GitLab

support for building .pdf docs

Give slightly more accurate line numbers for certain pattern parse errors.

add
	setSrcLocP :: SrcLoc -> P a -> P a

typo in Word64 code pointed out by Alastair Reid.

comment explaining why we check the perms on . before reading ./.ghci.

.raw_s is a haskellish_suffix.

document :set -package

- only read ~/.ghci if it is owned by the current user and isn't
  writable by anyone else.

- Only read ./.ghci if both . and ./.ghci are owned by the current
  user and aren't writable by anyone else.  I think this is
  sufficient: we don't need to check .. and ../.. etc. because "."
  always refers to the same directory while a process is running.

- Don't load .so libraries in a package if that package is already
  linked with GHCi.  This stops us re-linking libm, libreadline etc.

- Allow packages to be loaded from within GHCi using
	:set -package <name>
  NOTE: this will unload all modules currently loaded into the
  interpreter.  I did this to be on the safe side - I think perhaps
  it isn't necessary, but I haven't thought it through fully yet.

- fix CompManager.cmUnload in the process.  It was wrong in several
  ways.

MERGE WITH 5.00

Add mutual recursion between types test

Add more TC tracing

touch# should be declared to have_side_effects.  This fixes the
remaining problem with safe foreign imports and ForeignPtrs.

****	MERGE WITH 5.00 BRANCH     ********

	--------------------------------
	Fix a black hole when type checking type decls
	--------------------------------

GHC was falling into a black hole when type checking a recursive
group of type declarations including a chain of type synonyms.

  type PhraseFun = PMap -> Float
  type PMap      = () -> Player
  data Player    = P.MkT P.PhraseFun

Reason: too much consistency checking in TcMonoType.
Easily fixed using the existing wimp_out hack, but it's a mess.
This commit fixes it for the 5.00 branch but I'll do something
better in the head shortly.

Comments only

Script to change "cygwin1.dll" to something else, so that we can have our
own version of the cygwin DLL that doesn't clash with any installed DLLs,
for great justice.

add
	flushPackageCache :: [PackageConfig] -> IO ()

for modifying the list of available packages.

Since overlapping rules (when one is more specific than the other)
seem to be handled well, add remaining fromIntegral rules to have
optimal conversions in all cases (given available primops).

Comments only

****	MERGE WITH 5.00 BRANCH     ********

	--------------------------------
	Make CPR work only for small products
	--------------------------------

GHC was barfing when doing CPR for programs involving very large
products.  A one-line fix in MkId.mkDataCon makes it only do CPR for
smaller products (I chose 10).  Comments

-- We do not treat very big tuples as CPR-ish:
--	a) for a start we get into trouble because there aren't
--	   "enough" unboxed tuple types (a tiresome restriction,
--	   but hard to fix),
--	b) more importantly, big unboxed tuples get returned mainly
--	   on the stack, and are often then allocated in the heap
--	   by the caller.  So doing CPR for them may in fact make
--	   things worse.

This should fix the error
	Failed to find interface decl for
		PrelGHC.(#,,,,,,,,,,,,,,,,,,,,,,,,
			  ,,,,,,,,,,,,,,,,,,,,,,,,,,,
			  ,,,,,,,,,,,,,,,,,,,,,,,#)

I hope.

win32 tweak

Use infix syntax on lhs of some rules.

silence gcc 2.96 warning

oops, fix braino

redirect CPP output to /dev/null; gcc 2.96 seems to behave differently
that 2.95 when the -E and -MMD flags are given.

Arrange to call touch# on each ForeignObj# argument after a 'safe'
foreign import call.

This turned out to be the easiest place to implement it: if we'd done
it in CorePrep or CoreToStg, it would have been awkward to generate
the simplest code.  At least doing it in the desugarer gives the
simplifier a crack at it later.

****	MERGE WITH 5.00 BRANCH     ********

	--------------------------------
	Monomorphism restriction for implicit parameters
	--------------------------------

This commit tidies up the way in which monomorphic bindings
are dealt with, incidentally fixing a bug to do with implicit
parameters.

The tradeoffs concerning monomorphism and implicit paramters are
now documented in TcSimplify.lhs, and all the strategic choices
are made there (rather than in TcBinds where they were before).

I've continued with choice (B) -- which Jeff first implemented --
because that's what Hugs does, lacking any other consensus.

Add another IP test

Oops, xrefs to varlistentries are unsupported apparently.  Xref the
section instead.

Correct pretty print of HsIPVar

mention, with a link, the +RTS -xc option.

document the +RTS -xc option

------------------------------------------------
	Dramatically improve the error messages arising
	from failed unifications triggered by 'improvement'
	------------------------------------------------

A bit more plumbing in FunDeps, and consequential wibbles elsewhere

Changes this:

    Couldn't match `Int' against `[(String, Int)]'
	Expected type: Int
	Inferred type: [(String, Int)]

to this:

    Foo.hs:8:
	Couldn't match `Int' against `[(String, Int)]'
	    Expected type: Int
	    Inferred type: [(String, Int)]
	When using functional dependencies to combine
	  ?env :: Int, arising from a type signature at Foo.hs:7
	  ?env :: [(String, Int)],
	    arising from use of implicit parameter `?env' at Foo.hs:8
	When generalising the types for ident

Improve error message when two signature tyvars get unified

Error message wibble

Tidy up output from -ddump-types

Improve error message

Wibble

disable the hsc2hs suffix rules when booting from .hc files.

Add a fundep fail test

Remove bogus test

Add implicit param test