2014-12-15 12:46:34 +01:00
|
|
|
# $NetBSD: hacks.mk,v 1.6 2014/12/15 11:46:35 jperkin Exp $
|
2009-10-08 03:24:51 +02:00
|
|
|
|
|
|
|
|
.if !defined(OCAML_HACKS_MK)
|
2011-03-29 15:45:01 +02:00
|
|
|
OCAML_HACKS_MK= defined
|
2009-10-08 03:24:51 +02:00
|
|
|
|
|
|
|
|
.include "../../mk/compiler.mk"
|
|
|
|
|
|
2014-01-17 11:03:23 +01:00
|
|
|
### [Fri Jan 17 18:06:18 JST 2014 : pho]
|
2014-11-25 12:09:51 +01:00
|
|
|
### The linker that comes with Darwin 9 & prior does not recognize the option
|
2014-01-17 11:03:23 +01:00
|
|
|
### "-no_compact_unwind" since compact unwind information is only
|
|
|
|
|
### introduced (and made default) as of Darwin 10. Note that this
|
|
|
|
|
### can't be worked around with transformation
|
|
|
|
|
### "rm:-Wl,-no_compact_unwind" because the linker option is baked
|
|
|
|
|
### into utils/config.ml and will later be used by the OCaml compiler
|
|
|
|
|
### itself.
|
|
|
|
|
###
|
2014-11-25 12:09:51 +01:00
|
|
|
.if !empty(MACHINE_PLATFORM:MDarwin-[0-9].*-*)
|
2014-01-17 11:03:23 +01:00
|
|
|
PKG_HACKS+= no_compact_unwind
|
|
|
|
|
SUBST_CLASSES+= no_compact_unwind
|
|
|
|
|
SUBST_MESSAGE.no_compact_unwind= Removing -no_compact_unwind from linker options
|
|
|
|
|
SUBST_FILES.no_compact_unwind= configure
|
|
|
|
|
SUBST_SED.no_compact_unwind= -e 's/-Wl,-no_compact_unwind//g'
|
|
|
|
|
SUBST_STAGE.no_compact_unwind= pre-configure
|
|
|
|
|
.endif
|
|
|
|
|
|
2009-10-08 03:24:51 +02:00
|
|
|
### [ Wed Oct 7 9:16:42 GMT 2009 : dmcmahill ]
|
|
|
|
|
### On NetBSD/i386, gcc optimisation, at least for version 4.4.1 (lang/gcc44),
|
|
|
|
|
### produces a binary which segfaults. The binary is used during part of the
|
|
|
|
|
### build process.
|
|
|
|
|
###
|
Update to ocaml 4.02.0. Main changes (apart from ocamldoc and bugfixes, see
also upstream changelog):
Language features:
- Attributes and extension nodes
- Generative functors
- Module aliases
* Alternative syntax for string literals {id|...|id} (can break comments)
- Separation between read-only strings (type string) and read-write byte
sequences (type bytes). Activated by command-line option -safe-string.
Build system for the OCaml distribution:
- Use -bin-annot when building.
- Use GNU make instead of portable makefiles.
- Updated build instructions for 32-bit Mac OS X on Intel hardware.
Shedding weight:
* Removed Camlp4 from the distribution, now available as third-party software.
* Removed Labltk from the distribution, now available as a third-party library.
Type system:
* Keep typing of pattern cases independent in principal mode
- Allow opening a first-class module or applying a generative functor
in the body of a generative functor. Allow it also in the body of
an applicative functor if no types are created
* Module aliases are now typed in a specific way, which remembers their
identity. In particular this changes the signature inferred by
"module type of"
- Slight change in the criterion to distinguish private
abbreviations and private row types: create a private abbreviation for
closed objects and fixed polymorphic variants.
* Compare first class module types structurally rather than
nominally. Value subtyping allows module subtyping as long as the internal
representation is unchanged.
Compilers:
- More aggressive constant propagation, including float and
int32/int64/nativeint arithmetic. Constant propagation for floats
can be turned off with option -no-float-const-prop, for codes that
change FP rounding modes at run-time.
- New back-end optimization pass: common subexpression elimination (CSE).
(Reuses results of previous computations instead of recomputing them.)
- New back-end optimization pass: dead code elimination.
(Removes arithmetic and load instructions whose results are unused.)
- Optimization of sequences of string patterns
- Experimental native code generator for AArch64 (ARM 64 bits)
- Optimization of integer division and modulus by constant divisors
- Add "-open" command line flag for opening a single module before typing
* "-o" now sets module name to the output file name up to the first "."
(it also applies when "-o" is not given, i.e. the module name is then
the input file name up to the first ".")
* better sharing of structured constants
- new flag to keep locations in cmi files
- issue warning 3 when referring to a value marked with
the [@@ocaml.deprecated] attribute
- a new format implementation based on GADTs
* Constant exception constructors no longer allocate
- avoid unnecessary boxing in let
- Better compilation of optional arguments with default values
- ocamlopt -opaque option for incremental native compilation
Toplevel interactive system:
- New "#show_*" directives
Runtime system:
- New configure option "-no-naked-pointers" to improve performance by
avoiding page table tests during block darkening and the marking phase
of the major GC. In this mode, all out-of-heap pointers must point at
things that look like OCaml values: in particular they must have a valid
header. The colour of said headers should be black.
- Fixed bug in native code version of [caml_raise_with_string] that could
potentially lead to heap corruption.
- Blocks initialized by [CAMLlocal*] and [caml_alloc] are now filled with
[Val_unit] rather than zero.
- Fixed a major performance problem on large heaps (~1GB) by making heap
increments proportional to heap size by default
- Structural equality treats exception specifically
- efficient comparison/indexing of exceptions
- avoid using unsafe C library functions (strcpy, strcat, sprintf)
- An ISO C99-compliant C compiler and standard library is now assumed.
(Plus special exceptions for MSVC.) In particular, emulation code for
64-bit integer arithmetic was removed, the C compiler must support a
64-bit integer type.
Standard library:
* Add new modules Bytes and BytesLabels for mutable byte sequences.
- add List.sort_uniq and Set.of_list
- a faster version of "raise" which does not maintain the backtrace
- support "Unix.kill pid Sys.sigkill" under Windows
- speed improvement for Buffer
- efficient creation of uninitialized float arrays
- Improve documentation regarding finalisers and multithreading
- Trigger warning 3 for all values marked as deprecated in the documentation.
2014-10-09 21:08:28 +02:00
|
|
|
#.if !empty(MACHINE_PLATFORM:MNetBSD-*-i386)
|
|
|
|
|
#. if !empty(CC_VERSION:Mgcc-4.4.*)
|
|
|
|
|
#PKG_HACKS+= optimisation
|
2014-12-15 12:46:34 +01:00
|
|
|
#BUILDLINK_TRANSFORM+= opt:-O[0-9]*:-O
|
Update to ocaml 4.02.0. Main changes (apart from ocamldoc and bugfixes, see
also upstream changelog):
Language features:
- Attributes and extension nodes
- Generative functors
- Module aliases
* Alternative syntax for string literals {id|...|id} (can break comments)
- Separation between read-only strings (type string) and read-write byte
sequences (type bytes). Activated by command-line option -safe-string.
Build system for the OCaml distribution:
- Use -bin-annot when building.
- Use GNU make instead of portable makefiles.
- Updated build instructions for 32-bit Mac OS X on Intel hardware.
Shedding weight:
* Removed Camlp4 from the distribution, now available as third-party software.
* Removed Labltk from the distribution, now available as a third-party library.
Type system:
* Keep typing of pattern cases independent in principal mode
- Allow opening a first-class module or applying a generative functor
in the body of a generative functor. Allow it also in the body of
an applicative functor if no types are created
* Module aliases are now typed in a specific way, which remembers their
identity. In particular this changes the signature inferred by
"module type of"
- Slight change in the criterion to distinguish private
abbreviations and private row types: create a private abbreviation for
closed objects and fixed polymorphic variants.
* Compare first class module types structurally rather than
nominally. Value subtyping allows module subtyping as long as the internal
representation is unchanged.
Compilers:
- More aggressive constant propagation, including float and
int32/int64/nativeint arithmetic. Constant propagation for floats
can be turned off with option -no-float-const-prop, for codes that
change FP rounding modes at run-time.
- New back-end optimization pass: common subexpression elimination (CSE).
(Reuses results of previous computations instead of recomputing them.)
- New back-end optimization pass: dead code elimination.
(Removes arithmetic and load instructions whose results are unused.)
- Optimization of sequences of string patterns
- Experimental native code generator for AArch64 (ARM 64 bits)
- Optimization of integer division and modulus by constant divisors
- Add "-open" command line flag for opening a single module before typing
* "-o" now sets module name to the output file name up to the first "."
(it also applies when "-o" is not given, i.e. the module name is then
the input file name up to the first ".")
* better sharing of structured constants
- new flag to keep locations in cmi files
- issue warning 3 when referring to a value marked with
the [@@ocaml.deprecated] attribute
- a new format implementation based on GADTs
* Constant exception constructors no longer allocate
- avoid unnecessary boxing in let
- Better compilation of optional arguments with default values
- ocamlopt -opaque option for incremental native compilation
Toplevel interactive system:
- New "#show_*" directives
Runtime system:
- New configure option "-no-naked-pointers" to improve performance by
avoiding page table tests during block darkening and the marking phase
of the major GC. In this mode, all out-of-heap pointers must point at
things that look like OCaml values: in particular they must have a valid
header. The colour of said headers should be black.
- Fixed bug in native code version of [caml_raise_with_string] that could
potentially lead to heap corruption.
- Blocks initialized by [CAMLlocal*] and [caml_alloc] are now filled with
[Val_unit] rather than zero.
- Fixed a major performance problem on large heaps (~1GB) by making heap
increments proportional to heap size by default
- Structural equality treats exception specifically
- efficient comparison/indexing of exceptions
- avoid using unsafe C library functions (strcpy, strcat, sprintf)
- An ISO C99-compliant C compiler and standard library is now assumed.
(Plus special exceptions for MSVC.) In particular, emulation code for
64-bit integer arithmetic was removed, the C compiler must support a
64-bit integer type.
Standard library:
* Add new modules Bytes and BytesLabels for mutable byte sequences.
- add List.sort_uniq and Set.of_list
- a faster version of "raise" which does not maintain the backtrace
- support "Unix.kill pid Sys.sigkill" under Windows
- speed improvement for Buffer
- efficient creation of uninitialized float arrays
- Improve documentation regarding finalisers and multithreading
- Trigger warning 3 for all values marked as deprecated in the documentation.
2014-10-09 21:08:28 +02:00
|
|
|
#. endif
|
|
|
|
|
#.endif
|
2009-10-08 03:24:51 +02:00
|
|
|
|
|
|
|
|
.endif # OCAML_HACKS_MK
|
