Changes in 3.0.8 (since 3.0.7)
* Notable changes
** Cross-module inlining
Although historically Guile has treated modules as glorified hash
tables, most modules are actually _declarative_ -- they just define
functions and variables and provide them for other modules to use, and
don't manipulate modules as first-class objects. See "Declarative
Modules" in the manual, for more discussion.
Since version 3.0.0, Guile has taken advantage of declarative semantics
to allow a top-level definition to be inlined within its uses in the
same compilation unit, provided the binding is never assigned and
defined exactly once. Guile 3.0.8 extends this to allow some
exported declarative definitions to be inlined into other modules.
This facility is mostly transparent to the user and is enabled at the
default -O2 optimization level. "Small" definitions are available for
cross-module inlining (-Oinlinable-exports, included at -O2). The
actual inlining decision is performed by Guile's partial evaluation pass
(the -Ocross-module-inlining modifier to -Opeval, included at -O2 also),
subject to effort and size growth counters.
Note however that as with macros, when a definition changes in module A,
a separately compiled module B that uses that definition doesn't
automatically get recompiled. This is a limitation in Guile that we
would like to fix.
As another limitation, cross-module inlining is only available for
imports from modules which have already been compiled at -O2 (or
otherwise with -Oinlinable-exports).
When determining whether to enable this facility by default, we weighed
the usability problems of stale inlined bindings against the benefit of
allowing module boundaries to no longer be optimization boundaries, we
ended up on the "let's do it!" side of the equation. However we welcome
feedback from users as to what should be the default behavior, until
such a time as we have a proper notion of when a compiled file is stale
or not.
** Avoid the need for a custom GMP allocator
In Guile 3.0.6, we fixed a longstanding bug in Guile's use of the
library that Guile uses to implement bignums (large integers), GMP
(https://gmplib.org). See the Guile 3.0.6 release notes. However this
left us with a suboptimal Guile, in which each large integer had to have
a finalizer to free the memory allocated by GMP. Finalizers take time
and space, and so they limit allocation rate, causing bignum performance
to drop. Though you could set an environment variable to go back to the
older, faster behavior, it wasn't the default.
In Guile 3.0.8 we fix this problem comprehensively by avoiding embedding
GMP's mpz_t values in Guile bignums. Instead we embed the bignum digits
directly, avoiding the need for finalizers or custom allocators. This
removes the need for the GUILE_INSTALL_GMP_MEMORY_FUNCTIONS environment
variable mentioned in the Guile 3.0.6 release notes. We also deprecate
the scm_install_gmp_memory_functions variable.
* New interfaces and functionality
** Typed vector copy functions in (srfi srfi-4 gnu)
The functions `u8vector-copy' `s8vector-copy' `u16vector-copy'
`s16vector-copy' `u32vector-copy' `s32vector-copy' `u64vector-copy'
`s64vector-copy' `f32vector-copy' `f64vector-copy' `c32vector-copy'
`c64vector-copy' `u8vector-copy!' `s8vector-copy!' `u16vector-copy!'
`s16vector-copy!' `u32vector-copy!' `s32vector-copy!'
`u64vector-copy!' `s64vector-copy!' `f32vector-copy!'
`f64vector-copy!' `c32vector-copy!' `c64vector-copy!' have been
added. See SRFI-4 - Guile extensions" in the manual.
** New function srfi-4-vector-type-size in (srfi srfi-4 gnu)
See SRFI-4 - Guile extensions" in the manual.
** `bytevector-fill!' supports partial fill through optional arguments
This is an extension to the r6rs procedure. See "Manipulating
Bytevectors" in the manual.
** `vector-copy!' and `vector-copy' from (rnrs base) included in core
Compared to the previous versions, these accept range arguments. See
"Accessing and Modifying Vector Contents" in the manual.
** New function bitvector-copy
See "Bit vectors" in the manual.
** (system foreign) supports C99 complex types
The types `complex-float' and `complex-double' stand for C99 `float
_Complex' and `double _Complex` respectively.
* Other new optimizations
** Better optimization of "let" in right-hand-side of "letrec"
** Allow constant-folding for calls to "expt"
Thanks to Maxime Devos.
** Add ,optimize-cps REPL meta-command
This meta-command is like ,optimize, but at a lower level.
** Improve alias analysis in common subexpression elimination
** Avoid argument-count checks for well-typed calls to known procedures
This speeds up calls to lexically bound procedures.
** Avoid return-value-count checks for calls to known-return-arity procedures
This new optimization, enabled at -O2, speeds up returns from calls to
lexically bound procedures.
* Build system changes
** Update Gnulib (bugs.gnu.org/49930)
Update gnulib to 8f4538a53d64054ae2fc8b86c0f87c418c6176e6.
** Compile libguile with -flto if available
By default, if the compiler supports link-time optimization via the
-flto flag, Guile will add it to CFLAGS. This results in a libguile
that is approximately 15% smaller. Pass --disable-lto to configure to
inhibit this behavior.
** Trim set of prebuilt .go files shipped in the tarball
Guile includes built Scheme files in its tarball to speed up the build,
for casual builders that are less concerned with reproducibility.
However they took a lot of space and we have now trimmed these down to a
more minimal set. As always, you can remove them and build entirely
from source via a `make -C prebuilt clean`.
* New deprecations
** Vector functions require vector arguments
Passing arrays that are not vectors (arrays for which `(vector? array)'
returns false) to functions `vector-move-left!', `vector-move-right!',
`vector->list', and `vector-copy' is deprecated. Use `array-copy!',
`array-copy', and `array->list' for such arguments.
** `scm_from_contiguous_typed_array' is deprecated
This function was added during the Guile 2.x series and was not
documented and is no longer used in Guile itself.
** Deprecate the "simple vector" concept, `scm_is_simple_vector'
This concept meant to indicate "vectors which aren't array slices". Use
scm_is_vector.
** Deprecate internal contiguous array flag
We still reserve space for the flag to preserve ABI but it has no
effect. As such we also remove the internal SCM_I_ARRAY_CONTIGUOUS,
SCM_SET_ARRAY_CONTIGUOUS_FLAG, SCM_CLR_ARRAY_CONTIGUOUS_FLAG,
SCM_I_ARRAY_CONTP preprocessor interfaces, as they were internal and
there is no longer a sensible way of using them.
** Deprecate symbol properties
Symbols used to have a "function slot" and a "property slot", inherited
from Emacs Lisp and early Lisps, which one would access with
'symbol-pref', 'symbol-fref', 'symbol-pset!', and 'symbol-fset!'. These
procedures have been discouraged in favor of object properties; they are
now deprecated. This saves a few words of memory per symbol.
* Bug fixes
** Fix compilation of (ash x N), where N is a literal, at -O1 and below
** Texinfo and XML parsers are now thread-safe (bugs.gnu.org/51264)
** Fix `filename-completion-function' in (ice-9 readline)
** Fix trace-calls-to-procedure (bugs.gnu.org/43102, bugs.gnu.org/48412)
** Fix bug in nftw function (bugs.gnu.org/44182)
** Fix optimization bug in CSE in eq-constant? if both branches same
** Fix readline initialization with invalid keymaps
** Fix crash when reading #nil (bugs.gnu.org/49305)
** Fix read error when reading #{}}#.
** Fix Darwin host detection in foreign-library facility.
** Fix unification of (x ...) patterns in `match'
** Fix scaling floats with leading zeroes in `format'
** Improve support for r7rs-style `(srfi N)' and r6rs-style `(srfi :N)
module names (bugs.gnu.org/39601, bugs.gnu.org/40371)
** Add support for the ARC architecture (bugs.gnu.org/48816)
** Build fix for const strerror result (bugs.gnu.org/43987)
** Fix typos in SRFI documentation (bugs.gnu.org/50127)
** Fix bounds check in `recvfrom!' (bugs.gnu.org/45595)
** Add support for riscv32
** Limit `ash' to left-shift by 2^32 bits (bugs.gnu.org/48150)
** Fix type confusion in heap-numbers-equal? calls from VM
0f21e48f52