* libguile/ports-internal.h (scm_t_port_internal)
* libguile/ports.h (scm_t_port): Embed scm_t_port in scm_t_port_internal
so that we have just one allocation.
* libguile/ports-internal.h (SCM_PORT_GET_INTERNAL): Adapt.
* libguile/ports.c (scm_i_port_property, scm_i_set_port_property_x)
(scm_c_make_port_with_encoding): Adapt.
* doc/ref/api-io.texi (I/O Extensions): Update documentation on
implementing port types. Document get_natural_buffer_sizes. Document
the new random_access_p.
* libguile/fports.c (scm_i_fdes_to_port, fport_random_access_p):
(scm_make_fptob): Instead of frobbing rw_random manually, implement a
random_access_p function.
* libguile/ports.c (default_random_access_p)
(scm_set_port_random_access_p): New functions.
scm_make_port_type, scm_c_make_port_with_encoding): Arrange for
random_access_p to work.
* libguile/ports.c (scm_specialize_port_encoding_x): Add some sanity
checks.
(scm_unget_bytes): Use scm_expand_port_read_buffer_x.
(port_clear_stream_start_for_bom_read): Use
scm_specialize_port_encoding_x.
(scm_fill_input): Use scm_expand_port_read_buffer_x.
(scm_expand_port_read_buffer_x): Rename from
scm_set_port_read_buffer_x and actually expand the buffer.
* libguile/ports.h: Adapt to scm_expand_port_read_buffer_x change.
* module/ice-9/ports.scm: Remove ports-in-scheme stuff, and instead
expose the ports internals via an auxiliary module. This will let
ports-in-scheme live in a module during Guile 2.2.
* libguile/ports.c (peek_utf8_codepoint, peek_latin1_codepoint):
(peek_iconv_codepoint, peek_codepoint): Refactor to push error
handling to the leaves, where errors happen. Just return
the (possibly substituted) codepoint, without an error code; if
there's really an error, we should raise it.
(scm_getc, scm_peek_char): Adapt.
* libguile/ports.c (scm_getc): If the port conversion strategy is
'error, signal an error before advancing the read pointer. This is a
change from previous behavior; before, we advanced the read pointer
under an understanding that that was what R6RS required. But, that
seems to be not the case.
* test-suite/tests/ports.test ("string ports"): Update decoding-error
tests to assume that read-char with an error doesn't advance the read
pointer.
* test-suite/tests/rdelim.test ("read-line"): Likewise.
* module/ice-9/ports.scm (peek-bytes): New helper.
(peek-byte): Use peek-bytes helper.
(decoding-error): Don't inline; no need.
(decode-utf8, bad-utf8-len): New helpers.
(peek-char-and-len/utf8): Use new helpers.
(peek-char-and-len): No fast paths, and not inline. Peek-char has its
own fast path.
(%peek-char): Use helpers to make fast path.
* libguile/ports.c (scm_specialize_port_encoding_x)
(scm_port_clear_stream_start_for_bom_read): New functions exported
to (ice-9 ports).
* module/ice-9/ports.scm (clear-stream-start-for-bom-read):
(fill-input, peek-char-and-len): Rework to handle BOM in fill-input
instead of once per peek-char.
* libguile/print.c (display_string_using_iconv): Remove BOM handling;
this is now handled by scm_lfwrite.
* libguile/ports.c (open_iconv_descriptors): Refactor to take encoding
as a symbol.
(prepare_iconv_descriptors): New helper.
(scm_i_port_iconv_descriptors): Remove scm_t_port_rw_active argument,
and don't sniff UTF-16/UTF-32 byte orders here. Instead BOM handlers
will call prepare_iconv_descriptors.
(scm_c_read_bytes): Call new port_clear_stream_start_for_bom_read
helper.
(port_maybe_consume_initial_byte_order_mark)
(scm_port_maybe_consume_initial_byte_order_mark): Remove. Leaves
Scheme %peek-char broken but it's unused currently so that's OK.
(peek_iconv_codepoint): Fetch iconv descriptors after doing fill-input
because it's fill-input that will sniff the BOM.
(peek_codepoint): Instead of handling BOM at every character, handle
in fill-input instead.
(maybe_consume_bom, port_clear_stream_start_for_bom_read)
(port_clear_stream_start_for_bom_write): New helpers.
(scm_fill_input): Slurp a BOM if needed.
(scm_i_write): Clear the start-of-stream-for-bom-write flag.
(scm_lfwrite): Write a BOM if needed.
* module/ice-9/ports.scm: Speed tweaks to %peek-char. Ultimately
somewhat fruitless; I can get 1.4s instead of 1.5s by only
half-inlining the UTF-8 case though.
* module/ice-9/ports.scm (EILSEQ, decoding-error, peek-char-and-len/utf8):
(peek-char-and-len/iso-8859-1, peek-char-and-len/iconv):
(peek-char-and-len, %peek-char): New definitions. Missing iconv1 for
peek-char, but enough to benchmark.
* libguile/ports.h (scm_sys_port_encoding, scm_sys_set_port_encoding):
New functions, to expose port encodings as symbols directly to (ice-9
ports).
(scm_port_maybe_consume_initial_byte_order_mark): New function.
* libguile/ports.c (scm_port_encoding): Dispatch to %port-encoding.
(scm_set_port_encoding_x): Dispatch to %set-port-encoding!.
(port_maybe_consume_initial_byte_order_mark): New helper, factored out
of peek_codepoint.
(scm_port_maybe_consume_initial_byte_order_mark, peek_codepoint): Call
port_maybe_consume_initial_byte_order_mark.
* module/ice-9/ports.scm (port-encoding): Implement in Scheme.
* libguile/ports.h (scm_t_port): Represent the conversion strategy as a
symbol, to make things easier for Scheme. Rename to
"conversion_strategy".
(scm_c_make_port_with_encoding): Change to take encoding and
conversion_strategy arguments as symbols.
(scm_i_string_failed_conversion_handler): New internal helper, to turn
a symbol to a scm_t_string_failed_conversion_handler.
(scm_i_default_port_encoding): Return the default port encoding as a
symbol.
(scm_i_default_port_conversion_strategy)
(scm_i_set_default_port_conversion_strategy): Rename from
scm_i_default_port_conversion_handler et al. Take and return Scheme
symbols.
* libguile/foreign.c (scm_string_to_pointer, scm_pointer_to_string): Use
scm_i_default_string_failed_conversion_handler instead of
scm_i_default_port_conversion_handler.
* libguile/print.c (PORT_CONVERSION_HANDLER): Update definition.
(print_normal_symbol): Use PORT_CONVERSION_HANDLER.
* libguile/r6rs-ports.c (make_bytevector_input_port):
(make_custom_binary_input_port, make_bytevector_output_port): Adapt to
changes in scm_c_make_port_with_encoding.
* libguile/strings.h:
* libguile/strings.c (scm_i_default_string_failed_conversion_handler):
New helper.
(scm_from_locale_stringn, scm_from_port_stringn):
(scm_to_locale_stringn, scm_to_port_stringn): Adapt to interface
changes.
* libguile/strports.c (scm_mkstrport): Adapt to
scm_c_make_port_with_encoding change.
* libguile/ports.c (scm_c_make_port): Adapt to
scm_c_make_port_with_encoding change.
(ascii_toupper, encoding_matches, canonicalize_encoding): Move down in
the file.
(peek_codepoint, get_codepoint, scm_ungetc): Adapt to port conversion
strategy change. Remove duplicate case in get_codepoint.
(scm_init_ports): Move symbol initializations to the same place.
* libguile/ports-internal.h (scm_t_port_internal): Remove encoding_mode
member.
* libguile/ports.h (scm_t_port): "encoding" member is now a SCM symbol.
* libguile/ports.c (scm_init_ports): Define symbols for the encodings
that we handle explicitly.
(encoding_matches): Adapt to check against an encoding as a symbol.
(canonicalize_encoding): Return an encoding as a symbol.
(scm_c_make_port_with_encoding, scm_i_set_default_port_encoding)
(decide_utf16_encoding, decide_utf32_encoding)
(scm_i_port_iconv_descriptors, scm_i_set_port_encoding_x)
(scm_port_encoding, peek_codepoint, scm_ungetc): Adapt to encoding
change.
* libguile/print.c (display_string_using_iconv, display_string):
* libguile/read.c (scm_read_character):
* libguile/strings.c (scm_from_port_stringn, scm_to_port_stringn): Adapt
to port encoding change.
* module/ice-9/ports.scm (fill-input): Rewrite to make changes like the
ones made to the C scm_fill_input: allow callers to specify a minimum
amount of buffering.
* libguile/ports.c (scm_i_set_pending_eof): Remove now-unused helper.
(peek_utf8_codepoint, peek_latin1_codepoint, peek_iconv_codepoint):
(peek_codepoint): Refactor the fundamental character readers in Guile
to peek into the read buffer instead of reading then unreading. This
will allow Scheme to use the port buffer to convert, when we port this
to Scheme.
(get_codepoint): Use peek_codepoint.
(scm_getc): Adapt.
(scm_peek_char): Use peek_codepoint.
* libguile/ports.h (scm_fill_input): Add "minimum_size" argument. Adapt
all callers to pass 0 as this argument.
* libguile/ports.c (scm_i_read): Inline into scm_fill_input.
(scm_fill_input): "minimum_size" argument ensures that there are a
certain number of bytes available, or EOF. Instead of shrinking the
read buffer, only fill by the read_buffering amount, or the
minimum_size, whichever is larger.
* libguile/r6rs-ports.c:
* libguile/read.c: Adapt scm_fill_input callers.
* libguile/ports.c (trampoline_to_c_read, trampoline_to_c_write): Since
C might assume that the indices are within bounds of the bytevector,
verify them more here.
(scm_port_random_access_p, scm_port_read_buffering)
(scm_set_port_read_buffer, scm_port_read, scm_port_write): New helpers
exposed to (ice-9 ports).
(scm_port_read_buffer, scm_port_write_buffer): Don't flush or validate
port mode; we do that in Scheme.
* module/ice-9/ports.scm: Implement enough of port machinery to
implement peek-byte in Scheme. Not yet exported.
* libguile/init.c (scm_i_init_guile): Initialize ports before
strports/fports, so that we have initialized the read/write
trampolines before making port types.
* libguile/ports.c (scm_init_ice_9_ports): Define the-eof-object here.
Update a comment.
* module/ice-9/ports.scm: Use the-eof-object definition from C.
* libguile/ports.h (scm_t_ptob_descriptor): Add "scm_read" and
"scm_write" members, for calling from Scheme.
(scm_set_port_scm_read, scm_set_port_scm_write): New procedures.
* libguile/ports.c (trampoline_to_c_read_subr)
(trampoline_to_c_write_subr): New static variables.
* libguile/ports.c (scm_make_port_type): Initialize scm_read and
scm_write members to trampoline to C.
(trampoline_to_c_read, trampoline_to_scm_read)
(trampoline_to_c_write, trampoline_to_scm_write): New helpers.
(scm_set_port_scm_read, scm_set_port_scm_write): New functions.
(default_buffer_size): Move definition down.
(scm_i_read_bytes, scm_i_write_bytes): Use new names for read and
write procedures.
(scm_init_ports): Initialize trampolines.
* libguile/ports.c (scm_peek_char): Optimize. A loop calling peek-char
on a buffered string port 10e6 times goes down from 50ns/iteration to
32ns/iteration.
* libguile/print.c (scm_write, scm_display):
* libguile/read.c (set_port_read_option): Remove port locking. Reading
and writing to the same port from multiple threads just must not
crash; it doesn't have to make sense.
* libguile/ports.h (scm_unget_bytes_unlocked, scm_unget_byte_unlocked):
Remove.
* libguile/ports.c (looking_at_bytes): Use scm_unget_bytes instead of
scm_i_unget_bytes_unlocked
(scm_unget_bytes): Rename from scm_i_unget_bytes_unlocked. Remove
other implementations of this function.
(scm_unget_byte): Likewise.
(scm_ungetc_unlocked, scm_peek_char): Use scm_unget_byte.
* libguile/read.c (read_token): Use scm_unget_byte.
