[dependabot]

Bumps the uv group with 5 updates in the / directory:

Package	From	To
certifi	2026.1.4	2026.10.19
filelock	3.25.1	4.2.1
nltk	3.10.1	3.10.4
opentelemetry-api	1.40.0	1.45.1
setuptools	83.0.0	84.1.0

Updates certifi from 2026.1.4 to 2026.10.19

Commits

• 662df63 Restore unaffected emSign roots (#439) (#440)
• 5cb238c Remove emsign (#437)
• 69494a7 Bump actions/download-artifact from 8.0.1 to 8.0.2 (#436)
• 029c7b0 Bump actions/upload-artifact from 7.0.1 to 7.0.2 (#435)
• 9d0a8f1 Add support for Python 3.15 (#432)
• db472ab Bump pypa/gh-action-pypi-publish from 1.14.1 to 1.14.2 (#431)
• f4bc676 2026.07.22 (#428)
• 4c91f9c Bump actions/setup-python from 6.3.0 to 7.0.0 (#427)
• 01a66c5 Bump actions/checkout from 7.0.0 to 7.0.1 (#426)
• eb355f4 Bump pypa/gh-action-pypi-publish from 1.14.0 to 1.14.1 (#425)
• Additional commits viewable in compare view

Updates filelock from 3.25.1 to 4.2.1

Release notes

Sourced from filelock's releases.

4.2.1

• On a singleton wrapper, close> and close> end only the holds that tasks acquired through that wrapper. A reader that arrives during close() gets the closed error instead of joining the ending hold. After synchronous code closes or force-releases the shared lock, the wrappers refuse to admit readers into the ended hold, and close() on such a wrapper returns. Closing one wrapper no longer ends a task's nested hold taken through another wrapper on the same path. #832
• ReadWriteLock locks again after a release() that an exception from a signal handler interrupted. The async read-write wrappers stop admitting readers into such a level. The async read-write locks forget a hold whose backend release failed after the backend let go, and they free the memory of each acquire that times out or is canceled. AsyncFileLock returns from release() after a few hundred acquires gave up behind a waiting one. An exception from a signal handler delivered while an AsyncFileLock call joins its queue no longer blocks every later call on that lock. #833
• Fix ReadWriteLock acting on -journal, -wal or -shm files another process puts beside its database: a write could land in another file, a reader could get in beside a writer, and an acquire could fail. SQLite now opens the database by name only in a directory no other account can write, and through a private hard link elsewhere. On Windows the directory's access control list decides. Where hard links are unavailable, keep the database in a directory other accounts cannot write, such as one under %LOCALAPPDATA%. The lock removes link directories through a checked descriptor rather than by name, reads a link directory's name only when the locking account owns it, and leaves alone a directory with other content renamed onto a link directory's name. A database whose schema takes more than a few kilobytes to load raises ValueError instead of running past the acquire timeout. #834
• Fix five bugs in opening the lock file. UnixFileLock no longer truncates a file it did not create, such as one renamed to the lock path, and no longer applies mode to it. An on_acquired hook that writes holder metadata now truncates the file itself. A lease (Linux) or an oplock (Windows) on the lock file or a soft lock's marker no longer holds acquire() past its timeout or blocking=False, and an oplock no longer holds release(); ReadWriteLock on Windows waits for its timeout and raises Timeout. On Windows, a mode without group and other bits, such as 0o600, now creates the lock file with an owner-only permission list; the ReadWriteLock database gets the same. The soft locks' writability check no longer follows a symlink swapped in at the lock path. On a filesystem without flock, a contender from another account now waits on a marker it cannot write instead of raising PermissionError. #835
• Fix SoftFileLock, MarkerSoftFileLock and SoftFileLease removing a live holder's marker in three cases: the holder and the contender run in Linux time namespaces with different boottime offsets; only one of them can read the boot id; a process other than the owner rewrote the marker. Soft markers no longer carry group or other write bits. A process name containing a carriage return or another line separator no longer makes the stale check raise ValueError. When /proc/self/ns/time exists but cannot be read, the stale check no longer compares start times and keeps the marker. #836
• The async read-write locks record their hold under a task owner, so synchronous code on the executor thread that took the hold can no longer re-enter or release it. AsyncSoftReadWriteLock gives back one level on release, so synchronous readers of a shared singleton keep theirs. SoftReadWriteLock tracks read levels per thread: release() from a thread that holds none raises RuntimeError. After its heartbeat recorded a lost hold, it refuses new acquires until that hold is released. #838
• StrictSoftFileLock creates <path>.filelock and its claims directory with the lock's mode and sets the sticky bit (a directory flag that lets only a file's owner remove it) on both. On the next acquisition it narrows a directory it owns that an earlier release created wider; it leaves a directory of another account as it is, so remove such leftovers by hand. It reaches claims through a descriptor opened without following symlinks; on Windows it holds both directories open and refuses a junction (a directory that points at another directory). Under a sticky parent such as /tmp it refuses a coordination directory of another account, as SoftReadWriteLock does for <path>.rw. One read grace now covers a whole claim scan, a failed claim cleanup no longer leaves the sentinel descriptor open, and a non-regular node at a sentinel private-record name raises SoftFileLockProtocolError. On Windows, contenders read claims without blocking their owner's unlink. #839

... (truncated)

Changelog

Sourced from filelock's changelog.

########### Changelog ###########

.. towncrier-draft-entries:: Unreleased

.. towncrier release notes start

---

4.2.4 (2026-10-19)

---

• Keep GitHub Release notes free of extra blank lines while retaining the PyPI and version comparison links. :pr:863

---

4.2.3 (2026-10-19)

---

• Creating a lock on a Python 3.15 interpreter without types.LazyImportType no longer triggers AttributeError at exit. :pr:860
• Handle EFTYPE when a no-follow open refuses a symlink in a soft reader/writer lock's protocol tree, so heartbeat reports lost ownership rather than a transient refresh error.

---

4.2.2 (2026-10-18)

---

• Fix a lock on a relative path that stayed held after a later with lock: ended, when an earlier acquire() had failed because the working directory no longer exists. That call still raises :class:FileNotFoundError, and it now leaves the lock's hold count unchanged. :pr:856

---

4.2.1 (2026-10-17)

---

• On a singleton wrapper, :meth:AsyncReadWriteLock.close() <filelock.AsyncReadWriteLock.close> and :meth:AsyncSoftReadWriteLock.close() <filelock.AsyncSoftReadWriteLock.close> end only the holds that tasks acquired through that wrapper. A reader that arrives during close() gets the closed error instead of joining the ending hold. After synchronous code closes or force-releases the shared lock, the wrappers refuse to admit readers into the ended hold, and close() on such a wrapper returns. Closing one wrapper no longer ends a task's nested hold taken through another wrapper on the same path. :pr:832
• :class:~filelock.ReadWriteLock locks again after a release() that an exception from a signal handler interrupted. The async read-write wrappers stop admitting readers into such a level. The async read-write locks forget a hold whose backend release failed after the backend let go, and they free the memory of each acquire that times out or is canceled. :class:~filelock.AsyncFileLock returns from release() after a few hundred acquires gave up behind a waiting one. An exception from a signal handler delivered while an :class:~filelock.AsyncFileLock call joins its queue no longer blocks every later call on that lock. :pr:833
• Fix :class:~filelock.ReadWriteLock acting on -journal, -wal or -shm files another process puts beside its database: a write could land in another file, a reader could get in beside a writer, and an acquire could fail. SQLite now opens the database by name only in a directory no other account can write, and through a private hard link elsewhere. On Windows the directory's access control list decides. Where hard links are unavailable, keep the database

... (truncated)

Commits

• 60331bb Release 4.2.1
• 834bf13 🧪 test(soft): skip Unix pin test on Windows
• 9d15e38 fix(soft-rw): isolate subclass singleton registries (#855)
• e1e825b 🐛 fix(open): keep lock files intact, honor deadlines and mode (#835)
• 5e3f2ec 🐛 fix(read-write): keep SQLite off shared names, bound schema (#834)
• bfa066d 🐛 fix(strict): keep claim directories private to the lock (#839)
• 977d2c3 🐛 fix(soft): wait on an empty marker instead of evicting (#846)
• 740ff66 🐛 fix(soft): break stale markers without emptying the lock path (#843)
• 86c6d5f 🐛 fix(async): honor cancellation and per-thread gates (#844)
• eeceaa8 🐛 fix(soft): keep live holders when owner identity differs (#836)
• Additional commits viewable in compare view

Updates nltk from 3.10.1 to 3.10.4

Release notes

Sourced from nltk's releases.

v3.10.4

Version 3.10.4 2026-10-20

This release routes every file path, archive, download, URL fetch and external program through one sandbox that keeps reads and writes inside the trusted data directories, refuses path traversal, symlink escapes and decompression bombs, blocks requests to internal network addresses, and runs Java and other third-party tools only from trusted absolute locations with validated options and input. Pickled models now load through a restricted unpickler, XML and JSON parsing are hardened, and regular expressions, recursion and slow algorithms in the tokenizers, parsers, stemmers and corpus readers are bounded against denial-of-service input. Text that NLTK writes to the terminal, to CSV files, to the WordNet browser and to third-party tool input is sanitized against injection.

Security changes you may notice

Every refusal below raises with a message that starts with "Security Violation" and names the check, so a failure is easy to search for (PR #3877).

• Data directories (nltk.pathsec, nltk.data): corpora, models and every other resource are read and written only inside the directories listed in nltk.data.path. A path that leaves a data root, a symlink that points out of one, and a NUL byte or encoded traversal in a resource name are refused. A data root, or a directory on the way to a resource, that another account can write to is skipped with a warning naming the directory and the remedy. Keep private data in a private directory and add it to nltk.data.path; /tmp is not a data root. Model save and load methods (perceptron, maxent, named-entity chunker, Weka, crf, hunpos, Stanford) accept destinations inside a data root only (PR #3797, #3753, #3850, #3761, #3755, #3759, #3760, #3763, #3806, #3807, #3808, #3809, #3813, #3814, #3815, #3848, #3851, #3852, #3853, #3863, #3871, #3888, #3945, #3961, #3969, #3981)
• Downloader: downloaded package archives stay private to the installing account (0600). An install another account must read is extracted: a download made as root or outside the installing account's home extracts on its own, and nltk.download(..., extract=True) / python -m nltk.downloader --extract (or --no-extract) decide explicitly. This also unpacks the packages the index leaves zipped (wordnet, omw-1.4 ...); the extracted files are readable by every account and are found before the archive. find() names an archive it cannot read and says how to extract it instead of reporting it "not found". The downloader installs only into directories no other account can write to, treats an archive another account owns or can write to as stale and downloads it afresh, extracts only from an archive no other account could have changed since its checksum was verified, bounds the size of what it fetches, and refuses index entries and archive members that would write outside the package directory (PR #3929, #3804, #3850, #3778)
• Pickles: nltk.data.load(..., format="pickle") reconstructs plain data only; a pickle that names any class or function is refused. Model loaders that need classes use an explicit allowlist of audited globals, so a gadget such as os.system can never be reconstructed from a model file. chat80 shelve values load the same way (PR #3794, #3754, #3757, #3802, #3817, #3823, #3824)
• XML, JSON and YAML: nltk.xmlsec refuses any document that declares an entity or references an external one (billion-laughs, external-entity and DTD attacks) and bounds the tree depth, and every corpus XML view, the downloader index, Toolbox and internals parse through it. JSON resources load with a depth and size bound; YAML (the markdown corpus front matter and nltk.data.load(..., format="yaml")) loads through a bounded safe loader. Oversized or too deeply nested input raises ValueError naming the limit (PR #3754, #3805, #3828, #3933, #3972, #3974)
• Regular expressions (nltk.redos): every regular expression NLTK compiles, including one you pass to RegexpTokenizer, RegexpStemmer, RegexpParser, re_show, TokenSearcher.findall, tgrep, nltk.help tag-set lookups and the concordance and nemo GUIs, now goes through nltk.redos on the regex engine. A pattern longer than 100,000 characters, nested more than 100 deep, with more than 1,000 capturing groups or an excessive repetition product is refused with nltk.redos.error, a subclass of re.error; a match that runs longer than nltk.redos.DEFAULT_TIMEOUT (5 seconds) raises TimeoutError instead of hanging. Patterns keep re syntax; match objects are the regex engine's (PR #3825, #3829, #3830, #3831, #3832, #3833, #3834, #3835, #3836, #3837, #3838, #3839, #3840, #3841, #3842, #3843, #3844, #3845, #3846, #3796, #3801, #3906, #3907, #3908, #3910)
• Regular expressions made linear: the leading-run and re-anchoring patterns that a timeout cannot catch were rewritten or bounded in the word, Treebank, blankline, TextTiling, toktok and syllable tokenizers, the readline of corpus streams, the CoNLL SRL, TIMIT, senseval, pl196x, NKJP, Alpino, YCOE, IEER, ACE, VerbNet, reviews and Lin readers, relextract, the valuation parser, the CCG lexicon parser, dependency grammars and chomsky_normal_form, each verified on the real corpus (PR #3947, #3934, #3932, #3938, #3935, #3933, #3869, #3944, #3911, #3909, #3953, #3895)
• Size, depth and time bounds: trees deeper than nltk.tree.MAX_TREE_DEPTH (500), WordNet and other relation walks deeper than nltk.util.MAX_RECURSION_DEPTH (500), tgrep deeper than 50, Toolbox deeper than 500, dependency graphs deeper than 200, are refused with ValueError naming the limit rather than exhausting the stack, and the stemmers, distance metrics, skipgrams, parsers and grammar transforms run in linear or bounded time with identical output for legitimate input. chart_parse stops after nltk.parse.chart.DEFAULT_MAX_TIME (5 seconds of wall-clock, needed only for feature grammars; max_time=None disables it). Tree.repr no longer recurses, so a deep accepted tree prints (PR #3827, #3885, #3895, #3767, #3784, #3917, #3958, #3960, #3977)
• External tools and Java: a tool binary is resolved from an absolute path or a trusted search path only, never from the current directory; the binary, its directory and the model and data files handed to it must not be writable by another account (symlinked and multi-call executables such as a packaged Graphviz dot are supported). java() accepts a main class as the first argument only (no -jar, @​argfile or .java source file), validates its options against an allowlist, and loads jars only from a data root unless NLTK_ALLOW_UNSAFE_JARS=1 opts out with a warning. CoreNLP server options are allowlisted. Stanford, Malt, Senna, hunpos, Prover9, megam, TADM, Weka and crf wrappers validate their paths and the text they pipe to the tool against line and argument injection (PR #3772, #3771, #3975, #3743, #3782, #3788, #3768, #3798, #3887, #3858, #3950, #3951, #3812, #3816, #3868, #3894, #3941, #3850, #3925)
• Network: the downloader, nltk.data.load of a URL and the CHILDES reader fetch through pathsec.urlopen, which refuses addresses that are not globally routable (loopback, private, link-local and multicast, including tunnelled and obfuscated forms), pins the validated address for the connection, allows file:// with a local authority only, and refuses a proxied fetch unless NLTK_ALLOW_PROXIED_URLOPEN=1 (or nltk.pathsec.ALLOW_PROXIED_FETCH = True) asserts the proxy is trusted. The CoreNLP client talks to its local server with requests and is not affected (PR #3847, #3810, #3750, #3850)
• Terminal, CSV, DOT and HTML output: every print in the library goes through nltk.termsec.safe_print, which renders control characters, escape sequences and unbalanced bidi overrides from corpus text as visible \xNN escapes instead of letting them reach the terminal. The CSV files the twitter and sentiment helpers write neutralise formula-leading and control-sequence cells, AlignedSent DOT labels are escaped, the WordNet browser escapes corpus text, and nltk.data.load's verbose output no longer prints the full resource URL (PR #3914, #3915, #3937, #3946, #3931, #3930, #3856, #3902)
• Corpus stores and encodings: the PanLex Lite and chat80 sqlite stores open read-only on a pinned descriptor with their schema screened and a statement step budget, chat80 shelves are read by a key manifest, Toolbox SFM markers are sanitised, and a resource encoding that is not a text codec (bz2_codec and friends) is refused (PR #3961, #3964, #3959, #3888, #3800, #3962)
• Package contents: the wheel no longer ships nltk.test.unit (the attack harnesses, advisory probes and their fixtures), so pip install nltk is about half the size and import nltk.test.unit works from a source checkout or the sdist only. The doctests under nltk.test still ship (PR #3954)

Security test and CI infrastructure

• Attack and regression harnesses, advisory probes, the sandbox conftest fixtures, the pre-commit guards (no unsandboxed open, every regex through redos, every pickle through picklesec, every print through safe_print), load-invariant CPU-time timing rules, CI matrix and runner fixes, pinned third-party downloads, and the AUTHORS update (PR #3744, #3749, #3762, #3764, #3770, #3776, #3791, #3795, #3803, #3818, #3820, #3821, #3822, #3849, #3855, #3860, #3861, #3862, #3864, #3865, #3866, #3876, #3884, #3899, #3904, #3905, #3912, #3917, #3924, #3926, #3927, #3939, #3940, #3943, #3948, #3949, #3963, #3968, #3971, #3976, #3982, #3983, #3984, #3985)

Other changes

• Added RTL language syntactic tree visualization support (PR #3472)
• Fix #3394: Ensure punkt_tab is downloaded when punkt is requested (PR #3505)
• Show README.md as the PyPI long description (PR #3746)
• Add stopwords.words("english-extended") (PR #3747)
• Krippendorff's alpha: treat placeholder labels as missing data (#2865, #2732) (PR #3752)
• Fix shared defaults in chat80 Concept (PR #3773)
• Fix Java wrappers decoding str output from java() (universal_newlines) (PR #3777)
• Prevent IndexError in Hungarian stemmer on empty string (PR #3785)
• Isolate MTE tag filters per view (PR #3792)
• Fix newline in ellipsis tokens in TweetTokenizer (PR #3819)
• Isolate default WordNet reference relations (PR #3872)
• Fix resolution paths that cancel distinct literals (PR #3878)
• Fix UTF-8 text preprocessing in NKJP corpus reader (PR #3879)
• Remove shadowed FeatStruct._walk stub and preserve its docstring (PR #3893)
• Count all increasing pairs in RIBES kendall_tau (PR #3900)
• Fix the TweetTokenizer ZWJ emoji pattern capturing adjacent whitespace (PR #3913)
• Surface the CoreNLP server's reason for a non-2xx answer (PR #3942)
• Omit unaligned (NULL target) pairs from Alignment strings, so IBM model output prints (PR #3967)
• Fix DependencyGraph cycle detection, which returned False on every parsed graph (PR #3978)
• Show, range and draw an alignment with an unaligned pair on every surface (PR #3987)

Thanks to the following contributors to 3.10.4: Zhewen Tan, Yanhaoxi, Yazan Balawneh (Cystack.ps),

... (truncated)

Changelog

Sourced from nltk's changelog.

Version 3.10.4 2026-10-20

This release routes every file path, archive, download, URL fetch and external program through one sandbox that keeps reads and writes inside the trusted data directories, refuses path traversal, symlink escapes and decompression bombs, blocks requests to internal network addresses, and runs Java and other third-party tools only from trusted absolute locations with validated options and input. Pickled models now load through a restricted unpickler, XML and JSON parsing are hardened, and regular expressions, recursion and slow algorithms in the tokenizers, parsers, stemmers and corpus readers are bounded against denial-of-service input. Text that NLTK writes to the terminal, to CSV files, to the WordNet browser and to third-party tool input is sanitized against injection.

Security changes you may notice

Every refusal below raises with a message that starts with "Security Violation" and names the check, so a failure is easy to search for (PR #3877).

• Data directories (nltk.pathsec, nltk.data): corpora, models and every other resource are read and written only inside the directories listed in nltk.data.path. A path that leaves a data root, a symlink that points out of one, and a NUL byte or encoded traversal in a resource name are refused. A data root, or a directory on the way to a resource, that another account can write to is skipped with a warning naming the directory and the remedy. Keep private data in a private directory and add it to nltk.data.path; /tmp is not a data root. Model save and load methods (perceptron, maxent, named-entity chunker, Weka, crf, hunpos, Stanford) accept destinations inside a data root only (PR #3797, #3753, #3850, #3761, #3755, #3759, #3760, #3763, #3806, #3807, #3808, #3809, #3813, #3814, #3815, #3848, #3851, #3852, #3853, #3863, #3871, #3888, #3945, #3961, #3969, #3981)
• Downloader: downloaded package archives stay private to the installing account (0600). An install another account must read is extracted: a download made as root or outside the installing account's home extracts on its own, and nltk.download(..., extract=True) / python -m nltk.downloader --extract (or --no-extract) decide explicitly. This also unpacks the packages the index leaves zipped (wordnet, omw-1.4 ...); the extracted files are readable by every account and are found before the archive. find() names an archive it cannot read and says how to extract it instead of reporting it "not found". The downloader installs only into directories no other account can write to, treats an archive another account owns or can write to as stale and downloads it afresh, extracts only from an archive no other account could have changed since its checksum was verified, bounds the size of what it fetches, and refuses index entries and archive members that would write outside the package directory (PR #3929, #3804, #3850, #3778)
• Pickles: nltk.data.load(..., format="pickle") reconstructs plain data only; a pickle that names any class or function is refused. Model loaders that need classes use an explicit allowlist of audited globals, so a gadget such as os.system can never be reconstructed from a model file. chat80 shelve values load the same way (PR #3794, #3754, #3757, #3802, #3817, #3823, #3824)
• XML, JSON and YAML: nltk.xmlsec refuses any document that declares an entity or references an external one (billion-laughs, external-entity and DTD attacks) and bounds the tree depth, and every corpus XML view, the downloader index, Toolbox and internals parse through it. JSON resources load with a depth and size bound; YAML (the markdown corpus front matter and nltk.data.load(..., format="yaml")) loads through a bounded safe loader. Oversized or too deeply nested input raises ValueError naming the limit (PR #3754, #3805, #3828, #3933, #3972, #3974)
• Regular expressions (nltk.redos): every regular expression NLTK compiles, including one you pass to RegexpTokenizer, RegexpStemmer, RegexpParser, re_show, TokenSearcher.findall, tgrep, nltk.help tag-set lookups and the concordance and nemo GUIs, now goes through nltk.redos on the regex engine. A pattern longer than 100,000 characters, nested more than 100 deep, with more than 1,000 capturing groups or an excessive repetition product is refused with nltk.redos.error, a subclass of re.error; a match that runs longer than nltk.redos.DEFAULT_TIMEOUT (5 seconds) raises TimeoutError instead of hanging. Patterns keep re syntax; match objects are the regex engine's (PR #3825, #3829, #3830, #3831, #3832, #3833, #3834, #3835, #3836, #3837, #3838, #3839, #3840, #3841, #3842, #3843, #3844, #3845, #3846, #3796, #3801, #3906, #3907, #3908, #3910)
• Regular expressions made linear: the leading-run and re-anchoring patterns that a timeout cannot catch were rewritten or bounded in the word, Treebank, blankline, TextTiling, toktok and syllable tokenizers, the readline of corpus streams, the CoNLL SRL, TIMIT, senseval, pl196x, NKJP, Alpino, YCOE, IEER, ACE, VerbNet, reviews and Lin readers, relextract, the valuation parser, the CCG lexicon parser, dependency grammars and chomsky_normal_form, each verified on the real corpus (PR #3947, #3934, #3932, #3938, #3935, #3933, #3869, #3944, #3911, #3909, #3953, #3895)
• Size, depth and time bounds: trees deeper than nltk.tree.MAX_TREE_DEPTH (500), WordNet and other relation walks deeper than nltk.util.MAX_RECURSION_DEPTH (500), tgrep deeper than 50, Toolbox deeper than 500, dependency graphs deeper than 200, are refused with ValueError naming the limit rather than exhausting the stack, and the stemmers, distance metrics, skipgrams, parsers and grammar transforms run in linear or bounded time with identical output for legitimate input. chart_parse stops after nltk.parse.chart.DEFAULT_MAX_TIME (5 seconds of wall-clock, needed only for feature grammars; max_time=None disables it). Tree.repr no longer recurses, so a deep accepted tree prints (PR #3827, #3885, #3895, #3767, #3784, #3917, #3958, #3960, #3977)
• External tools and Java: a tool binary is resolved from an absolute path or a trusted search path only, never from the current directory; the binary, its directory and the model and data files handed to it must not be writable by another account (symlinked and multi-call executables such as a packaged Graphviz dot are supported). java() accepts a main class as the first argument only (no -jar, @​argfile or .java source file), validates its options against an allowlist, and loads jars only from a data root unless NLTK_ALLOW_UNSAFE_JARS=1 opts out with a warning. CoreNLP server options are allowlisted. Stanford, Malt, Senna, hunpos, Prover9, megam, TADM, Weka and crf wrappers validate their paths and the text they pipe to the tool against line and argument injection (PR #3772, #3771, #3975, #3743, #3782, #3788, #3768, #3798, #3887, #3858, #3950, #3951, #3812, #3816, #3868, #3894, #3941, #3850, #3925)
• Network: the downloader, nltk.data.load of a URL and the CHILDES reader fetch through pathsec.urlopen, which refuses addresses that are not globally routable (loopback, private, link-local and multicast, including tunnelled and obfuscated forms), pins the validated address for the connection, allows file:// with a local authority only, and refuses a proxied fetch unless NLTK_ALLOW_PROXIED_URLOPEN=1 (or nltk.pathsec.ALLOW_PROXIED_FETCH = True) asserts the proxy is trusted. The CoreNLP client talks to its local server with requests and is not affected (PR #3847, #3810, #3750, #3850)
• Terminal, CSV, DOT and HTML output: every print in the library goes through nltk.termsec.safe_print, which renders control characters, escape sequences and unbalanced bidi overrides from corpus text as visible \xNN escapes instead of letting them reach the terminal. The CSV files the twitter and sentiment helpers write neutralise formula-leading and control-sequence cells, AlignedSent DOT labels are escaped, the WordNet browser escapes corpus text, and nltk.data.load's verbose output no longer prints the full resource URL (PR #3914, #3915, #3937, #3946, #3931, #3930, #3856, #3902)
• Corpus stores and encodings: the PanLex Lite and chat80 sqlite stores open read-only on a pinned descriptor with their schema screened and a statement step budget, chat80 shelves are read by a key manifest, Toolbox SFM markers are sanitised, and a resource encoding that is not a text codec (bz2_codec and friends) is refused (PR #3961, #3964, #3959, #3888, #3800, #3962)
• Package contents: the wheel no longer ships nltk.test.unit (the attack harnesses, advisory probes and their fixtures), so pip install nltk is about half the size and import nltk.test.unit works from a source checkout or the sdist only. The doctests under nltk.test still ship (PR #3954)

Security test and CI infrastructure

• Attack and regression harnesses, advisory probes, the sandbox conftest fixtures, the pre-commit guards (no unsandboxed open, every regex through redos, every pickle through picklesec, every print through safe_print), load-invariant CPU-time timing rules, CI matrix and runner fixes, pinned third-party downloads, and the AUTHORS update (PR #3744, #3749, #3762, #3764, #3770, #3776, #3791, #3795, #3803, #3818, #3820, #3821, #3822, #3849, #3855, #3860, #3861, #3862, #3864, #3865, #3866, #3876, #3884, #3899, #3904, #3905, #3912, #3917, #3924, #3926, #3927, #3939, #3940, #3943, #3948, #3949, #3963, #3968, #3971, #3976, #3982, #3983, #3984, #3985)

Other changes

• Added RTL language syntactic tree visualization support (PR #3472)
• Fix #3394: Ensure punkt_tab is downloaded when punkt is requested (PR #3505)
• Show README.md as the PyPI long description (PR #3746)
• Add stopwords.words("english-extended") (PR #3747)
• Krippendorff's alpha: treat placeholder labels as missing data (#2865, #2732) (PR #3752)
• Fix shared defaults in chat80 Concept (PR #3773)
• Fix Java wrappers decoding str output from java() (universal_newlines) (PR #3777)
• Prevent IndexError in Hungarian stemmer on empty string (PR #3785)
• Isolate MTE tag filters per view (PR #3792)
• Fix newline in ellipsis tokens in TweetTokenizer (PR #3819)
• Isolate default WordNet reference relations (PR #3872)
• Fix resolution paths that cancel distinct literals (PR #3878)
• Fix UTF-8 text preprocessing in NKJP corpus reader (PR #3879)
• Remove shadowed FeatStruct._walk stub and preserve its docstring (PR #3893)
• Count all increasing pairs in RIBES kendall_tau (PR #3900)
• Fix the TweetTokenizer ZWJ emoji pattern capturing adjacent whitespace (PR #3913)
• Surface the CoreNLP server's reason for a non-2xx answer (PR #3942)
• Omit unaligned (NULL target) pairs from Alignment strings, so IBM model output prints (PR #3967)
• Fix DependencyGraph cycle detection, which returned False on every parsed graph (PR #3978)
• Show, range and draw an alignment with an unaligned pair on every surface (PR #3987)

Thanks to the following contributors to 3.10.4: Zhewen Tan, Yanhaoxi, Yazan Balawneh (Cystack.ps), Abdulmoyn Sommakieh, Golitsin Vyacheslav, Mingzhe Yu, 宁岑, Tai An, 区梓灏, Tom Aarsen, Raffi Enficiaud,

... (truncated)

Commits

• 5c41fc0 Merge pull request #3990 from alvations/deflake-span-teeth-control-20261021
• a517c64 Make the span_tokenize teeth control compute-bound so Windows reads its quadr...
• b8cfc96 Merge pull request #3989 from alvations/release-notes-thanks-20261021
• dd1254e Merge pull request #3988 from alvations/harness-3913-zwj-20261021
• 8fd879f Credit every contributor since 3.10.3 in the release notes and AUTHORS.md
• 7e33e78 Merge remote...

  Description has been truncated

[timeleft--]

Independent technical review at exactcbad10044b6ee4c92d54ff54191ec567ad26bde5/currentmain1754bdd: net3files are lock dependency closure and synchronized0.9.3 patch mirrors. Current-base repository Version policy PASS. Separate checkout/frozen install and92 release-channel/MCP/context/registration/packaged-MCP tests PASS (60.51s). Worker full1233tests on bothJJversions and actual hosted matrices also pass; no tests weakened.

Primary filelock4.0.0 release changes SoftReadWriteLock disk protocol; this optional transitive FAVA graph does not exercise that API or require migrating active lock files. Setuptools84 archive hardening is an optional torch build dependency; FAVA builds with hatchling. Runtime certifi/OTel and optional NLTK edges stay within actual consumer constraints, no direct FAVA API change identified. Upstream major numbering is not a FAVA major-release/install decision. Configured independent native review remains pending; this technical comment does not waive it or authorize publication/installation.

[yia-mw-agent]

Reviewed the full merge-base-to-head diff (base main, merge-base 1754bdd = current main) at cbad100. No material defects.

Helmet-Review-Request: #156 (comment)

Scope: 3 files. The Dependabot uv-group bump (certifi 2026.10.19, filelock 4.2.1, nltk 3.10.4, opentelemetry-api 1.45.1 with the matching opentelemetry-* family, setuptools 84.1.0) is confined to uv.lock. The version bump 0.9.2 -> 0.9.3 appears in pyproject.toml, src/fava_trails/init.py and the fava-trails entry in uv.lock.

Checks I ran at the exact head:

• Lock consistency: uv lock --check resolves with no change to the lock.
• Version policy: scripts/mw-version.py check --base origin/main reports "repo: 0.9.2 -> 0.9.3, Version policy passed."
• Full uv run --frozen pytest: 1230 passed. The 3 tests in tests/test_packaged_mcp.py failed in my first run only because uv was not on PATH in that harness. Rerun with uv on PATH, all 3 pass.
• GitHub checks at this head: CodeQL, Validate PR, Version policy and the test matrix (jj 0.28.0 and 0.45.1) all pass.

Notes (non-blocking):

• No direct dependency in pyproject.toml pins any of the five bumped packages, so only transitive resolution changes. The lock drops importlib-metadata and zipp, which opentelemetry-api 1.45.1 no longer needs, and adds opentelemetry-exporter-http-transport and opentelemetry-exporter-otlp-common as new transitive packages. This is consistent with the upstream metadata.
• filelock moves a major version (3.x -> 4.x). The test suite, including the packaged-wheel install and upgrade tests, exercises it without failures.