GitLab

Wibble...

The corrected example is as follows:

myS :: (forall t t1 t2. (t -> t2 -> t1) -> (t -> t2) -> t -> t1)
[NoDiscard] __AL 3
myS
  = \ @ t @ t1 @ t2 f :: (t -> t2 -> t1) g :: (t -> t2) x :: t -> f x (g x)

public class myS implements Code {
  public Object ENTER () {
    VM.COLLECT(3, this);
    final Object f = VM.POP();
    final Object g = VM.POP();
    final Object x = VM.POP();
    VM.PUSH(x);
    VM.PUSH(new Thunk(new myS$1(g, x)));
    return f;
  }
}
class myS$1 extends Code {
  final Object g;
  final Object x;
  public myS$1 (Object _g_, Object _x_) {
    g = _g_;
    x = _x_;
  }
  public Object ENTER () {
    VM.PUSH(x);
    return g;
  }
}

First attempt at at class lifter for the GHC GOO backend.

This included a cleanup of the Java/GOO abstract syntax
  - Name is now a string, not a list of string
  - Type is used instead of name in some places
      (for example, with new)
  - other minor tweeks.

Andy

---------
Example for myS f g x = f x (g x)

public class myS implements Code {
  public Object ENTER () {
    VM.COLLECT(3, this);
    final Object f = VM.POP();
    final Object g = VM.POP();
    final Object x = VM.POP();
    VM.PUSH(x);
    VM.PUSH(new Thunk(new Code(g, x)));
    return f;
  }
}
class myS$1 {
  final Object g;
  final Object x;
  public myS$1 (Object _g_, Object _x_) {
    g = _g_;
    x = _x_;
  }
  public Object ENTER () {
    VM.PUSH(x);
    return g;
  }
}

Reinstate the bytecode peephole optimiser.

findFilesForModule: search default library directories before looking
at user-supplied search paths.

More RULES for coercions and truncate.

Temporarily removed i018 test on mingw

Omitted some tests on mingw.

Omitted some tests on mingw (conc020 *should* work, but doesn't).

compile CType.lhs via-C, since the NCG's register allocator seems to
behave suboptimally when presented with the huge switch in this file.

Use the native code generator by default for compilations without -O.

Add tryTakeMVar test.

Removed LITERATE, added DOCBOOK_PREFIX, and improved some comments.

DocBook scripts from Cygnus DocBook tools; the first visible sign of freedom
from the tyranny of RPMs.

Tried to bring this test into the modern age. Failed. I've committed the
pieces to be picked up later.

Put back the '$(RM) $@' in the rule for building library modules when
splitting, until we fix this problem properly.

Add comment noting that StgLint is currently broken.

rename takeMaybeMVar to tryTakeMVar

Undo revision 1.34 of Prelude.hs, ie, reinstate fromDouble in class
Fractional (in standalone mode).  This reduces by a factor of 6 the
number of enters required for some simple expressions,
for example    1.0 :: Double.

Updated to cope with updated test.

Added RULES for realToFrac with types
   Double   -> Float
   Float    -> Double
   Double   -> Double
   Float    -> Float
   Rational -> Rational

This was unaccountably empty before.

Allow the c-t storage manager to reuse dead symbol table slots, as had
always been intended, but up to now has been disabled due to ultra-paranoid
debugging.

Zap CRUDE_PROFILING.  It was there mainly to test assess the effect of
the simplifier; is redundant.

Hack to work around bug in pre-4.06 mkdependHS.  Fixes last night's
nightly build failure.

Removing the duplex #include "Storage.h"

Removed info and texinfo support.

Used identifiers rather than paths to find the print and HTML stylesheets.
N.B. This means that building docs won't work unless you set
SGML_CATALOG_FILES to
/usr/lib/sgml/CATALOG:/usr/lib/sgml/stylesheets/nwalsh-modular/docbook.cat,
or wait for this lot to be added to the build system (RSN).

Get package names right.  Again.

Point out in the comment for GhcLibHcOpts that it also applies to the
standard library and hslibs.

Changed Perl test so that it simply allows vv5 & 6 and disallows any other;
also hardwire path to perl to /bin/perl under Windows, to avoid clashing with
other weird Perls that may be on the path.

Add -DNO_REGS to SRC_CC_OPTS.

Record details of the fptools/mk/build.mk required to build Hugs.

Fix various compile-time warnings in the assembler and evaluator.

Corrected a typo (omitted space).

Undefine NO_REGS.  Seems unneccessary since the driver supplies it anyway
when compiling the RTS, and this duplicate defn generates a lot of warnings.

Reorder tests in whatIs() to try for most common options first.

Typo: changed ($Var) to $(Var) in test to set SplitObjs.

Committing the Makefile, too, might be a good idea: Removed ctypes.

* Changed a backslash at EOL to the word "backslash". Using cpp in
  conjunction with Haskell is a real hack...

* Renamed the module Ctypes to Ctype, because CTypes is (soon) a
  module from package lang. I'm not sure if two files differing only
  in their case work under "the" OS. Just to be sure...

Detect threads that are blocked on themselves (i.e. black holed), and
raise NonTermination exceptions.  We only do this when the system is
deadlocked, so as not to introduce unnecessary overhead in the normal
case.