GitLab

doh

escape GHC_PKG_VERSION double quotes on all platforms when compiling with <=6.0.x

Trimmed a few function checks

Beautified test for ghc-pkg

Aesthetic fix: Redirect error messages for non-GNU ar to /dev/null

all target: if building GMP, only descend into gmp/ when way=""

Better tags generation, avoiding derived files (still a bit ad hoc)

Oops: always qualify when printing code

Updated BuildRequires tags. Alas, there seems to be no real standard here, so
your mileage may vary... At least the current specs should work on SuSE Linux.

SGML is dead, long live DocBook XML!

Note: The BuildRequires tags in the spec files are still incomplete
and the documentation about the DocBook tools needs to be updated,
too. Stay tuned...

-------------------------------
	Print built-in sytax right
	-------------------------------

Built-in syntax, like (:) and [], is not "in scope" via the GlobalRdrEnv
in the usual way.  When we print it out, we should also print it in unqualified
form, even though it's not in the environment.

I've finally bitten the (not very big) bullet, and added to Name the information
about whether or not a name is one of these built-in ones.  That entailed changing
the calls to mkWiredInName, but those are exactly the places where you have to
decide whether it's built-in or not, which is fine.


Built-in syntax => It's a syntactic form, not "in scope" (e.g. [])

Wired-in thing  => The thing (Id, TyCon) is fully known to the compiler,
		   not read from an interface file.
		   E.g. Bool, True, Int, Float, and many others

All built-in syntax is for wired-in things.

Fix recently-introduced improvement bug

Urk, since when has subtraction been associative.  /me dons the pointy
hat.

Add the :def . readFile example.

Add the location of the docbook XSL stylesheets on this RedHat 9 box
to the list of places to look.

wibble

tidy up EOL termination handling under mingw; as was, C-related dependencies were being accumulated in existing dependency files, not replaced

backquote GHC_PKG_VERSION defn for <= ghc-6.0.x; believe this is only needed under mingw

wibble

FP_READDIR_EOF_ERRNO: better use AC_DEFINE_UNQUOTED here

fix braino in previous commit: we should cast the function result to
the right type, not just assume (W_).

Fix typo that caused non -O to see unfoldings

Spelling mistake in comment

Pair of new macros for checking how dirent.h is implemented. Needed
to deal with versioning issues under mingw only, but won't do
any harm on other plats. The macros are:

- FP_READDIR_EOF_ERRNO
     checks what readdir() sets errno to upon reaching end of a directory.
     (that value is available as READDIR_ERRNO_EOF in mk/config.h). Up
     until recently, mingw's readdir() did set errno rather than just
     leaving it alone.
- FP_DIRENT_FLAT_LAYOUT
     checks how 'struct dirent' is defined; defines
     config.h:STRUCT_DIRENT_FLAT_LAYOUT to 1 if 'struct dirent' is terminated
     with a d_name array. Up until recently, mingw didn't define it as such.

To be hooked up to libraries/base/{include/HsBase.h,cbits/dirUtils.c} and
quite possibly merged to STABLE.

Fix an infinite loop in the cmm-optimiser in the native codegen, and
refactor: move isAssociativeMachOp into MachOp.

Terminate program if execPage fails, this is more honest and
simplifies things a bit.

Ooops, fixed previous commit...

stgMallocBytes never returns NULL

* Renamed __obscure_ccall_ret_code to obscure_ccall_ret_code, leading
  underscores are bad for user code.

* Let the code for obscure_ccall_ret_code live in dynamically
  allocated memory, just like the adjustors. This should fix problems
  on OpenBSD, which doesn't like executable code in the data section.

Removed a few "use of cast expressions as lvalues is deprecated"
warnings. Note that gcc is really pedantic nowadays, so we have to go
via a temporary to avoid "cast does not match function type" warnings.

If "hhc" is in the PATH, the "htmlhelp" target builds a *.chm file now. Note:
This is untested and "hhc" should really be detected via autoconf.

* Factored out labeling options (used by FO now, too)
* Cleaned up stylesheet handling

* Header cleanup
* Improved type of execPage
* Use prog_belch instead of fprintf(stderr, ...)

Perform initialisation of adjustor thunk layer at startup.

Moved createAdjustor and freeHaskellFunctionPtr to a header visible in
*.hc code.  The whole header layout is a little bit baroque, IMHO...

Hack to work around bug on Windows reported by Levent Erkok
<levent.erkok@intel.com>:

Prelude> let x=x in x
*** Exception: <<loop>>
Prelude> 3
Fail: thread blocked indefinitely

Details in the comment.

MERGE TO STABLE

Fixes

Simplify the "impossible branch" handling, and fix a bug in the
process.  CmmSwitch encodes the possibility of having impossible
branches (the destinations are Maybe BlockId rather than just BlockId)
so we don't need to encode impossible branches as dummy blocks
containing a jump to an impossible location (currently 0).

However, PprC and PprCmm weren't set up to cope with Nothings in a
CmmSwitch, so this commit fixes that too.

Give literal string labels a _str suffix, to make it less likely that
they'll clash with a symbol in scope in a C file.

Build fix