`Parser::set_logger` boxes the logger callback into a type-erased C
pointer, but `Parser` carries no lifetime parameter. This allows for
two kinds of UB:
- Use-after-free: a logger could capture a non-`'static` borrow, let the
borrowed value drop, and then dangle the next time the parser logged.
- Data race: a logger could capture a `!Send` value such as an `Rc`. The
parser could then be moved to another thread and logged from there
while the original thread still held a clone, racing on the reference
count.
As a fix, we just require that logger is `Send + 'static`. The
alternative here is to attach a lifetime parameter to `Parser`, but this
is highly breaking for what is mostly a debugging utility. As an escape
hatch `set_logger_unchecked` is added to the public API as an `unsafe fn`
to correctly communicate the risks and invariants that must be held.
`(P . Q* Y)` with zero `Q` dropped the leading `.`, so `Y` matched at any
position instead of being pinned to the parent's first named child. On a `?`/`*`
zero-skip, when the skipped step is the parent's first child and carries a
leading anchor, transfer that first-child requirement to the skip target.
on both sides
In `A . Q* . B`, a zero-matched `Q` made both anchors vacuous, so `A` and `B`
were no longer required to be immediate siblings. Only relax the following
anchor on a `?`/`*` zero-skip when the skipped step has no leading anchor of
its own; otherwise the adjacency transfers through the empty run.
Problem: A set of `array_*` macros (`array_push`, `array_extend`, etc) implicitly convert a `void*` into a different pointer type. In environments that compile these headers as C++, this implicit conversion is an error.
Solution: This commit adds an `_array_cast` macro that uses `decltype` to cast the `void*` to the proper type when compiling as C++.
A trailing `.` after an optional node, e.g. `(p (a)+ @a . (b)? @b .)`, sets
`is_last_child` on the `(b)?` step. When `(b)?` matched zero, the zero-skip
jumped past that step to completion, so the last-child requirement was never
enforced.
When a zero-skip would bypass a step carrying `is_last_child`, require that
the last matched node really is the last named child. Gated on the
`alternative_is_skip` edge marker.
A `.` anchor between a quantified pattern and a following node, e.g.
`(parent (comment)* @c . (decl))`, was treated as a leading anchor when the
quantifier matched zero. The zero-skip jumps a state directly onto the anchored
step, where `is_immediate` was enforced even though no sibling had matched before
it. Name that skip edge (`alternative_is_skip`) and, when a state follows it,
record that it skipped the quantifier (`skipped_quantifier`). The
immediate anchor is then supressoed for that state's next match.
An unanchored quantified sibling like `(program (comment)+ @doc (class))`
keeps one match state per matching sibling. A recent fix stopped over-pruning
them, but the per-node longest-match dedup is pairwise, so with n live states
matching became O(n^3).
Two states can only be capture subsets of one another if their captured
byte ranges overlap, so keep each group ordered by first-capture position
and stop the pairwise scan once the rest of the group is disjoint.
Refs neovim/neovim#40517
object rather than 4 separate optional function pointers.
Helps prevent misuse via mixing different allocators. Also update the
doc comment with relevant safety information.
This allows duplicating a query so that it can be modified by using
disable_capture or disable_pattern, without re-parsing the query. The
new `ts_query_copy` creates a deep copy of the entire query object.
A motivation for this is tags.scm code navigation queries. You might
want to have one version of the compiled query capture only definitions
and the other only capture references, since references are much much
more common than definitions in a typical codebase. Instead of
re-parsing and analyzing the query and then calling
`ts_query_disable_capture`, we can clone the built query all-at-once and
then disable captures / patterns.
Problem: In queries matching a parent node with anchored children
(siblings) where the first sibling has a quantifier and is not
captured, the check for fallible steps skips splitting the state because
the next node is a passthrough node (and not an is_immediate one). This
prevents the query from matching beyond the first occurrence.
Solution: In the check for fallible steps, skip the next steps if they
are passthrough steps.
Problem: In queries matching a parent node with anchored children
(siblings) where the first anchored sibling has a quantifier, the
deduplication logic for states is overly aggressive. The logic causes
the state split on the first capture (with the parent) to be dropped in
favor of the loopback state. This results in the query not being able to
match beyond the first occurrence.
Solution: Prevent the deduplication logic from dropping a state that is
not seeking an immediate match for one that is seeking an immediate
match, since the one not seeking for an immediate match could still
match later nodes.
Problem: In queries matching a parent node with anchored children
(siblings), the check for fallible steps only considers nodes with
children, which results in no state split being made for anchored
siblings (node1) . (node2). This prevents the query from matching beyond
the first occurrence.
Solution: Make the check for fallible steps consider also the case where
the next step is at the same depth and must be matched immediately
after.
parser can reuse a node.
Lookahead bytes can be used to decide what a node is parsed as, so it's
resaonable to consider this as part of a node's "range" when deciding
which edits affect it.
The alloc.h header checked for TREE_SITTER_HIDDEN_SYMBOLS, but the
canonical macro name used everywhere else (api.h, render.rs, setup.py)
is TREE_SITTER_HIDE_SYMBOLS. This mismatch meant that defining
TREE_SITTER_HIDE_SYMBOLS (as the Python binding build does) would not
actually hide the allocator symbols in alloc.h.
Fixestree-sitter/tree-sitter#5625
Signed-off-by: Georges Savoundararadj <savoundg@amazon.com>
The wasm store gated supertype_symbols / supertype_map_slices /
supertype_map_entries copies on abi_version > LANGUAGE_VERSION_WITH_RESERVED_WORDS,
but every other consumer (language.c, query.c) treats those tables as
present when abi_version >= LANGUAGE_VERSION_WITH_RESERVED_WORDS.
A Wasm grammar built at ABI exactly 15 with supertype_count > 0 ends up
with supertype_count copied into the native TSLanguage but supertype_map_slices
left NULL. ts_query__analyze_patterns then calls ts_language_subtypes,
which dereferences self->supertype_map_slices[supertype] and crashes.
`ts_query_cursor_next_capture` linearly scanned all finished states to
find the one with the earliest next capture byte offset. With deeply
nested code, this O(n) scan per capture caused the highlight crate to
hang for minutes on large files.
Replace the linear scan with a min-heap over the finished_states array,
keyed by (next_capture_byte_offset, pattern_index, id). The heap is
maintained lazily: ts_query_cursor__advance uses plain array_push
(preserving FIFO insertion order), and next_capture sifts new elements
into the heap on entry via a tracked heap_size boundary. This preserves
the documented "order found" guarantee for next_match while giving
next_capture O(log n) per call.
Commit 1f6eac55 ("query: Use uint32_t for capture list IDs") widened
QueryState.capture_list_id to uint32_t and removed the 65536 pool cap,
but left the pool function signatures as uint16_t. This caused silent
truncation when the pool exceeded 65535 entries, leading to a segfault.
Solaris does not provide <endian.h> or <sys/endian.h>, but it does expose
byte-order definitions and conversion helpers via <sys/isa_defs.h> and
<sys/byteorder.h>.
Add a __sun branch so the portable header defines __BYTE_ORDER and the
htobe*/le*toh conversions on Solaris.
Co-authored-by: Amaan Qureshi <git@amaanq.com>
`ts_decode_utf16_le` and `ts_decode_utf16_be` passed a byte length to
`U16_NEXT_LE` and `U16_NEXT_BE`, but those macros count `uint16_t` code
units. If a lead surrogate was the last code unit in a chunk, the decoder
could peek past the chunk and combine it with adjacent memory.
This commit passes the code-unit length to the UTF-16 macros, and fails
with `TS_DECODE_ERROR` when a chunk is too short to contain even one full
code unit. This lets the lexer retry with a fresh chunk or advance through
the invalid byte as it already does.
Co-authored-by: Will Lillis <will.lillis24@gmail.com>
Co-authored-by: Amaan Qureshi <git@amaanq.com>
- Cache tree byte range length to avoid redundant FFI calls per test
- Combine 5 separate XML attribute write! calls into one
- Pre-allocate format_sexp output buffer to avoid repeated growth
The `LookaheadIterator` visits symbols in group order for small parse
states, which can produce `reduce_actions` in a different order than the
original linear symbol scan (pre c1379718). Since reductions are applied
sequentially and the last reduction version survives, different orderings
lead to different error recovery outcomes (e.g. losing nodes from ERROR
trees).
Sort the reduce_actions array by symbol (descending) after collection.
The array is typically 1-5 entries, so the insertion sort cost is
negligible and the full optimization speedup is preserved.
