GitLab

Add -static to the linker command line when -static is passed to GHC.

At some point we might want a way to say that Haskell libraries should
be linked statically but everything else should be linked as normal,
but since we aren't really supporting dynamic Haskell libraries at
this time, we don't need to address that yet.

More informative crash in rdrNameModule

Documentation for #size

add `size` macro

include stdlib.h (to get at NULL)

Always return 8-byte-aligned memory from arenaAlloc().  Fixes problems
with profiling on sparc-sun-solaris2, and might fix problems with
Windows too.

MERGE

interactiveLoop{mingw32}: always show prompt under mingw.

hIsTerminalDevice returns True iff stdin is connected to a Console
under Win32. Which it is when running in a command shell
(DOS,bash etc.), but Emacs sub-shells don't get a Console of their
own when invoked. So, just disable this feature (whose motivation
somewhat escapes me.)

Expose printAllThreads() in non-DEBUG builds;
of some use when debugging ConcHask code and
want a dump of the current state of the system's
threads.

Not provided via a library, so to use, do

 foreign import call unsafe printAllThreads :: IO ()

Add -ddump-minimal-imports (not sure why this was missing)

MERGE

warning removal

Documentation for #offset

Add #offset for just grabbing the byte offset of a struct field.

add GhcLibsWithObjectIO

awaitEvent: if select() reports EBADF,
always unblock all waiting threads, even if
the prior invocation of select() also
elicited an EBADF.

The 'smart' that was there previously runs
the risk of working against us (if the EBADFs are
coming from different fds), so the above
scheme is preferable (and simpler.)

add --enable-objectio configuration option

Change "rm" in example makefile to "rm -f",
in case there is no .o file to remove

Use tcLookupId, not tcLookupGlobalId, in Inst.newMethodFromName

    This really only affects the behaviour with -fno-implicit-prelude

	*** MERGE TO STABLE BRANCH ***

a)	Correct precedence for application in derived
	Read/Show

  b)	Spaces round '=' in derived Show for records

	*** MERGE TO STABLE BRANCH ***

find ~=> $(FIND)

if $(STAMP_PKG_CONF) is non-null, then wait for it before generating
dependencies.  This avoids a race with 'make -j' where package.conf
could get modified while the compiler is reading it.

Don't re-run ghc-pkg if the package conf files haven't changed.  We
use a stamp file rather than adding an explicit dependency on
ghc/driver/package.conf{,.inplace} for reasons explained in a comment.

Use $(HADDOCK) rather than $(HADDOCK_INPLACE)

Add $(HADDOCK) pointing to the Haddock to use (src tree or installed)

- Invoke FPTOOLS_HADDOCK to find an installed Haddock
- Add an option --use-src-tree-haddock to mirror --use-src-tree-happy

Add FPTOOLS_HADDOCK to detect an installed version of Haddock.

Last bunch of OpenBSD-implicated autoconf-changes
from Donald Bruce Stewart <dons@cse.unsw.edu.au>.

Add a note about pattern-matching with -fno-implicit-prelude

Add a comment to note that Stg.h must be included before system
headers (this was always the case, but is now more important because
it is often necessary to #include system headers where it wasn't
previously).

Long explanation for small commit:

With the recent reorganization of #includes, things get a little bit
fragile: When gcc with -O is used, glibc's <stdio.h> defines a few
inline functions (see <bits/stdio.h>), and "Stg.h" defines some global
variables which reside in registers. But the latter must happen
*before* any function definition has been seen, otherwise the
generated code could be invalid. Consequently gcc complains like:

   In file included from Stg.h:182,
                    from mkNativeHdr.c:12:
   Regs.h:293: global register variable follows a function definition
   Regs.h:302: global register variable follows a function definition

The solution is quite simple: Always #include "Stg.h" *before* any
system headers.

OpenBSD needs #include <sys/types.h> for msghdr compile tests.

forkProcess():
 be less ambitious, and don't kill any main_threads.
 Killing them, and possibly replacing them on the
 main_threads with the current one, turned out to
 run into problems of its own (see comments.)

 The hope/wish is that the current behaviour of
 forkProcess will suffice.

Remove the check for >4 args when doing a foreign import "wrapper" on
Sparc: we support unlimited arguments now (from Sven Panne).

Code to make the general case of createAdjustor() work on Sparc, from
Sven Panne.  Previously only 4 args were supported.

Sparc NCG fix from Sven Panne: 8-byte align the stack ptr when doing a ccall.

forkProcess():
 - fix bug which left run_queue_tl in a bad state.
 - be better behaved wrt 'main threads', i.e.,
   if the killing thread isn't the main thread,
   make sure it's hooked up to main_threads +
   correctly signal the completion/killing of
   any main threads.

%files for the profiling rpm was missing .p_hi files that were nested deeply.
Fixed by calculating %files for prof rpm similar to how it is calculated
for the main rpm.

Back off from including the interface file version in the module init
label - we might not recompile modules which depend on the current
one, even if its version changes.  Thanks to Sigbjorn for pointing
this out.

We still include the way, however, so we'll still catch cases of
linking modules compiled in different ways.

Remove a bogus test

#include wibbles

cosmetic only