GitLab

for efficiency. More documentation to follow.

We now always have base4, and don't need a workaround for GHC 6.4.2.

Signed-off-by: Paolo Capriotti <p.capriotti@gmail.com>

This is motivated by the fact that before this change marking e.g.
GHC.List.filter as INLINABLE caused the compiler to diverge when
you tried to make use of the function.

The response is to say that a RULE on an imported function introduces
a dependency edge between the FVs of its LHS and RHS for the purposes
of computing loop breakers. This will not perfectly prevent all those
potential inlinings that could cause the compiler to non-terminate,
but it works well enough for the particular case we are interested in.

We might be in the "knot" of a type declaration: see Trac #5968.

variables work in Template Haskell

Triggered by fixing Trac #5968.

Fixes Trac #5951

Add support for ELF objects on ARM to the runtime linker. While the list of
relocation types in the "ELF for the ARM Architecture" specification spans four
pages, we thankfully only see a handful of these in the wild. Thus, at the
moment we only support the following .rel relocation types,

 * R_ARM_ABS32
 * R_ARM_TARGET1
 * R_ARM_REL32
 * R_ARM_CALL
 * R_ARM_JUMP24
 * R_ARM_MOVT_ABS
 * R_ARM_MOVW_ABS_NC
 * R_ARM_THM_CALL
 * R_ARM_THM_JUMP24
 * R_ARM_THM_MOVT_ABS
 * R_ARM_THM_MOVW_ABS_NC
 * R_ARM_THM_JUMP8
 * R_ARM_THM_JUMP11
Signed-off-by: Ben Gamari <bgamari.foss@gmail.com>

More informative commentary as to why we allow family interactions with inerts when simplEqsOnly is on.

pretty much as we do for mkTcEqPred. Ugly and delicate but we have to rething the subkinding
story. For now I am doing the same thing as mkTcEqPred.

inert sets do not tolerate more than a single constraint per family head
we have to allow family interactions /with the inerts/ (not with top-level)
when the context says simplEqsOnly.

          (ii) evidence loops because of caching fixed

If the old stack is only half full, then the next chunk we allocate
will be twice as large, to accommodate large requests for stack.  In
that case, make sure that the chunk we allocate is at least as large
as the usual chunk size - there's no point allocating a 2k chunk
(double the default initial chunk size of 1k) if in the normal case we
would allocate a 32k chunk.

If we overflow the current stack chunk and copy its entire contents
into the next stack chunk, we could end up with two UNDERFLOW_FRAMEs.
We had a special case to catch this in the case when the old stack
chunk was the last one (ending in STOP_FRAME), but it went wrong for
other chunks.

I found this bug while poking around in the core dump attached to
options and running the nofib suite: imaginary/wheel_sieve2 crashed
with +RTS -kc600 -kb300.

I don't know if this is the cause of all the symptoms reported in

This allows us to give a sensible error message when a Template Haskell
splice generates an occurrence without a binding site.

Fixes Trac #5959 and #5971

The absence of this was causing a loop when typechecking an interface
where the superclass context mentioned an associated type
   class C (T a) => D a where
     data T a

Fixes Trac #5970

No reason user should turn them off, just want for ghc developer
debugging uses.

   (i) Replaced a lot of clunky and fragile EvVar handling code with
       a more uniform ``flavor transformer'' API in the canonicalizer
       and the interaction solver. Now EvVars are just fields inside
       the CtFlavors.
   (ii) Significantly simplified our caching story
This patch does not validate yet and more refactoring is on the way.

A Forest is just a list, and we have a Show instance for
lists already.  We don't need a special Show instance for
Forests.  If we change our minds we can add a newtype,
but I don't think it's worth it.

Fixes Trac #5960