fix: remove the shared indentation of the commit body

docs/configuration.md

@@ -157,6 +157,8 @@ For compatibility, cmp-git also supports `filter_fn` on a provider, as an alias
 
 - `sha` has `git.commits.sha_length` characters (default 7). The default `insertText` inserts `sha`. The default
   `label` shows the first 7 characters.
+- `description` is the body of the commit, without the blank lines at its start and the whitespace at its end. When
+  all lines with text share an indentation, that indentation is removed, and lines with only whitespace become empty.
 - `author_name` and `author_mail` are the name and the email of the author, with the `.mailmap` applied.
   `commit_timestamp` is the commit date.
 - GitHub issues and pull requests always have `body` and `updatedAt`.

lua/cmp_git/provider/capabilities/commits.lua

@@ -4,13 +4,18 @@ local utils = require("cmp_git.utils")
 
 local M = {}
 
+local byte = string.byte
 local find = string.find
+local match = string.match
 local sub = string.sub
 local trim_range = utils.trim_range
 
 -- The NUL-terminated fields of each commit in the `M.request` format
 local FIELDS = 6
 
+-- The bytes of an indentation: tab and space
+local INDENT = { [9] = true, [32] = true }
+
 ---@param config cmp_git.Config.GitCommits
 ---@return cmp_git.CommandSpec
 function M.request(config)
@@ -34,6 +39,53 @@ local function trimmed(s, first, last)
     return sub(s, trim_range(s, first, last))
 end
 
+---The trimmed body without the indentation that all its lines with text share, as `textwrap.dedent` of Python does.
+---Some bodies are indented as a whole, and the trim removes the indentation of the first line only. Lines with only
+---whitespace do not count, and they become empty when an indentation is removed. A tab and a space do not match.
+---Without a shared indentation, the result is the trimmed body.
+---@param s string
+---@param first integer
+---@param last integer
+---@return string
+local function dedented(s, first, last)
+    local field_first = first
+    first, last = trim_range(s, first, last)
+    if first > last then
+        return ""
+    end
+
+    local line_first = first
+    while line_first > field_first and INDENT[byte(s, line_first - 1)] do
+        line_first = line_first - 1
+    end
+    -- The first line has text, so the shared indentation is a part of its indentation
+    if line_first == first then
+        return sub(s, first, last)
+    end
+
+    local lines = vim.split(sub(s, line_first, last), "\n", { plain = true })
+    local margin = sub(s, line_first, first - 1)
+    for _, line in ipairs(lines) do
+        if find(line, "%S") then
+            local indent = match(line, "^[ \t]*")
+            local shared = 0
+            while shared < #margin and byte(margin, shared + 1) == byte(indent, shared + 1) do
+                shared = shared + 1
+            end
+            if shared == 0 then
+                return sub(s, first, last)
+            end
+            margin = sub(margin, 1, shared)
+        end
+    end
+
+    local width = #margin
+    for i, line in ipairs(lines) do
+        lines[i] = find(line, "%S") and sub(line, width + 1) or ""
+    end
+    return table.concat(lines, "\n")
+end
+
 ---@param result string
 ---@param config cmp_git.Config.GitCommits
 ---@return cmp_git.Commit[]
@@ -42,7 +94,7 @@ function M.parse(result, config)
     local now = os.time()
     local sha_length = config.sha_length
     -- The position of the NUL after each field of the current commit. The fields are read in place, so that each
-    -- field is copied once, after the trim.
+    -- field is copied once, after the trim. `dedented` also splits a body whose first line is indented into lines.
     local stops = {}
     local start = 1
 
@@ -73,7 +125,7 @@ function M.parse(result, config)
         local commit = {
             sha = sha,
             title = trimmed(result, stops[1] + 1, stops[2] - 1),
-            description = trimmed(result, stops[2] + 1, stops[3] - 1),
+            description = dedented(result, stops[2] + 1, stops[3] - 1),
             author_name = sub(result, stops[3] + 1, stops[4] - 1),
             author_mail = sub(result, stops[4] + 1, stops[5] - 1),
             commit_timestamp = timestamp,

tests/provider/capabilities/test_commits.lua

@@ -73,11 +73,67 @@ T["commit_capability_parse_marker_collision_and_malformed_records"] = function()
     eq(commits[1].description, "body with ###CMP_GIT_END### marker", "commit parser allows old entry marker text")
 end
 
--- The parser before it read the fields in place. The test below checks that the new parser gives the same output.
+T["commit_capability_parse_dedents_the_body"] = function()
+    local function description(body)
+        local raw = commit_record("abcdef1234567890", "subject", body, "Ada", "ada@example.com", 1234567890)
+        return commit_capability.parse(raw, commit_test_config())[1].description
+    end
+
+    eq(
+        description("  ## what\n  - a\n\n  ## how\n  - enable it if\n    it is on PATH\n"),
+        "## what\n- a\n\n## how\n- enable it if\n  it is on PATH",
+        "a body that is indented as a whole keeps its relative indentation"
+    )
+    eq(description("  first\n    second"), "first\n  second", "the first line has the shared indentation")
+    eq(description("    first\n  second"), "  first\nsecond", "a later line has the shared indentation")
+    eq(description("\t\tfirst\n\t\t\tsecond"), "first\n\tsecond", "tabs are an indentation")
+    eq(description("first\n  second"), "first\n  second", "a first line without indentation")
+    eq(description("  first\nsecond"), "first\nsecond", "a later line without indentation")
+    eq(description("\tfirst\n  second"), "first\n  second", "a tab and a space do not match")
+    eq(description("  first\n \t \n\n  second"), "first\n\n\nsecond", "lines with only whitespace become empty")
+    eq(
+        description("first\n \t \n  second"),
+        "first\n \t \n  second",
+        "lines with only whitespace stay without a dedent"
+    )
+    eq(description("\n \n\t\n  first\n  second\n\n"), "first\nsecond", "blank lines around the body are removed")
+    eq(description(" \r first\n  second"), "first\n second", "the indentation starts after other whitespace")
+    eq(description(""), "", "an empty body")
+    eq(description(" \n\t "), "", "a body with only whitespace")
+end
+
+-- The parser before it read the fields in place, with the dedent of the body. The test below checks that the new
+-- parser gives the same output.
 local function old_trim(s)
     return (s:gsub("^%s*(.-)%s*$", "%1"))
 end
 
+---The body without the indentation that all its lines with text share, written with line lists instead of the index
+---arithmetic of the parser. Without a shared indentation, it is the trimmed body.
+local function old_body(s)
+    local trimmed = old_trim(s)
+    -- The spaces and tabs right before the first byte that is not whitespace
+    local indent = s:match("([ \t]*)%S") or ""
+    local lines = vim.split(indent .. trimmed, "\n", { plain = true })
+    local margin
+    for _, line in ipairs(lines) do
+        if line:find("%S") then
+            local lead = line:match("^[ \t]*")
+            margin = margin or lead
+            while lead:sub(1, #margin) ~= margin do
+                margin = margin:sub(1, -2)
+            end
+        end
+    end
+    if not margin or margin == "" then
+        return trimmed
+    end
+    for i, line in ipairs(lines) do
+        lines[i] = line:find("%S") and line:sub(#margin + 1) or ""
+    end
+    return table.concat(lines, "\n")
+end
+
 local function old_split_nul(input)
     local fields = {}
     local start = 1
@@ -102,7 +158,7 @@ local function old_parse(result, config)
             table.insert(commits, {
                 sha = old_trim(fields[i]):sub(1, config.sha_length),
                 title = old_trim(fields[i + 1]),
-                description = old_trim(fields[i + 2]),
+                description = old_body(fields[i + 2]),
                 author_name = fields[i + 3] or "",
                 author_mail = fields[i + 4] or "",
                 commit_timestamp = timestamp,
@@ -146,6 +202,11 @@ T["commit_capability_parse_matches_the_old_parser"] = function()
         "Ünïcödé",
         "\031",
         "a" .. string.rep(" ", 50) .. "b",
+        "  ## what\n  - a\n \t \n    b\n",
+        "  indented\n",
+        "\n  - after a blank line\n    more\n",
+        "\t- tab line\n",
+        "   \n",
     }
     local function text()
         local parts = {}

tests/sources/test_git.lua

@@ -86,6 +86,60 @@ T["git_requests_answer_in_a_repository_without_commits"] = function()
     eq(next(git.cache_trailer_values), nil, "trailer values are not cached")
 end
 
+T["git_commit_documentation_keeps_the_body_layout"] = function()
+    if vim.fn.executable("git") ~= 1 then
+        helpers.skip("git is not installed")
+    end
+
+    -- The body of issue #54. It is indented as a whole, and one line is indented more.
+    local repo = helpers.git_repo()
+    repo.commit(table.concat({
+        "chore(astronvim): auto install Treesitter language packs",
+        "",
+        "  ## what",
+        "  - auto install Treesitter language packs",
+        "",
+        "  ## how",
+        "  - enable config `auto_install` if `tree-sitter` is installed and",
+        "    available on PATH",
+    }, "\n"))
+    vim.fn.chdir(repo.dir)
+
+    local git = Git.new(config.normalize({}).git)
+    local line = "See :"
+    local context = {
+        bufnr = helpers.buffer({ "feat: subject", "", line }),
+        cursor = { row = 3, character = #line },
+        line = line,
+    }
+    local lists = {}
+    git:get_commits(function(list)
+        table.insert(lists, list)
+    end, context, ":")
+    helpers.wait_for(function()
+        return #lists > 0
+    end, "commits call back")
+
+    eq(#lists, 1, "commits call back once")
+    eq(#lists[1].items, 1, "commits give the commit")
+    local item = lists[1].items[1]
+    eq(item.documentation, {
+        kind = "markdown",
+        value = table.concat({
+            "# chore(astronvim): auto install Treesitter language packs",
+            "",
+            "## what",
+            "- auto install Treesitter language packs",
+            "",
+            "## how",
+            "- enable config `auto_install` if `tree-sitter` is installed and",
+            "  available on PATH",
+            "",
+            "Authored by Ada Lovelace (ada@example.com), committed on " .. os.date("%c", item.data.commit_timestamp),
+        }, "\n"),
+    }, "the documentation keeps the lines of the body without their shared indentation")
+end
+
 T["git_requests_answer_without_git"] = function()
     -- `command.build` gives no job when git is not installed
     helpers.stub(vim.fn, "executable", function()