fenv: Add an example program to show problems with forkIO.
fenv: Add appropriate #ifdef's to rounding direction helpers.
An implementation need not support all of the rounding modes, so only
compile in support for those that are available.
An implementation need not support all of the rounding modes, so only
compile in support for those that are available.
fenv: Separate exception testing from environment save/restore.
floating: Add a Sortable constraint for PrimFloat.
Every instance of this class should be sortable, thus we don't need to
pollute type signatures elsewhere.
Every instance of this class should be sortable, thus we don't need to
pollute type signatures elsewhere.
floating: Add a new Data.Floating.Types module.
This module combines all the modules under it, which are no longer
exposed.
This module combines all the modules under it, which are no longer
exposed.
trivial: Remove redundant imports and pointless variables.
The dependency graph is bad enough right now as it is, removing redunant
imports should help a bit.
The dependency graph is bad enough right now as it is, removing redunant
imports should help a bit.
roundable: Split out the Roundable class.
This class, by itself, is not really a floating point thing. Move it
into its own module.
This class, by itself, is not really a floating point thing. Move it
into its own module.
poset: Add fixity declarations.
floating: Move Data.Floating.CMath.Instances under Data.Floating.Types.
Part of a massive cleanup of altfloat's module layout.
Part of a massive cleanup of altfloat's module layout.
floating: Merge Data.Floating.Types and Data.Floating.Classes.
The module dependency graph of altfloat has become too complicated for
mere humans to understand. This is the first part of a massive cleanup.
The module dependency graph of altfloat has become too complicated for
mere humans to understand. This is the first part of a massive cleanup.
fenv: Add support for suppressing exceptions across a computation.
floating: Add epsilon to the RealFloat class.
cmath: Fix bad typo in atan2 and atan2f.
We were actually calling atan and atanf, causing horrible problems.
We were actually calling atan and atanf, causing horrible problems.
complex: Add initial complex support.
cmath: Fix section headings to match the C standard.
floating: Add instances for CDouble and CFloat.
floating: Fix off-by-one error in scaleRational.
2^52 is 53 bits wide.
2^52 is 53 bits wide.
floating: Split out formatting from the Double module.
floating: Make fromRational work properly.
The current implementation of fromRational (for both floating types) is
a simple floating point division, which does not work correctly when
either the numerator or the denominator cannot be represented exactly.
We do a correct conversion by scaling the rational by a power of two to
get a large integer that makes maximum use of the available precision.
Since such an integer is exactly representable, we can use scalb to get
the actual value.
The current implementation of fromRational (for both floating types) is
a simple floating point division, which does not work correctly when
either the numerator or the denominator cannot be represented exactly.
We do a correct conversion by scaling the rational by a power of two to
get a large integer that makes maximum use of the available precision.
Since such an integer is exactly representable, we can use scalb to get
the actual value.
floating: Add scalb as the altfloat version of scalbln.
I don't think this function is really useful in the context of
manipulating the floating point environmnent, so it is in the
PrimFloat class. Move logb down there too, since they're a pair.
I don't think this function is really useful in the context of
manipulating the floating point environmnent, so it is in the
PrimFloat class. Move logb down there too, since they're a pair.
floating: Add another characteristic function and generalize types.
floating: Add some floating point characteristic functions.
floating: Punt the FFI newtype wrappers.
Newtypes aren't supposed to change underlying representation, so they
should be unnecessary. Indeed, things still seem to work without them.
Newtypes aren't supposed to change underlying representation, so they
should be unnecessary. Indeed, things still seem to work without them.
altfloat-0.3
doc: Add haddock documentation for the RealFloat class.
floating: Rename tgamma to gamma.
The name "tgamma" is historical legacy due to older systems having
already used the name "gamma". Since altfloat has no such legacy, we
don't need such silliness.
The name "tgamma" is historical legacy due to older systems having
already used the name "gamma". Since altfloat has no such legacy, we
don't need such silliness.
floating: Move the Double and Float modules.
floating: Use nearbyint instead of round in fquotRem.
The remainder function rounds halfway values to even, while the round
function rounds halfway values away from zero. This is obviously wrong,
so use nearbyint which also rounds halfway to even, at least in the
pure case.
In the FEnv case, changing the rounding mode now affects the quotient,
which is probably wrong. So this change probably just trades an old
bug for a new, less important one.
The remainder function rounds halfway values to even, while the round
function rounds halfway values away from zero. This is obviously wrong,
so use nearbyint which also rounds halfway to even, at least in the
pure case.
In the FEnv case, changing the rounding mode now affects the quotient,
which is probably wrong. So this change probably just trades an old
bug for a new, less important one.
floating: Add rint function to complement nearbyint.
examples: Add a demonstration of why FEnv is still dangerous.
fenv: Add support for raising exceptions explicitly.
cfloat: Use FE_ALL_EXCEPT instead of -1.
compat: Make an effort to support fma properly.
Perform arithmetic as the long double type in the hopes that there is
extra precision with which to compute the correct answer.
Perform arithmetic as the long double type in the hopes that there is
extra precision with which to compute the correct answer.
floating: Punt the class version of C's remquo function.
This function is not very Haskell-ish, and any performance benefits are
likely lost in the ugly FFI wrapper. If anyone cares, they can still
use the binding in Data.Floating.CMath themselves.
This function is not very Haskell-ish, and any performance benefits are
likely lost in the ugly FFI wrapper. If anyone cares, they can still
use the binding in Data.Floating.CMath themselves.
floating: Add a separate class for the basic floating types.
This makes the RealFloat instance for FEnv less horrible. Once complex
numbers are added, this will also prevent one from packing FEnv values
into a Complex.
This makes the RealFloat instance for FEnv less horrible. Once complex
numbers are added, this will also prevent one from packing FEnv values
into a Complex.
compat: Add note to source file about non-conformance.
Make it clear that these functions really are not an appropriate
replacement for a proper C99 math library.
Make it clear that these functions really are not an appropriate
replacement for a proper C99 math library.
doc: Acknowledge the existence of floating point environment access.
Also fix comment about multi-threaded programs.
Also fix comment about multi-threaded programs.
compat: Implement wrappers for C99 float-valued functions.
build: Fix build type in cabal file.
compat: Add configure warning when replacements are used.
cfloat: Add compatibility wrappers for some C99 functions.
It was reported that some C99 math functions are not present on FreeBSD.
Some of those functions can be implemented "well enough" fairly easily,
and whether or not to build them can be controlled by autoconf.
These implementations are very simple, and in some cases don't satisfy
the all the requirements of the C99 standard.
It was reported that some C99 math functions are not present on FreeBSD.
Some of those functions can be implemented "well enough" fairly easily,
and whether or not to build them can be controlled by autoconf.
These implementations are very simple, and in some cases don't satisfy
the all the requirements of the C99 standard.
build: Add source repository to cabal file.
fenv: Add explicit definitions for infinity, nan and pi.
The default definitions of infinity and nan raise exceptions.
The default definitions of infinity and nan raise exceptions.
fenv: Fix embarrassing typo in the Fractional instance.
fenv: Re-export Control.Applicative since users need it.
floating: Add exception support to fenvEval.
license: Fix copyright years in source files.
floating: Use a strict data constructor for FEnv.
This seems a bit more natural than using seq in the definition of pure.
This seems a bit more natural than using seq in the definition of pure.
floating: Add initial support for floating point exceptions.
floating: Add nearbyint to the RealFloat class.
This is meaningful now that we have support for changing the rounding
mode.
This is meaningful now that we have support for changing the rounding
mode.
build: Update cabal dependency information for GHC 6.12.
GHC 6.12 has split the integer package into two: integer-gmp and
integer-simple. We can build with any one of these packages since we
only use one bloody function from them.
GHC 6.12 has split the integer package into two: integer-gmp and
integer-simple. We can build with any one of these packages since we
only use one bloody function from them.
float: Remove use of decodeFloatInteger.
The new integer packages in GHC 6.12 have removed this function,
presumably because an Int# is sufficient for storing the components of a
Float. Use decodeFloat_Int# instead.
The new integer packages in GHC 6.12 have removed this function,
presumably because an Int# is sufficient for storing the components of a
Float. Use decodeFloat_Int# instead.
floating: Use bang patterns for unlifted bindings.
GHC 6.12 complains very loudly if we do not have these, and claims that
it will be an error to omit them in 6.14.
GHC 6.12 complains very loudly if we do not have these, and claims that
it will be an error to omit them in 6.14.
cfloat: remove INCLUDE pragmas as GHC 6.12 hates them.
build: Add missing Data.Floating.Float to cabal file.
floating: Add initial support for the floating point environment.
build: Add homepage.
cfloat: Generalize double_format to allow specifying precision.
float: Add class instances for Float.
Now that the interface for Double seems to be pinned down, we can fill
in the blanks for Float.
Now that the interface for Double seems to be pinned down, we can fill
in the blanks for Float.
cmath: Add bindings for float-valued functions.
altfloat-0.2
license: Add missing copyright headers.
doc: Add rationale behind the Roundable class.
doc: Fix oops in description of Ord class.
poset: Add commutativity note for max and min.
poset: Rename isSignificant to isOrdered.
doc: Move library description to a better home.
The cabal file is a much more appropriate place than the
Data.Floating.Types module. It will appear on the index page of haddock
documentation and on the library's description page on hackage.
The cabal file is a much more appropriate place than the
Data.Floating.Types module. It will appear on the index page of haddock
documentation and on the library's description page on hackage.
build: Hackage-ify the cabal file.
Hackage does not like BSD2, so use OtherLicense. Also specify an
explicit version of base.
Hackage does not like BSD2, so use OtherLicense. Also specify an
explicit version of base.
floating: Add Roundable instance for Integral a => Ratio a.
poset: Add instance for Integral a => Ratio a.
doc: Minor haddock markup fix.
double: Add an Enum instance.
floating: Add a class which supports rounding functions.
floating: Fix broken isFinite implementation.
Zero is finite too, too.
Zero is finite too, too.
floating: Get rid of the extra Instances module.
Now that we've killed IncoherentInstances, the Poset and Sortable
instances for Double can go back where they belong.
Now that we've killed IncoherentInstances, the Poset and Sortable
instances for Double can go back where they belong.
poset: Remove IncoherentInstances and define our own Ord class.
The "every Ord is a Poset" logic turned out to cause major problems when
trying to define instances of a class such as
class Poset a => Fooable a.
The result was that the type checker still tried to find an Ord instance
even when a Poset instance was available.
So we flip the logic around and create our own Ord class, then say
"Every instance of Ord is an instance of Prelude.Ord". This means we
need to manually add Poset instances for everything, but it should solve
most issues.
The "every Ord is a Poset" logic turned out to cause major problems when
trying to define instances of a class such as
class Poset a => Fooable a.
The result was that the type checker still tried to find an Ord instance
even when a Poset instance was available.
So we flip the logic around and create our own Ord class, then say
"Every instance of Ord is an instance of Prelude.Ord". This means we
need to manually add Poset instances for everything, but it should solve
most issues.
poset: Use IncoherentInstances to resolve overlaps.
The Ord instances for Poset are causing major problems with overlapping.
Adding IncoherentInstances on all the other Poset instances seems to
work, at least for now. It remains to be seen whether or not the
"Every Ord is a Poset" idea is actually worth the hassle, or whether
there's a better way to do this.
Creating our own Ord class inside the Poset module and defining our own
instances for it means we don't get to (easily) take advantage of the
very rich libraries built on top of Ord.
The Ord instances for Poset are causing major problems with overlapping.
Adding IncoherentInstances on all the other Poset instances seems to
work, at least for now. It remains to be seen whether or not the
"Every Ord is a Poset" idea is actually worth the hassle, or whether
there's a better way to do this.
Creating our own Ord class inside the Poset module and defining our own
instances for it means we don't get to (easily) take advantage of the
very rich libraries built on top of Ord.
floating: Update export lists.
double: Split out foreign declarations into their own module.
doc: Update haddock comments for Data.Poset.
poset: Export the Ord class (but not its methods).
doc: Elaborate on why Ord instances for floating types is a bad idea.
floating: Add fquotRem, the analogue of C's remquo.
floating: Add fma function.
double: Implement specific comparison functions.
double: Remove superfluous parentheses.
build: Update cabal script for distribution.
misc: Fix FFI pragmas.
license: Add copyright and license information.
cfloat: Update cfloat.h.
doc: Add a detailed overview of the purpose of the library.
docs: Update haddock markup and add a trivial function.
double: Add Poset and Sortable instances.
poset: Add a class for partially ordered datatypes.
build: Add an autoconf configure script to determine CFLAGS.
floating: Add description of floating classes.
floating: Add remainder functions.
cfloat: Allow a printf format specifier to be passed explicitly.
double: implement fromInteger using toFloating.
floating: Allow conversion from any Real to Double or Float.