GitLab

more Ptr/FunPtr documentation

Add documentation from the FFI spec.

Change the interface (but not the implementation) to match the FFI spec:

* added charIsRepresentable, *CAString and *CWString

* currently (and in violation of the spec), fooCString = fooCAString

* The WString versions use UTF-16 under Windows and UTF-32 elsewhere
  (valid only if the compiler defines __STDC_ISO_10646__).

Typo.

Yet another fix for the fact that CTime is not an instance of Integral anymore,

Note dependency of Foreign.C on Foreign.C.*

Add missing instances of Data.Bits.Bits for nhc98.

Solaris/nhc98 fixes: HAVE_TZNAME rather than HAVE_TM_ZONE, and need a
"time.h" specification on every foreign import.

Changes related to arithmetic types:

* Renamed macros NUMERIC_FOO to ARITHMETIC_FOO to match C99-speak

* ARITHMETIC_TYPEs now have a Real instance, otherwise they are quite useless.
  Note that this differs from the FFI spec, but the spec should very probably
  changed in this respect.

* Some changes to fix the wrong assumption that CTime/CClock are integral types,
  C99 in fact guarantees only that they are arithmetic types. This has been
  accomplished by using

     realToInteger = round . realToFrac :: Real a => a -> Integer

  instead of fromIntegral for CTime/CClock. I'm not sure if we could do better,
  going via Double seems to be overkill, but I couldn't think of a better way.
  GHC could e.g. use RULES here. Improvements welcome.

Added missing dependencies of Haskell files to header files they include

The FFI report does not mention that CClock and CTime have instances for
Bounded, Real, Integral, and Bits.

timezone fix by Antony Courtney

Add a comment about fdGetMode, which doesn't work properly on Windows

__hscore_PrelHandle_{send,recv}: make these mingw32-only

- Win32: when using the threaded RTS, bypass the Async IO stuff and
  just make blocking calls to read()/write().

  This gives a significant performance boost to programs doing lots of
  multithreaded I/O: in fact, a test program I have which does I/O
  over 500 pipes simultaneously goes twice as fast with this change,
  and is even faster than the non-threaded RTS (Windows only - Unix
  changes are in the pipeline too).

- openFd: take an extra parameter to specify socketness of the file
  descriptor rather than assuming that all streams are sockets.  Some
  streams (eg. pipes) aren't sockets.

Add __hscore_PrelHandle_{send,recv}

The lock arrays are too small on Windows, leading to buffer overruns
and crashes when a program opens too many files.

The problem is that on Windows, we shouldn't use FD_SETSIZE to get the
max number of file descriptors: this is set to 64 in the mingw
includes.  The real maximum is 2048 (according to the crt sources), so
we now hardwire that in.

Also, put in a runtime check that we aren't overruning this array.

MERGE TO STABLE

Inline the default method for newArray; big perf boost; comments with the pragma

Small performance hack in maxBound::Word.

Remove spaces around the "%" when showing ratios.

This is not absolutely strictly according to the H98 report,
but it was suggested by Doug McIlroy

   According to the standard prelude, a list of Ratios prints like
   this: [1 % 1,1 % 2,1 % 3,1 % 4], which strongly suggests the
   grouping [1 % (1,1) % (2,1) % (3,1) % 4].

   Even standing alone outside of lists, the style is at odds with
   mathematical custom.  When did you last see 1/2 written as 1 / 2?

RawSystem inclusion belongs inside the __GLASGOW_HASKELL__ ifdef.

When .hc bootstrapping, don't build the Concurrent_stubs

Make the documentation and the error messages match the way the threaded
RTS is used nowadays
(the -threaded flag to ghc rather than ./configure --enable-threaded-rts)

MERGE TO STABLE

Move the definition of rawSystem into a separate file which we
#include in the places it is needed.  This is slightly better than
copying the code, since we now need it in three places
(ghc/utils/runghc is the 3rd).

Hugs: use the same version of fixIO as GHC

New version of fixIO which does eager blackholing.

- comments on INLINability of unsafePerformIO

- change unsafeInterleaveIO from NOINLINE to INLINE(!)  we believe
  this is safe.  Interestingly, there's now a good reason to use
  unsafeInterleaveIO.

Fix a file descriptor leak in openFile: if openFd fails, then we
weren't closing the newly created descriptor.

indent: don't use tab characters

Tiny fix in Haddock markup

Use consistent capitalisation

fix a warning

track Data/Generics/* twists

We decided that we want the gunfold primitive back.
This avoids some hassle with bottoms and strict datatypes.
The compiler now also derives gunfold.

Added withArrayLen and withArrayLen0

This package is named "base", not "core".

* Merged Martin Sj-A?gren's patch for multiline descriptions-b
* Nuked some TABs in favour of space + some small reformatting
* Updated copyright

Add caveat about finalizers (don't refer to Handles from finalizers).

For nhc98, at last include the Typeable macros which define the instance decls.

Implemented renaming for Data.Typeable according to
http://www.haskell.org//pipermail/libraries/2004-March/001846.html