These are not needed after #35129 but making uncrustify still play nice
with them was a bit tricky.
Unfortunately `uncrustify --update-config-with-doc` breaks strings
with backslashes. This issue has been reported upstream,
and in the meanwhile auto-update on every single run has been disabled.
Problem:
The previous fix in #34314 relies on copying the tree in `tree_root` to
ensure the `TSNode`'s tree cannot be mutated. But that causes the
problem where two calls to `tree_root` return nodes from different
copies of a tree, which do not compare as equal. This has broken at
least one plugin.
Solution:
Make all `TSTree`s on the Lua side always immutable, avoiding the need
to copy the tree in `tree_root`, and make the only mutation point,
`tree_edit`, copy the tree instead.
Problem:
TSNode contains a `const TSTree*` and a `const void *id`. The `id`
points to Tree-sitter's internal type `Subtree`, which resides inside
the `TSTree` but may be deallocated if the `TSTree` is mutated (which
is likely why it is `const`).
The Lua method `TSTree:edit()` mutates the tree, which can deallocate
`id`.
See #25254 and #31758.
Solution:
To avoid this, we now make a copy of the tree before pushing its root to
the Lua stack. This also removes the fenv from TSLuaTree, as it was only
used when pushing the tree root to the Lua stack.
We also copy the tree in `node_tree`.
`ts_tree_copy()` just increments a couple of reference counters, so it's
relatively cheap to call.
Apparently after parsing with options in tree-sitter, the options data
persists in the parser object, and thus successive calls to
`ts_parser_parse()` will act like `ts_parser_parse_with_options()`. This
is problematic because `languagetree.lua` makes coroutine-environment
assumptions based on if a nullptr has been returned by the parser
function. This commit makes it so that the parse options state is reset
upon a regular parse (would be nice if this was done upstream).
Fixes#33277
Remove the `set_timeout` functions for `TSParser` and instead add a timeout
parameter to the regular parse function. Remove these deprecated tree-sitter
API functions and replace them with the preferred `TSParseOptions` style.
Problem: Cannot disable individual captures and patterns in treesitter queries.
Solution:
* Expose the corresponding tree-sitter API functions for `TSQuery` object.
* Add documentation for `TSQuery`.
* Return the pattern ID from `get_captures_at_pos()` (and hence `:Inspect!`).
Problem: No way to check the version of a treesitter parser.
Solution: Add version metadata (ABI 15 parsers only) as well as parser state count and supertype information (ABI 15) in `vim.treesitter.language.inspect()`. Also graduate the `abi_version` field, as this is now the official upstream name.
---------
Co-authored-by: Christian Clason <c.clason@uni-graz.at>
Problem:
When running an initial parse, parse() returns an empty table rather
than an actual range. In `languagetree.lua`, we manually check if
a parse was incremental to determine the changed parse region.
Solution:
- Always return a range (in the C side) from parse().
- Simplify the language tree code a bit.
- Logger no longer shows empty ranges on the initial parse.
**Problem:** Parsing can be slow for large files, and it is a blocking
operation which can be disruptive and annoying.
**Solution:** Provide a function for asynchronous parsing, which accepts
a callback to be run after parsing completes.
Co-authored-by: Lewis Russell <lewis6991@gmail.com>
Co-authored-by: Luuk van Baal <luukvbaal@gmail.com>
Co-authored-by: VanaIgr <vanaigranov@gmail.com>
Regression from 2a7d0ed614, which removed
header that is only needed if wasmtime support is enabled. Prevent this
from happening again by wrapping the include in a `HAVE_WASMTIME` check.
**Problem:** Currently node names with non-alphanumeric, non
underscore/hyphen characters (only possible with anonymous nodes) are
not given a proper error message. See tree-sitter issue 3892 for more
details.
**Solution:** Apply a different scanning logic to anonymous nodes to
correctly identify the entire node name (i.e., up until the final double
quote)
**Problem:** Tree-sitter 0.24.0 introduced a new symbol type to denote
supertype nodes (`TSSymbolTypeSupertype`). Now, `language.inspect()`
(and the query `omnifunc`) return supertype symbols, but with double
quotes around them.
**Solution:** Mark a symbol as "named" based on it *not* being an
anonymous node, rather than checking that it is a regular node (which a
supertype also is not).
**Problems:**
- `vim.treesitter.language.inspect()` returns duplicate
symbol names, sometimes up to 6 of one kind in the case of `markdown`
- The list-like `symbols` table can have holes and is thus not even a
valid msgpack table anyway, mentioned in a test
**Solution:** Return symbols as a map, rather than a list, where field
names are the names of the symbol. The boolean value associated with the
field encodes whether or not the symbol is named.
Note that anonymous nodes are surrounded with double quotes (`"`) to
prevent potential collisions with named counterparts that have the same
identifier.
This commit also marks `child_containing_descendant()` as deprecated
(per upstream's documentation), and uses `child_with_descendant()` in
its place. Minimum required tree-sitter version will now be `0.24`.
Problem:
Tree cursors can only be efficient when they are re-used.
Short-lived cursors are very slow.
Solution:
Reimplement functions that use short-lived cursors.
Problem: Installing treesitter parser is hard (harder than
climbing to heaven).
Solution: Add optional support for wasm parsers with `wasmtime`.
Notes:
* Needs to be enabled by setting `ENABLE_WASMTIME` for tree-sitter and
Neovim. Build with
`make CMAKE_EXTRA_FLAGS=-DENABLE_WASMTIME=ON
DEPS_CMAKE_FLAGS=-DENABLE_WASMTIME=ON`
* Adds optional Rust (obviously) and C11 dependencies.
* Wasmtime comes with a lot of features that can negatively affect
Neovim performance due to library and symbol table size. Make sure to
build with minimal features and full LTO.
* To reduce re-compilation times, install `sccache` and build with
`RUSTC_WRAPPER=<path/to/sccache> make ...`
Problem: `has-ancestor?` is O(n²) for the depth of the tree since it iterates over each of the node's ancestors (bottom-up), and each ancestor takes O(n) time.
This happens because tree-sitter's nodes don't store their parent nodes, and the tree is searched (top-down) each time a new parent is requested.
Solution: Make use of new `ts_node_child_containing_descendant()` in tree-sitter v0.22.6 (which is now the minimum required version) to rewrite the `has-ancestor?` predicate in C to become O(n).
For a sample file, decreases the time taken by `has-ancestor?` from 360ms to 6ms.
Instead of painfully messing with timing to determine if queries were
reparsed, we can simply keep a counter next to the call to ts_query_new
Also memoization had a hidden dependency on the garbage collection of
the the key, a hash value which never is kept around in memory. this was
done intentionally as the hash does not capture all relevant state for the
query (external included files) even if actual query objects still
would be reachable in memory. To make the test fully deterministic in
CI, we explicitly control GC.
Problem:
`TSNode:_rawquery()` is complicated, has known issues and the Lua and
C code is awkwardly coupled (see logic with `active`).
Solution:
- Add `TSQueryCursor` and `TSQueryMatch` bindings.
- Replace `TSNode:_rawquery()` with `TSQueryCursor:next_capture()` and `TSQueryCursor:next_match()`
- Do more stuff in Lua
- API for `Query:iter_captures()` and `Query:iter_matches()` remains the same.
- `treesitter.c` no longer contains any logic related to predicates.
- Add `match_limit` option to `iter_matches()`. Default is still 256.
Query patterns can contain quantifiers (e.g. (foo)+ @bar), so a single
capture can map to multiple nodes. The iter_matches API can not handle
this situation because the match table incorrectly maps capture indices
to a single node instead of to an array of nodes.
The match table should be updated to map capture indices to an array of
nodes. However, this is a massively breaking change, so must be done
with a proper deprecation period.
`iter_matches`, `add_predicate` and `add_directive` must opt-in to the
correct behavior for backward compatibility. This is done with a new
"all" option. This option will become the default and removed after the
0.10 release.
Co-authored-by: Christian Clason <c.clason@uni-graz.at>
Co-authored-by: MDeiml <matthias@deiml.net>
Co-authored-by: Gregory Anders <greg@gpanders.com>
- `TSQuery`: userdata object for parsed query.
- `vim.treesitter.Query`: renamed from `Query`.
- Add a new field `lang`.
- `TSQueryInfo`:
- Move to `vim/treesitter/_meta.lua`, because C code owns it.
- Correct typing for `patterns`, should be a map from `integer`
(pattern_id) to `(integer|string)[][]` (list of predicates or
directives).
- `vim.treesitter.QueryInfo` is added.
- This currently has the same structure as `TSQueryInfo` (exported
from C code).
- Document the fields (see `TSQuery:inspect`).
- Add typing for `vim._ts_parse_query()`.
We already have an extensive suite of static analysis tools we use,
which causes a fair bit of redundancy as we get duplicate warnings. PVS
is also prone to give false warnings which creates a lot of work to
identify and disable.
long is 32 bits on windows, while it is 64 bits on other architectures.
This makes the type suboptimal for a codebase meant to be
cross-platform. Replace it with more appropriate integer types.
This involves two redesigns of the map.c implementations:
1. Change of macro style and code organization
The old khash.h and map.c implementation used huge #define blocks with a
lot of backslash line continuations.
This instead uses the "implementation file" .c.h pattern. Such a file is
meant to be included multiple times, with different macros set prior to
inclusion as parameters. we already use this pattern e.g. for
eval/typval_encode.c.h to implement different typval encoders reusing a
similar structure.
We can structure this code into two parts. one that only depends on key
type and is enough to implement sets, and one which depends on both key
and value to implement maps (as a wrapper around sets, with an added
value[] array)
2. Separate the main hash buckets from the key / value arrays
Change the hack buckets to only contain an index into separate key /
value arrays
This is a common pattern in modern, state of the art hashmap
implementations. Even though this leads to one more allocated array, it
is this often is a net reduction of memory consumption. Consider
key+value consuming at least 12 bytes per pair. On average, we will have
twice as many buckets per item.
Thus old implementation:
2*12 = 24 bytes per item
New implementation
1*12 + 2*4 = 20 bytes per item
And the difference gets bigger with larger items.
One might think we have pulled a fast one here, as wouldn't the average size of
the new key/value arrays be 1.5 slots per items due to amortized grows?
But remember, these arrays are fully dense, and thus the accessed memory,
measured in _cache lines_, the unit which actually matters, will be the
fully used memory but just rounded up to the nearest cache line
boundary.
This has some other interesting properties, such as an insert-only
set/map will be fully ordered by insert only. Preserving this ordering
in face of deletions is more tricky tho. As we currently don't use
ordered maps, the "delete" operation maintains compactness of the item
arrays in the simplest way by breaking the ordering. It would be
possible to implement an order-preserving delete although at some cost,
like allowing the items array to become non-dense until the next rehash.
Finally, in face of these two major changes, all code used in khash.h
has been integrated into map.c and friends. Given the heavy edits it
makes no sense to "layer" the code into a vendored and a wrapper part.
Rather, the layered cake follows the specialization depth: code shared
for all maps, code specialized to a key type (and its equivalence
relation), and finally code specialized to value+key type.
