25174011784 | 14.6
| after | 23712637272 | 13.1
Cabal:
| version | bytes allocated | cpu_seconds
---------------------------------
| before | 18754966920 | 12.6
| after | 18198208864 | 11.6
|
| | |
|
|/
|
|
| |
See https://ghc.haskell.org/trac/ghc/wiki/ImplementingTreesThatGrow
|
| |
|
| |
|
|\ |
|
| | |
|
| | |
|
| | |
|
| | |
|
|/
|
|
|
| |
This is made possible by the elimination of "silent superclass
parameters" in GHC
|
| |
|
|
|
|
| |
Fixes #383
|
| |
|
| |
|
|
|
|
|
|
|
| |
This should at the very least not abort when something weird happens. It
does feel like we should have a type that carries these errors until the
end however as the user might not see them unless they are printed at
the end.
|
|
|