GitLab

Notes on .exe files

Bale out a little earlier on renamer errors

Comments

merge rev. 1.78.2.7

merge rev. 1.33.8.1

merge rev. 1.19.2.1

Update the list of packages already loaded to match recent changes.

When a command-line evaluation returns an exception, revert to the old
interactive context, because the new context will bind 'it' to a name
that doesn't exist in the linker's symbol table.

fix logic in prev. commit

Re-instate the checking for the values of errno constants at configure
time.  The problem with doing it using foreign calls is simply that
this tickles a bad case in the code gen machinery, which in this case
results in an extra 10-20k of goop ending up in pretty much every
binary, and it impacts GC performance too.

This has some portability implications, but the situation is no worse
than before.  To reliably cross-compile for a new platform you need to
build a set of .hc files for the libraries using a config.h generated on
the *target* machine.  (at some point we'll formalise the
cross-compilation story, but that's another thing on the todo list...)

Fix a scheduling/GC bug, spotted by Wolfgang Thaller.  If a main
thread completes, and a GC runs before the return (from rts_evalIO())
happens, then the thread might be GC'd before we get a chance to
extract its return value, leading to barf("main thread has been GC'd")
from the garbage collector.

The fix is to treat all main threads which have completed as roots:
this is logically the right thing to do, because these threads must be
retained by virtue of holding the return value, and this is a property of
main threads only.

The variable in a foreign declaration should be 'var' not 'varid',
which means that the following should be legal:

	foreign export "plusInt" (+) :: Int -> Int -> Int

(this example is even in the FFI spec!)

Better location for type/class cycle errors

Better reporting of unused imports

Remember to import the things used by foreign export

Dont warn of missing signatures in interface mode

Better error locations for instance errors

Fix assertion handling

Hopefully fix recent breakage in mangling/splitting: toss the .size
directives over the fence into the next block along with .globl and
.type directives (but carefullly avoid tossing the .size directives
for a function, which appear at the *end* of a function block).

Clean out stamp-pkg-conf*

Put the stamp-pkg-conf file in $(GHC_DRIVER_DIR), so that if someone
does 'make clean' in ghc without also cleaning in libraries, we don't
get into an inconsistent state w.r.t. which packages have been
installed.

More notes about building on windows

Stale interface files left by modules which no longer exist can cause
build problems, so in make clean we now delete all the interface files
we can find, rather than just the ones we know were generated from
existing source files.

remove HEAP_HWM_WORDS; it probably hasn't been used for about 5 years

Don't throw away .size and .type directives.  This lets us profile
with cachegrind and get meaningful results from vg_annotate (*very*
useful).

Make sure that permissions on installed files are right by adding
`%defattr(-,root,root)' to the `%files' entries.
Not a critical patch - Manuel built the rpms as root anyway, but this
makes it so that you are not dependent on building the rpm as root.

Fix the install-datas target when $(INSTALL) is ./install-sh

Remove vestiges of GHC.Prim.assert

Fix records with infix constructors (parser/should_compile/read010).
Also tidy up the parser a bit:

 - clean up the tycon productions

 - check the current s/r conflicts (29) against reality,
   and update the comment.

Fix LIE-plumbing bogon that killed spectral/fibheaps

Add support for slurping in nofib dumps created with
EXTRA_RUNTEST_OPTS=-cachegrind, and output tables for number of
instructions, memory reads/writes, and cache misses.

remove old heimdall and cacheprof support, and add cachegrind support
(-cachegrind option).

Fix up exception handling when reading an interface file, and make it
compile with 4.08.x again.

GhcExceptions weren't being caught by readIface, so an error when
reading an interface could be unintentionally fatal (errors should be
soft when reading the old interface file for the current module).
Also, the Interrupted exception should not be caught by readIface,
because we want ^C to behave as normal when reading interface files
(currently it causes an interface-file read error rather than
interrupting the whole compiler).

Some exception-related compatibility functions have been moved from
Util to Panic.

ifdefs for mingw32 to get rid of signal functions.

Fix search-and-replace-o:  s/%lt/&lt

MERGE TO STABLE

--------------------------------------
	Another attempt to make unbound type
	variables in RULES work right
	--------------------------------------

Sigh.  I'm trying to find the unbound type variables on the LHS of a
RULE.  I thought I could just gather free vars, but that does not work
well on an un-zonked LHS, because a big lambda might bind a type variable
that looks different (pre-zonking) but isn't really.

Oh well, back to plan B which is more work but more robust.

Now the zonking phase (in TcHsSyn) arranges to zonk types in a different
way (zonkTypeCollecting) on a rule LHS than in ordinary expressions
(zonkTypeZapping).  This is less dependent on the exact form of the LHS
(good) but involves another mutable variable (not unclean, but it's sad
to have to admit that mutable variables do sometimes allow you to make
non-invasive changes).

and for good measure, use STGCALL1 rather than a bare C call.

remove bogus return too.

NON_ENTERABLE_ENTRY_CODE() should really barf() rather than attempting
a clean shutdown.

MERGE TO STABLE

The GC wasn't properly marking pending signal handlers, which could
lead to "EVACUATED object entered!" errors.  Also, a race occurs if a
signal arrives during GC.  Two fixes:

  (a) mark all pending signal handlers during GC, and
  (b) block signals during GC

MERGE TO STABLE