GitLab

Previously, the `-no-user-package` and `-no-global-package` flags
affected the "initial" stack only, while `user-package` and
`global-packages` appended to the end of the stack.

This commit changes the behavior of those flags, so that they are always
applied to the stack as a whole.

The effect of the GHC_PACKAGE_PATH environment variable has also been
changed: terminating it with a separator now adds the default package
dbs (user and global) instead of the initial stack.

Rename package database flags in both GHC and ghc-pkg so that they are
consistent with Cabal nomenclature.

Add a version check to the build system so that the correct set of
package db flags are used when the bootstrapping GHC has version < 7.5.

Introduce new flags to allow any package database stack to be set up.
The `-no-user-package-conf` and `-no-global-package-conf` flags remove
the corresponding package db from the initial stack, while
`-user-package-conf` and `-global-package-conf` push it back on top of
the stack.

We were hitting a problem when reading the LANGUAGE/OPTIONS pragmas
from GHC.TypeLits, where the buffer ended "{-". The rules for the
start-comment lexeme check that "{-" is not followed by "#", but the
test returned False when there was no next character. Therefore we
were lexing this as as an open-curly lexeme (only consuming the "{",
and not reaching the end of the buffer),
which meant the options parser think that it had reached the end of
the options.

Now we correctly lex as "{-".

Surround a name in backticks when printing an infix declaration in GHCi.

When unifying two type variables we must unify their kinds.
The pure *matcher* was doing so, but the pure *unifier* was not.
This patch fixes Trac #6015, where an instance lookup was failing
when it should have succeeded.

I removed a bunch of code aimed at support sub-kinding. It's
tricky, ad-hoc, and I don't think its necessary any more.
Anything we can do to simplify the sub-kinding story is welcome!

This is really a small change, but it touches a lot of files quite
significantly. The real goal is to put the implicitly-bound kind
variables of a data/class decl in the right place, namely on the
LHsTyVarBndrs type, which now looks like

  data LHsTyVarBndrs name
    = HsQTvs { hsq_kvs :: [Name]
             , hsq_tvs :: [LHsTyVarBndr name]
      }

This little change made the type checker neater in a number of
ways, but it was fiddly to push through the changes.

Fixes cgrun045(ghci) amongst others

* Undoes Max's very aggressive function-inlining change
  (see comments with Trac #6048)

* Resticts function application discount to functions
  that occur just once in the body. It was the multiple
  occurrences that led to the exponential behavour in
  Trac #6048.

See Note [Function application discount] in CoreUnfold.

Module binary sizes are down 2% on average, which is good.
Allocations wobble about a bit, but only on a few benchmarks
and not by much, so it seems a price worth paying to avoid
exponential behaviour!

                         Allocs
            Min           -1.2%
            Max           +2.8%
 Geometric Mean           +0.0%

Fixes Trac #6078

* Put the result type in the Stop continuation
* No need for the alts type in Select

The result is a modest but useful simplification

This turns out to be important becuase we don't have
a form for superclass selection in TcCoercion (we could
but we don't).

Se comments with xCtFlavor_cache, the Given case.

This is kosher, and turns out to be vital when we have
more complicate evidence terms.

When inlining, we are making a copy of the expression, so we have to
be careful about duplicating work.  Previously we were using
exprIsCheap for that, but it is willing to duplicate a cheap primop --
and that is terribly bad if it happens inside some inner array loop
(Trac #5623).  So now we use a new function exprIsWorkFree.  Even
then there is some wiggle room:
   see Note [exprIsWorkFree] in CoreUtils

This commit does make wheel-sieve1 allocate a lot more, but we decided
that's just tough; it's more important for inlining to be robust
about not duplicating work.

We get these in a full build on Win64

This is the result of Simon and Dimitrios doing a code walk through.
There is no change in behaviour, but the structure is much better.
Main changes:

* Given constraints contain an EvTerm not an EvVar

* Correspondingly, TcEvidence is a recursive types that uses
  EvTerms rather than EvVars

* Rename CtFlavor to CtEvidence

* Every CtEvidence has a ctev_pred field.  And use record fields
  consistently for CtEvidence

* The solved-constraint fields of InertSet (namely inert_solved and
  inert_solved_funeqs) contain CtEvidence, not Ct

There is a long cascade of follow-on changes.

I'm not sure if we still need to encode the error, but validate is happy
with this fix, at least.

On Win64, "install -s" calls a strip that doesn't understand 64bit
binaries. For some reason, this means the DLLs end up non-executable,
which means executables that use them just segfault.

Otherwise the dynHelloWorld test segfaults when run from an installed
compiler on cygwin.