diff options
| -rw-r--r-- | doc/advanced_topics.md | 2 | ||||
| -rw-r--r-- | lua/conform/formatters/injected.lua | 54 |
2 files changed, 45 insertions, 11 deletions
diff --git a/doc/advanced_topics.md b/doc/advanced_topics.md index 71aadf9..bc66f4f 100644 --- a/doc/advanced_topics.md +++ b/doc/advanced_topics.md @@ -22,6 +22,8 @@ The way this "aftermarket" range formatting works is conform will format the ent ## Injected language formatting (code blocks) +Requires: Neovim 0.9+ + Sometimes you may have a file that contains small chunks of code in another language. This is most common for markup formats like markdown and neorg, but can theoretically be present in any filetype (for example, embedded SQL queries in a host language). For files like this, it would be nice to be able to format these code chunks using their language-specific formatters. The way that conform supports this is via the `injected` formatter. If you run this formatter on a file, it will use treesitter to parse out the blocks in the file that have different languages and runs the formatters for that filetype (configured with `formatters_by_ft`). The formatters are run in parallel, one job for each language block. diff --git a/lua/conform/formatters/injected.lua b/lua/conform/formatters/injected.lua index 3706ae9..edc2c66 100644 --- a/lua/conform/formatters/injected.lua +++ b/lua/conform/formatters/injected.lua @@ -75,7 +75,8 @@ return { }, condition = function(self, ctx) local ok = pcall(vim.treesitter.get_parser, ctx.buf) - return ok + -- Require Neovim 0.9 because the treesitter API has changed significantly + return ok and vim.fn.has("nvim-0.9") == 1 end, format = function(self, ctx, lines, callback) local conform = require("conform") @@ -93,24 +94,55 @@ return { local options = self.options --- Disable diagnostic to pass the typecheck github action --- This is available on nightly, but not on stable - --- Stable doesn't have any parameters, so it's safe to always pass `true` + --- Stable doesn't have any parameters, so it's safe to always pass `false` ---@diagnostic disable-next-line: redundant-parameter - parser:parse(true) + parser:parse(false) local root_lang = parser:lang() local regions = {} - for lang, child_tree in pairs(parser:children()) do - local formatter_names = conform.formatters_by_ft[lang] - if formatter_names and lang ~= root_lang then - for _, tree in ipairs(child_tree:trees()) do - local root = tree:root() - local start_lnum = root:start() + 1 - local end_lnum = root:end_() - if start_lnum <= end_lnum and in_range(ctx.range, start_lnum, end_lnum) then + + for _, tree in pairs(parser:trees()) do + local root_node = tree:root() + local start_line, _, end_line, _ = root_node:range() + + -- I don't like using these private methods, but critically we do _not_ want to format + -- "combined" injections (they contain the metadata "injection.combined"). These injections + -- will merge all of their regions into a single LanguageTree. If we then try to format the + -- range defined by that LanguageTree, we will likely end up with a range that contains all + -- sorts of content. As a concrete example, consider the following markdown: + -- This is some text + -- <!-- Here is a comment --> + -- Some more text + -- <!-- Another comment --> + -- Since the html injection is combined, the range will contain "Some more text", which is not + -- what we want. + -- To avoid this, don't parse with injections. Instead, we use private methods to run the + -- injection queries ourselves, and then filter out the combined injections. + for _, match, metadata in + ---@diagnostic disable-next-line: invisible + parser._injection_query:iter_matches(root_node, text, start_line, end_line + 1) + do + ---@diagnostic disable-next-line: invisible + local lang, combined, ranges = parser:_get_injection(match, metadata) + local has_formatters = conform.formatters_by_ft[lang] ~= nil + if lang and has_formatters and not combined and #ranges > 0 and lang ~= root_lang then + local start_lnum + local end_lnum + -- Merge all of the ranges into a single range + for _, range in ipairs(ranges) do + if not start_lnum or start_lnum > range[1] + 1 then + start_lnum = range[1] + 1 + end + if not end_lnum or end_lnum < range[4] then + end_lnum = range[4] + end + end + if in_range(ctx.range, start_lnum, end_lnum) then table.insert(regions, { lang, start_lnum, end_lnum }) end end end end + -- Sort from largest start_lnum to smallest table.sort(regions, function(a, b) return a[2] > b[2] |