* libguile/weak-set.c (resize_set): When releasing a lock temporarily to do
an allocation, re-lock it afterwards, instead of unlocking again.
* libguile/weak-table.c (resize_table): Likewise.
* libguile/pairs.h (scm_cons, scm_car, scm_cdr): Define these as inline
functions.
(scm_is_pair): Move here from inline.h.
* libguile/pairs.c: Remove definitions here, and define gsubrs.
* libguile/inline.h: Remove scm_is_pair implementation.
* libguile/inline.c: Include pairs.h to residualize inlines from
pairs.h.
* test-suite/tests/ports.test ("%file-port-name-canonicalization")["relative
canonicalization with /", "relative canonicalization from ice-9",
"absolute canonicalization from ice-9"]: New tests.
* libguile/filesys.c (scm_i_relativize_path): Don't attempt to
canonicalize when encountering an entry of IN_PATH that is the empty
string.
* test-suite/tests/ports.test (with-load-path): New macro.
("%file-port-name-canonicalization"): New test prefix.
* test-suite/tests/cse.test (pass-if-cse): Fix-letrec and canonicalize
the output, so that unreferenced failure continuations get trimmed.
("cse"): Fix the two tests regarding bailout info.
* module/language/tree-il/effects.scm (make-effects-analyzer): The
analyzer will take an optional second argument, a lookup procedure of
type sym -> exp. This can let the analyzer dig into calls to
lexically bound procedures.
* module/language/tree-il/cse.scm (singly-valued-expression?, cse):
Allow CSE to propagate lexicals to tail positions, if the expression
is singly-valued.
* module/language/tree-il/peval.scm (peval): If we can lift one common
test, see if we can lift others as well.
* test-suite/tests/peval.test: Add a test.
* module/language/tree-il/peval.scm (peval): Optimize common tests in
chains of "if" expressions, like those generated by matchers.
* test-suite/tests/peval.test ("partial evaluation"): Add a test.
* module/language/tree-il/analyze.scm (proc-ref?)[special?]: New
procedure.
Return #t for any toplevel-ref of `_'.
* test-suite/tests/tree-il.test ("warnings")["format"]("non-literal
format string using gettext as top-level _"): New test.
* module/language/tree-il/analyze.scm (proc-ref?): Check for
SPECIAL-NAME in the <module-ref> and <lexical-ref> cases too.
* test-suite/tests/tree-il.test ("warnings")["format"]("non-literal
format string using gettext as module-ref _", "non-literal format
string using gettext as lexical _"): New tests.
* libguile/frames.c (scm_frame_source, scm_frame_previous):
* libguile/stacks.c (scm_make_stack):
* module/ice-9/boot-9.scm (exception-printers):
* module/system/vm/frame.scm (frame-call-representation): Fix more
assumptions that frame-procedure is a program, or even a procedure.
* libguile/frames.c (scm_frame_instruction_pointer):
* module/system/vm/frame.scm (frame-bindings):
(frame-next-source, frame-call-representation): Fix a few locations
that thought that the frame-procedure will always be a VM
procedure. This will not not be the case when traversing the stack of
an application of a non-procedure.
* libguile/vm-i-system.c (call, tail-call, mv-call): Instead of
special-casing structs and smobs at these call sites, just set up the
stack, and jump to a generic apply loop if the proc is not a program.
* libguile/vm-engine.c: The generic apply loop is here. Also, the boot
program is now simply a boot continuation, and can handle any number
of arguments.
* libguile/vm.c (make_boot_program): Update the code that makes the boot
continuation.
* libguile/fports.c (scm_setvbuf): Use `scm_take_from_input_buffers'
directly instead of `scm_drain_input'; use `scm_unget_byte' instead of
`scm_unread_string' to put the drained input back to PORT. This
leaves PORT's line/column numbers unchanged, whereas they'd previously
be decreased by the `scm_unread_string' call.
* libguile/ports.c (scm_take_from_input_buffers): Update description and
variable names to refer to "bytes", not "chars".
* test-suite/tests/ports.test ("setvbuf"): New test prefix.
* module/texinfo.scm (texi-command-specs): Add a new kind of texinfo
command, inline-text-args, a sort of a cross between inline-args,
which are unparsed, and inline-text, which is. Perhaps this should
supersede inline-args at some point. In any case, add acronym as an
inline-text-args element.
(inline-content?, arguments->attlist, complete-start-command)
(parse-inline-text-args, make-dom-parser): Adapt for
inline-text-args.
* module/texinfo/serialize.scm (inline-text-args): Add serialization for
@acronym.
* test-suite/tests/texinfo.test ("test-texinfo->stexinfo"): Add some
tests.
According to the new benchmarks, this leads a 5% speed improvement when
reading small strings, and a 27% improvement when reading large strings.
* libguile/read.c (READER_STRING_BUFFER_SIZE): Change to 128; update
comment to mention codepoints.
(scm_read_string): Make `str' a list of strings, instead of a string.
Store characters read in buffer `c_str'. Cons to STR when C_STR is
full, and concatenate/reverse at the end.
* benchmark-suite/benchmarks/read.bm (small, large): New variables.
Set %DEFAULT-PORT-ENCODING to "UTF-8".
("read")["small strings", "large strings"]: New benchmarks.
* libguile/read.c (read_token): Remove unneeded `const' before `size_t'.
(read_complete_token): Remove `overflow_buffer' parameter; return
`char *' instead of `int'. Allocate the overflow buffer with
`scm_gc_malloc_pointerless' instead of `scm_malloc'. Return either
the overflow buffer or BUFFER.
(scm_read_number, scm_read_mixed_case_symbol,
scm_read_number_and_radix): Rename `buffer' to `local_buffer', and
`overflow_buffer' to `buffer'. Remove `overflow'. Adjust code to new
`read_complete_token'.
* libguile/vm.c:
(vm_error):
(vm_error_bad_instruction):
(vm_error_unbound):
(vm_error_unbound_fluid):
(vm_error_not_a_variable):
(vm_error_not_a_thunk):
(vm_error_apply_to_non_list):
(vm_error_kwargs_length_not_even):
(vm_error_kwargs_invalid_keyword):
(vm_error_kwargs_unrecognized_keyword):
(vm_error_too_many_args):
(vm_error_wrong_num_args):
(vm_error_wrong_type_apply):
(vm_error_stack_overflow):
(vm_error_stack_underflow):
(vm_error_improper_list):
(vm_error_not_a_pair):
(vm_error_not_a_bytevector):
(vm_error_not_a_struct):
(vm_error_no_values):
(vm_error_not_enough_values):
(vm_error_continuation_not_rewindable):
(vm_error_bad_wide_string_length):
(vm_error_invalid_address):
(vm_error_object):
(vm_error_free_variable): New internal helpers, implementing VM error
handling.
* libguile/vm-engine.h (VM_ASSERT): New helper macro.
(ASSERT, CHECK_OBJECT, CHECK_FREE_VARIABLE):
(PRE_CHECK_UNDERFLOW, PUSH_LIST): Use the new helper.
* libguile/vm-i-loader.c:
* libguile/vm-i-scheme.c:
* libguile/vm-i-system.c: Use VM_ASSERT and the out-of-line error
handlers.
* libguile/vm-engine.c (vm_engine): Remove inline error handlers, and
remove a couple of local vars. Use VM_ASSERT. Have halt handle the
return itself.
* libguile/stacks.c: update make-stack and narrow_stack to handle
prompt tags that are not symbols.
* test-suite/tests/eval.test: add tests for trimming a stack with
a prompt tag.
* module/ice-9/vlist.scm (make-block): If we are making a hash table,
allocate it inline with the contents. Otherwise don't even add a
pointer to the block.
(block-hash-table?): New internal accessor.
(block-ref*): Remove. Vhash entries are no longer wrapped.
(block-ref):
(block-hash-table-next-offset):
(block-hash-table-set-next-offset!):
(block-hash-table-ref):
(block-hash-table-set!):
(block-hash-table-add!): Adapt to take content vector explicitly, and
to expect the hash table inline with the contents. Some of these
accessors are new. Adapt callers.
(assert-vlist): New helper.
(vlist-cons): Update comment.
(vhash?): Update scheme to allocate the hash table and chain links
inline with the contents.
(%vhash-fold*, %vhash-assoc): Rewrite to be more performant.
* module/language/tree-il/peval.scm (<operand>, make-operand)
(make-bound-operands, peval): Avoid emitting needless aliases in
degenerate cases of let.
(visit-operand): If we visit an operand with a fresh counter and have
to abort, record that fact.
* test-suite/tests/peval.test ("partial evaluation"): Add a test.
* module/ice-9/vlist.scm (set-block-next-free!): Define this instead of
increment-block-next-free!.
(block-append!): Refactor to take an offset, and only append if the
offset is the next free value, and there is space in the block.
(block-cons): Refactor to not be a loop. The partial evaluator would
have to understand effects analysis in order to be able to unroll it,
and there's at most one recursion.
Recovers the performance loss resulting from the previous commit.
