SWI-Prolog Changelog from version 10.1.7 to 10.1.8

[May 28 2026]

DOC: macOS .pkg installer build, signing, notarization and cert renewal Add doc/MacOSXInstaller.md: an end-to-end, reproducible guide to building the signed and notarized macOS .pkg --- one-time keychain and Developer-ID/notarytool setup, the build invocation, result verification, certificate renewal, and the failure modes hit on the Apple-Silicon + macOS 26 host.
ADDED: macOS: `ninja pkg' builds a signed, notarized split-layout .pkg installer Add a `pkg' target producing a single Developer-ID-signed and Apple-notarized .pkg that delivers swipl-win.app to /Applications and swipl.framework to /Library/Frameworks.
ADDED: macOS: swipl.framework and the split swipl-win.app bundle layout Package the macOS runtime as a standalone swipl.framework and arrange the split bundle layout the installer ships: swipl-win.app in /Applications, swipl.framework in /Library/Frameworks.

[May 25 2026]

FIXED: Reset cached JIT index decisions when the supervisor changes. The per-argument assessed flags, jiti_tried, pindex_verified, fixed_indexes and unindexed memoize index decisions for the current clause shape. A supervisor change is the signal that this shape has changed enough that those verdicts may be stale. Without a reset, a "not indexable" verdict made on a transient state (e.g., while an autoload triggered through trapUndefined() re-enters Prolog) could stick and disable JIT indexing for the eventual stable clause set.

[May 18 2026]

MODIFIED: PL_predicate() now takes UTF-8 strings as argument.
DOC: Modernized calc.c, cacl.pl example. Replaceing PR 1493

[May 17 2026]

DOC: Unicode bracket-pair atoms are consistent with {} Document that an empty Unicode bracket pair (optionally layout-only) reads as the two-character atom, that atom before '(' is a functor, and '<open><close>'(X) writes as <open>X<close> with the bare atom unquoted, all governed by the brace_terms write option.
TEST: Unicode bracket-pair atoms (read/write consistency with {}) Covers empty/layout pair reading as the atom, '<o><c>'(X) as a compound, unquoted bare-atom output, write_canonical functor form, round-tripping and that {} is unchanged.
ENHANCED: Make Unicode bracket-pair atoms consistent with {} A matching Unicode bracket pair (Ps/Pe) now behaves like {}: an empty pair, optionally with layout in between, reads as the atom '<open><close>' (and as a functor when followed by '('), while '<open><close>'(X) writes as <open>X<close> and the bare atom is written unquoted. E.g. ⟨⟩ reads as '⟨⟩', ⟨⟩(x) as '⟨⟩'(x), and '⟨⟩'(hello) writes as ⟨hello⟩.

[May 15 2026]

FIXED: Avoid crash trying to destroy a running engine from another thread.

[May 13 2026]

ADDED: code_type/2 and char_type/2 property pattern_syntax Exposes Unicode UAX #31 Pattern_Syntax membership through the standard ctype dispatch. Useful for libraries that want to align tokenisation or quoting with the immutable syntax set, and for inspecting which code points the write_term/2 pattern_syntax_solo option will treat as stable.
ADDED: write_term/2 option pattern_syntax_solo, used by write_canonical/1 A new boolean write_term/2 option quotes single-character atoms whose code point is not in the immutable UAX #31 Pattern_Syntax set, so atoms like '€', '·', '🎉' round-trip safely across Unicode versions instead of relying on classifications that may shift in future releases. write_canonical/1 enables the option by default; plain write/1 and writeq/1 are unchanged.
FIXED: compare/3 on wide atoms must order by code point, not code unit On platforms with 16-bit wchar_t (Windows), a supplementary character is stored as a UTF-16 surrogate pair whose lead unit lives in U+D800..U+DBFF. compareUCSAtom() compared raw wchar_t values, so a non-BMP atom like '\U0001D11E' wrongly ordered below BMP atoms in U+DC00..U+FFFF — e.g. compare(C, '\U0001D11E', '豈') (with U+F900) returned < on Windows where Linux (32-bit wchar_t) gave the correct >.
FIXED: Reject surrogate code points in UTF-8 and UTF-16 stream decoders. A well-formed UTF-8 sequence never encodes a surrogate (RFC 3629), and a well-formed UTF-16 sequence never produces a lone trail surrogate. Both decoders previously let such malformed input through, so a file containing the bytes ED A0 80 read with encoding(utf8) would deliver U+D800 as a code point — leaking an invalid Unicode value into atoms and strings even though the new PL_get_char / Sputcode checks reject surrogates everywhere else.
FIXED: Reject UTF-16 surrogate code points at every Prolog API surface. Extends the recent fix to atom_codes/2 to put_code/1,2, put_char/1,2, put/1,2 and the format/2,3 ~c directive: PL_get_char() and the ~c argument check now use VALID_CODE_POINT(), so a lone surrogate is rejected with type_error(character_code, Code) (or format_argument_type for ~c) before it can reach the stream layer. As defence in depth, Sputcode() in pl-stream.c now calls reperror() on a surrogate before encoding-specific handling, preventing foreign callers from emitting invalid UTF-8 or raw-wchar bytes via Sputcode(0xD800, s).
FIXED: Do not allow surrogate pair code points in atom_codes/2.

[May 12 2026]

FIXED: Solo characters do not needs quotes
MODIFIED: Code walker not to track below call/1. This allows using call(Goal) in places that are false possitives for the detection of undefined predicates.

[May 11 2026]

FIXED: Confusable detection when using zero-arity compounds.
ENHANCED: char_type/2 and code_type/2 accept prolog_solo The kernel has a f_is_prolog_solo helper (and a corresponding SOLO category in the syntax map) but did not expose it through the char_type/2 catalogue. Register prolog_solo alongside prolog_symbol, prolog_atom_start and friends in src/os/pl-ctype.c, with a new atom in src/ATOMS, and document the class in the char_type/2 reference (man/builtin.plx).
ENHANCED: library(check) gets list_confusable_identifiers/{0,1} A new consistency check that walks every clause in the modules in module_class (default [user]) and warns on atoms whose written form is a UTS #39 spoof:
ENHANCED: '\''$needs_quotes'\''/1 + fix unquoted_atom for UCS atoms Exposes the kernel's own quote decision as a semidet '\''$needs_quotes'\''/1 so library code (notably library(check)'s confusable-identifier linter) can ask "would writeq quote this atom?" without re-parsing the name.

Package clib

[May 28 2026]

FIXED: bsd-crypt.c: unconditional include of "crypt.h" breaks build when no system crypt.h is available On glibc >= 2.39 the crypt() function is no longer in libc; when libcrypt-dev is also absent (e.g. some PPA build chroots), CMake's HAVE_CRYPT and HAVE_LIBCRYPT both fail and the build falls back to the bsd-crypt.c implementation, which then failed because it unconditionally included "crypt.h". The DES_ONLY define already excludes the only consumers of anything declared there, so guard the include with HAVE_CRYPT_H, matching the pattern in crypt.c.

Package cpp

[May 19 2026]

DOC: document the PlEncoding API in pl2cpp.plx Document the PlEncoding enum and the ENC_INPUT/ENC_OUTPUT defaults, the trailing-PlEncoding constructor/method forms and the deprecated argument-order constructors. Add a PlFunctor section and state that PlModule/PlPredicate/PlFunctor/PlCompound functor names are UTF-8. Replace the stale EncLatin1/EncUTF8/EncLocale enumerators, correct the unify_*() wrapper descriptions and escape & for the PDF build.

Package http

[May 28 2026]

TEST: proxy: reserve the `unused' port by holding a bound (unlistened) socket Concurrent ctest jobs that allocate ports dynamically (bind to port 0) could grab the port the proxy tests had pre-allocated for the `unused' role, causing connections that should be refused to instead succeed against an unrelated test server. Keeping a bound socket on that port for the duration of the suite reserves the port number while still producing ECONNREFUSED on connect(), since no tcp_listen/2 is called. SO_REUSEADDR is intentionally omitted on the held socket.

Package libedit

[May 20 2026]

ADDED: Bind Ctrl+Left/Right to word motion Maps the xterm sequences ESC[1;5D and ESC[1;5C, sent by the Epilog terminal, to ed-prev-word and em-next-word. In vi mode the same sequences are also bound in the command (alternative) keymap to vi-prev-word and vi-next-word.

Package ltx2htm

[May 12 2026]

FIXED: #8 Handling e.g., \\const{\%h}

Package pldoc

[May 18 2026]

CLEANUP: As we go through luatex, we no longer need utf8proc as dependency

Package utf8proc

[May 17 2026]

ADDED: library(uniname) exporting unicode_name/2 Wraps the compact name table in a SWI plugin (uniname4pl.c) and a library (uniname.pl). unicode_name(?CodePoint, ?Name) is nondet: (+,-) is the semidet forward name, (-,+) the semidet reverse lookup, and (-,-) enumerates every named code point on backtracking via a stateful foreign iterator (no quadratic blow-up; >100k solutions in ~50 ms). Registered as the utf8proc:uniname ctest.
ADDED: Prototype compact Unicode code-point -> name table Adds etc/gen_uniname.pl, the generated uniname_data.{c,h} and a small decoder (uniname.c) implementing the ICU / GNU libunistring scheme: algorithmic ranges (Hangul synthesised from Jamo; CJK/Tangut and every PREFIX-<cphex> family derived from the code point) carry no per-code- point data, and the remaining ~34.6k names are stored as a flat LEB128 stream of a code-point delta plus frequency-ordered word ids into a shared word table.

[May 11 2026]

TEST: extend confusable_demo.pl with non-predicate-name cases Adds clauses that put confusable atoms in argument positions, in nested compound functors, in list elements, and in clause bodies, so list_confusable_identifiers/0 exercises every code path of its sub_term/2 walk:
CHANGE: move list_confusable_identifiers/0 into library(check) The hook was duplicated from library(check)'s existing list_* checkers and didn't share their per-atom source-location walking. Move the predicate to library(check) directly, guarded by :- if(exists_source(library(unicode_security))), so it uses the same prolog_program_clause/2 + message_context/4 infrastructure as list_strings/0 and friends.
ENHANCED: hook list_confusable_identifiers/0 into library(check) Adds a linter that walks predicates of every user-class module and flags two UTS #39 concerns:
DOC: fix PlDoc Markdown conventions in unicode_security.pl PlDoc Markdown is not LaTeX:
CHANGE: unicode_script_extensions/2 and unicode_identifier_status/2 are semidet Follow-up to commit 8f00f14, applying the same "no info -> fail" rule to the remaining per-code-point predicates:
CHANGE: unicode_script/2 and unicode_identifier_type/2 are now semidet Previously both predicates absorbed missing entries with a default: unicode_script/2 fell through to 'common' for unassigned code points, and unicode_identifier_type/2 returned [] for code points without a listing in IdentifierType.txt. Both also raised a domain_error for code points outside U+0000..U+10FFFF.
CHANGE: don't vendor UCD source files; add fetch instructions The generator etc/gen_uts39.pl runs only on demand (regen-uts39 target) and the checked-in uts39_data.{c,h} are authoritative for default builds, so keeping the .txt inputs in the repo just bloats history. Remove them, add data/*.txt to .gitignore, and add a data/README.md that says where to obtain them.
ENHANCED: unify range tables as (start, len, value) All UTS #39 / UAX #24 range tables (script, scx, idstatus, idtype) now share one type:
DOC: predicate docstrings and README paragraph for unicode_security PlDoc comments for each exported predicate in unicode_security.pl, plus a one-paragraph README pointer to library(unicode_security) and the regen-uts39 workflow.
TEST: add PLUnit suite for library(unicode_security) Covers all eight predicates: unicode_script/2, unicode_script_extensions/2, unicode_identifier_status/2, unicode_identifier_type/2, unicode_skeleton/2, unicode_confusable/{2,3}, unicode_resolved_scripts/2 and unicode_restriction_level/2.
ENHANCED: implement UTS #39 runtime in library(unicode_security) Implements the runtime algorithms over the generated tables and exports the public Prolog API:
ENHANCED: add UTS #39 / UAX #24 generator and generated tables etc/gen_uts39.pl reads the vendored UCD files under data/ and writes uts39_data.{c,h}. The output is deterministic, so re-running the generator on unchanged inputs produces a zero diff.
ENHANCED: scaffold library(unicode_security) plugin Adds an empty second plugin alongside unicode4pl in the same package. The plugin compiles, the Prolog module loads, the doc section is generated and linked into utf8proc.tex. No predicates yet; this is the integration scaffold for the upcoming UTS #39 implementation.
ENHANCED: vendor UTS #39 and UAX #24 data files for unicode_security Source files for the upcoming library(unicode_security) generator:

Package xpce

[May 25 2026]

FIXED: Hold a code reference on Frame across the FLASH_END SDL event flash_end_callback() pushes MY_EVENT_FLASH_END carrying a raw FrameObj pointer. Same UAF shape as the Timer case: if the frame is destroyed and reused before the consumer in CtoEvent() runs, the F_FREED check reads stale memory. Take a code reference around the queue and skip the push entirely when the frame is already being freed.
FIXED: Hold a code reference on Timer across the SDL event queue tm_callback() pushes MY_EVENT_TIMER carrying a raw Timer pointer. If the Timer is destroyed and its memory freed/reused before the main thread consumes the event, the F_FREEING|F_FREED check in sdl_timer_event() reads stale memory. Mirror the FDWatch pattern: addCodeReference() before SDL_PushEvent(), delCodeReference() in the consumer so the object stays in F_FREEING (memory alive) until the queued event is drained.
FIXED: Register Epilog pty slave_fd as a drainable console (POSIX) So that ws_discard_input() drops keystrokes the user typed at an Epilog terminal window while a modal confirmer is up. Uses tcflush(TCIFLUSH) on the slave fd, which drops kernel-buffered bytes without consuming them and therefore does not race with the Prolog engine reading the same fd.
FIXED: Register stdin as a drainable console when it is a terminal So that ws_discard_input() drops pending bytes typed at the terminal while a modal confirmer is up. Skipped silently when stdin is redirected (e.g. Epilog spawning swipl with stdin closed).
FIXED: ws_discard_input() reads stale fd after socket-based dispatch hook Replace the static dispatch_fd cache with a small console registry, populated by the host on init (stdin) and by terminal.c for Epilog ptys. ws_dispatch() now only watches an fd transiently for the duration of one call, never caching it as a global; ws_discard_input() iterates the registry and uses tcflush()/FlushConsoleInputBuffer() instead of read(), avoiding races with engines reading the same fd.

[May 20 2026]

ADDED: Send xterm Ctrl+Left/Right escape sequences from the terminal Ctrl modifier on cursorLeft/cursorRight now produces ESC[1;5D / ESC[1;5C so libedit (or any xterm-compatible reader) can bind these to word motion.

[May 17 2026]

ADDED: PceEmacs: normalize_region/normalize_buffer M-x commands to apply a Unicode normalization form (nfc, nfd, nfkc, nfkd) to the region or the whole buffer. Conditional on the optional library(unicode).
ADDED: PceEmacs/epilog: insert_symbol command (C-x 8 RET / C-x 8 s) C-x 8 is a prefix in PceEmacs and the epilog terminal; it opens the symbol picker targeting the invoking editor/terminal.
ADDED: library(pce_symbol_picker) for picking Unicode symbols A non-modal singleton (and modal pick_symbol/1) to browse Unicode blocks and curated code ranges, type the picked symbol into the focused window, and remember recents. Supports code_range/3 user ranges, matching pairs, and a filter that searches either block names or Unicode character names.
FIXED: man class hierarchy: correct builtin/user class icons
ADDED: display_manager <->focus_message: hook on keyboard-focus change Fires the given message when a frame gains or loses the keyboard focus, so tools can track the active application window.
ENHANCED: text<-pointed: add round option (caret vs hit-test) <-pointed takes an extra argument selecting caret-style rounding (snap to nearest gap) versus exact hit-testing.
ENHANCED: text_item: optional clear-field image Add a ->clear_image slot to class text_item. When set (use @default for the standard clear-text.svg icon), a clickable image appears at the right of the entry field while it is non-empty; clicking it clears the field. The space is reserved as soon as the image is set so the field does not reflow when the icon appears. Restricted to style `normal'.

[May 15 2026]

ADDED: window->pdf: print the full window contents Graphical->pdf saves <-area, i.e. only the visible part of a window. Override it on window to use <-bounding_box (the union of all graphicals) so scrolled or oversized content is printed in full, with the page offset so the bounding box maps onto the page.
FIXED: graphical->pdf: negate the page offset pdfGraphical() created a page sized to <-area but set d_offset() to +area->x,+area->y. Since drawing primitives add the offset, a graphical at a non-zero position was shifted off the page, producing a blank PDF. Negate the offset so the graphical's area maps onto the page origin.
FIXED: text<-pointed: hit-test with Pango layout, not per-char c_width get_pointed_text() summed c_width() per character to map a click to a character index. For proportional fonts and strings needing font fallback (e.g. the symbol picker showing emoji/Greek/math), that sum drifts from the actual rendered glyph positions, so clicks selected the wrong symbol.

[May 14 2026]

FIXED: xpce printf %c: handle non-ASCII code points Routed %c through snprintf which collapses the int argument to a single byte and produced invalid UTF-8 for any code point above 0x7F. Handle %c separately: Put() takes a uchar_t (uint32_t) which already carries a full code point, and put_void_str promotes the target string to wide as needed. Width and left-align flags from the format spec are still honoured.