feat: Support context recognition for injected languages #388
Conversation
|
@lewis6991, any chance you could review this pull request anytime soon? I'd like to add Svelte support as well, but without this PR, it doesn't make much sense. Thanks! |
|
Can you rebase and add a functional test to |
kwaszczuk
force-pushed
the
respect-injections
branch
from
f5d1d83 to
826999e
Compare
kwaszczuk
force-pushed
the
respect-injections
branch
from
826999e to
3c78f63
Compare
|
Done. I've added one test for the plain HTML without any injections, and two more for JavaScript within |
| ---@param row integer | ||
| ---@param col integer | ||
| ---@return LanguageTree[] | ||
| local function get_parent_langtrees(bufnr, range) |
There was a problem hiding this comment.
Does this work properly for nested injections? it looks to me this will only return at most 2 langtrees. The root and the most inner.
There was a problem hiding this comment.
You are right, I misunderstood language_for_range implementation. I changed the code to manually traverse children langtrees, please take a look.
To test the nested injection, I came up with the markdown test case replacing HTML one. Context both for <html> and <script> displays properly when I run vim manually, but in tests injecting language in Markdown seems to be broken (look at test job for the reference). Do you have any idea what could be the problem here?
There was a problem hiding this comment.
@lewis6991 It took me a while, but I finally get it to work.
Looks like tests are not performing full initialization of nvim-treesitter plugin, but just require'nvim-treesitter.configs'.setup { ... }. Because of that some modules are not initialized, including nvim-treesitter.query_predicates which contains set-lang-from-info-string! directive configuration required by Markdown queries.
My straightforward solution for that is to explicitly initialize the plugin via require'nvim-treesitter'.setup(). I don't think it is the greatest solution, as we end up invoking setup() twice (just on different modules), which seems weird, but I couldn't find any other way that would not induce setup duplication.
kwaszczuk
force-pushed
the
respect-injections
branch
from
37bbfe8 to
02f27f8
Compare
kwaszczuk
force-pushed
the
respect-injections
branch
from
792381c to
c17f9b6
Compare
|
Thanks. Let go with this for now. Can tidy up stuff later if we need to. |
Currently, the plugin only recognizes the outermost language of a file, disregarding any injected languages within the file. A notable example are
.htmlfiles, where JavaScript content within<script>tags is ignored.The root cause of this behavior lies in the fact that, during context evaluation for a line, only the root language tree is considered. Fortunately, the
nvim-treesitterAPI provides a solution by allowing us to retrieve all language trees associated with a given range (LanguageTree:language_for_range({range})).This PR proposes a simple yet effective change: iterating over parent nodes of each language tree detected for the range, as opposed to doing so only for the root language tree.
I believe, introduced functionality aims to enhance the plugin's behavior without overcomplicating the source code. It particularly improves compatibility with JavaScript frameworks like Svelte or Vue, where language injection is common, if not inevitable.
To illustrate the impact of this PR, below there is a comparison of the plugin's behavior on
test/test.htmlfile before (left side) and after introducing the changes (right side):