gityo 1.0.10 → 1.0.11

package/README.md

@@ -154,10 +154,27 @@ Example:
   "autoAcceptMessage": false,
   "postCommand": "push",
   "autoRunPostCommand": false,
-  "instructions": "Write short, clear commit messages."
+  "instructions": "Write short, clear commit messages.",
+  "maxDiffTokens": 24000,
+  "perFileCap": 400
 }
 ```
 
+### Large changes
+
+When the diff is small, it is sent to the model as-is. When it is too large for one request, gityo minimizes it first:
+
+- regenerates the diff with minimal context lines
+- drops lock files and minified/generated files from the payload
+- caps each file's patch and lists every changed file with its line counts so the model still sees the full picture
+
+If the minimized diff is still too large, each remaining part is summarized in parallel and a final commit message is generated from those summaries.
+
+Two optional config knobs control this:
+
+- `maxDiffTokens` — estimated token budget for the diff sent to the model (default `24000`)
+- `perFileCap` — max diff lines kept per file when minimizing (default `400`)
+
 Example instructions file:
 
 ```md

package/dist/index.js

@@ -1710,7 +1710,7 @@ function useColor() {
 	if (process$1.env.FORCE_COLOR || process$1.env.CLICOLOR_FORCE !== void 0) return true;
 }
 new Command();
-var version$4 = "1.0.10";
+var version$4 = "1.0.11";
 Object.freeze({ status: "aborted" });
 function $constructor$3(name$16, initializer$6, params) {
 	function init$1(inst, def) {
@@ -175938,6 +175938,8 @@ const configSchema = object$4({
 	models: record$3(string$6().min(1), modelSchema),
 	autoAcceptMessage: boolean$5(),
 	instructions: string$6().min(1),
+	maxDiffTokens: number$5().int().min(1e3),
+	perFileCap: number$5().int().min(50),
 	postCommand: _enum$4(["push", "push-and-pull"]).nullable(),
 	autoRunPostCommand: boolean$5()
 }).partial();
@@ -175947,6 +175949,8 @@ function resolveConfig(input) {
 		models: parsed.models,
 		instructions: parsed.instructions,
 		autoAcceptCommitMessage: parsed.autoAcceptMessage ?? false,
+		maxDiffTokens: parsed.maxDiffTokens,
+		perFileCap: parsed.perFileCap,
 		postCommand: parsed.postCommand === void 0 ? "push" : parsed.postCommand,
 		autoRunPostCommand: parsed.autoRunPostCommand ?? false
 	};
@@ -181074,17 +181078,23 @@ async function getChangedFiles(git) {
 		...splitNull(untracked)
 	])).sort((left, right) => left.localeCompare(right));
 }
-async function getCommitDiff(git) {
+async function getCommitDiff(git, contextLines = 3) {
+	const unified = [`-U${contextLines}`];
 	const staged = await git.raw([
 		"diff",
 		"--cached",
-		"--no-ext-diff"
+		"--no-ext-diff",
+		...unified
 	]);
 	if (staged.trim().length > 0) return {
 		diff: staged,
 		hasStaged: true
 	};
-	const [unstaged, untracked] = await Promise.all([git.raw(["diff", "--no-ext-diff"]), git.raw([
+	const [unstaged, untracked] = await Promise.all([git.raw([
+		"diff",
+		"--no-ext-diff",
+		...unified
+	]), git.raw([
 		"ls-files",
 		"--others",
 		"--exclude-standard",
@@ -181094,6 +181104,7 @@ async function getCommitDiff(git) {
 		diff: [unstaged, ...await Promise.all(splitNull(untracked).map((file$4) => git.raw([
 			"diff",
 			"--no-index",
+			...unified,
 			"--",
 			"/dev/null",
 			file$4
@@ -181112,20 +181123,178 @@ function createLiveGit(baseDir) {
 	});
 	return git;
 }
+const DEFAULT_MAX_DIFF_TOKENS = 24e3;
+const DEFAULT_PER_FILE_CAP = 400;
+const CHARS_PER_TOKEN = 4;
+const MINIMIZED_CONTEXT_LINES = 1;
+const FILE_SECTION_PREFIX = "diff --git ";
+const HUNK_HEADER_PREFIX = "@@ ";
+const NOISE_FILE_PATTERNS = [/\.(lock|min\.(?:js|mjs|cjs|css)|map)$/i, /-lock\.[^/]+$/i];
+function estimateTokens(text$1) {
+	return Math.ceil(text$1.length / CHARS_PER_TOKEN);
+}
+async function minimizeDiff(git, options) {
+	const { diff } = await getCommitDiff(git, MINIMIZED_CONTEXT_LINES);
+	const sections = splitFileSections(diff).map(parseSection);
+	const kept = [];
+	const excludedPaths = /* @__PURE__ */ new Set();
+	for (const section of sections) if (isNoiseFile(section.path)) excludedPaths.add(section.path);
+	else kept.push(section);
+	return {
+		toc: buildToc(sections, excludedPaths, options.allFiles),
+		body: kept.map((section) => capSection(section, options.perFileCap)).join("\n")
+	};
+}
+function splitDiffIntoChunks(diff, maxTokens) {
+	return packIntoChunks(splitFileSections(diff).flatMap((section) => estimateTokens(section) <= maxTokens ? [section] : splitOversizedSection(section, maxTokens)), maxTokens);
+}
+function isNoiseFile(path$4) {
+	return NOISE_FILE_PATTERNS.some((pattern) => pattern.test(path$4));
+}
+function splitFileSections(diff) {
+	return ("\n" + diff.trimStart()).split("\ndiff --git ").slice(1).map((part) => FILE_SECTION_PREFIX + part);
+}
+function parseSection(section) {
+	let additions = 0;
+	let deletions = 0;
+	let binary = false;
+	for (const line of section.split("\n").slice(1)) {
+		if (line.startsWith("Binary files") || line.startsWith("GIT binary patch")) {
+			binary = true;
+			continue;
+		}
+		if (line.startsWith("+++") || line.startsWith("---")) continue;
+		if (line.startsWith("+")) additions++;
+		else if (line.startsWith("-")) deletions++;
+	}
+	return {
+		path: extractPath(section.split("\n")[0]),
+		content: section,
+		additions,
+		deletions,
+		binary
+	};
+}
+function extractPath(header) {
+	const match = header.match(/^diff --git a\/(.+) b\/(.+)$/);
+	return match ? match[2] : "";
+}
+function capSection(section, cap) {
+	const lines = section.content.split("\n");
+	const bodyLines = lines.length - 1;
+	if (bodyLines <= cap) return section.content;
+	return [...lines.slice(0, cap + 1), `[... ${bodyLines - cap} more lines truncated]`].join("\n");
+}
+function buildToc(sections, excludedPaths, allFiles) {
+	const lines = ["Changed files:"];
+	const seen = /* @__PURE__ */ new Set();
+	for (const section of sections) {
+		seen.add(section.path);
+		if (section.binary) {
+			lines.push(`- ${section.path} (binary)`);
+			continue;
+		}
+		lines.push(`- ${section.path} (+${section.additions} -${section.deletions})`);
+	}
+	for (const path$4 of excludedPaths) {
+		seen.add(path$4);
+		lines.push(`- ${path$4} (excluded generated file)`);
+	}
+	for (const file$4 of allFiles) if (!seen.has(file$4)) lines.push(`- ${file$4}`);
+	return lines.join("\n");
+}
+function packIntoChunks(pieces, maxTokens) {
+	const chunks = [];
+	let current = "";
+	for (const piece of pieces) {
+		if (current.length > 0 && estimateTokens(current + piece) > maxTokens) {
+			chunks.push(current);
+			current = "";
+		}
+		current += piece;
+	}
+	if (current.length > 0) chunks.push(current);
+	return chunks;
+}
+function splitOversizedSection(section, maxTokens) {
+	const [header, ...lines] = section.split("\n");
+	const hunks = [];
+	let currentHunk = [];
+	for (const line of lines) {
+		if (line.startsWith(HUNK_HEADER_PREFIX) && currentHunk.length > 0) {
+			hunks.push(currentHunk);
+			currentHunk = [];
+		}
+		currentHunk.push(line);
+	}
+	if (currentHunk.length > 0) hunks.push(currentHunk);
+	return packIntoChunks(hunks.flatMap((hunk) => {
+		const piece = [header, ...hunk].join("\n");
+		return estimateTokens(piece) <= maxTokens ? [piece] : hardSplit(piece, maxTokens);
+	}), maxTokens);
+}
+function hardSplit(text$1, maxTokens) {
+	const maxChars = maxTokens * CHARS_PER_TOKEN;
+	const chunks = [];
+	let current = "";
+	for (const line of text$1.split("\n")) {
+		if (estimateTokens(line) > maxTokens) {
+			if (current.length > 0) {
+				chunks.push(current);
+				current = "";
+			}
+			for (let index$2 = 0; index$2 < line.length; index$2 += maxChars) chunks.push(line.slice(index$2, index$2 + maxChars));
+			continue;
+		}
+		const candidate = current.length === 0 ? line : `${current}\n${line}`;
+		if (estimateTokens(candidate) > maxTokens) {
+			chunks.push(current);
+			current = line;
+			continue;
+		}
+		current = candidate;
+	}
+	if (current.length > 0) chunks.push(current);
+	return chunks;
+}
 var system_prompt_default = "You are a strict Git commit message generator. ONLY output the commit message, nothing else.\n\nREQUIRED FORMAT:\n- First line: <type>(<scope>): <subject> (max 50 characters)\n- Types: feat, fix, docs, style, refactor, perf, test, chore, ci\n- Subject: imperative mood, lowercase, no period\n- If body needed: blank line + detailed explanation (wrapped at 72 chars)\n\nSTRICT RULES:\n1. MUST use conventional commits format: type(scope): subject\n2. Subject line MUST be under 50 characters\n3. Subject MUST be in imperative mood (e.g., \"add\", \"fix\", \"update\")\n4. Subject MUST be lowercase\n5. Subject MUST NOT end with a period\n6. MUST provide clear context about WHAT changed and WHY\n7. MUST be specific about affected functions, components, or modules\n8. MUST NOT use vague terms like \"Update\", \"Fix stuff\", \"Changes\"\n9. If including a body, MUST have exactly one blank line between subject and body\n10. Body lines MUST wrap at 72 characters\n\nOUTPUT INSTRUCTION:\nReturn ONLY the commit message. No explanations, no extra text, no markdown formatting.\n\nEXAMPLE:\nfeat(llm): add support for streaming responses\n\nImplement streaming response handling for commit message\ngeneration to improve user experience with large diffs.\nAdds new stream event handlers and updates the API contract.\n\nAnalyze the diff and generate a message following ALL rules above.";
-async function generateCommitMessage(languageModel, instructions, diff) {
+const summarizePrompt = `You summarize parts of a large git diff for a commit message generator.
+Describe WHAT changed, in which files or modules, and any observable intent.
+Be factual and terse. Do NOT write a commit message. Maximum 120 words.`;
+async function ask(languageModel, system, messages) {
 	return (await generateText({
 		model: languageModel,
-		instructions: system_prompt_default,
-		messages: [{
-			role: "user",
-			content: `Changes:\n${diff}`
-		}, {
-			role: "user",
-			content: instructions || "Generate a concise git commit message based on the above instructions and diff."
-		}]
+		instructions: system,
+		messages
 	})).text.trim();
 }
+async function generateCommitMessage(languageModel, instructions, diff) {
+	return ask(languageModel, system_prompt_default, [{
+		role: "user",
+		content: `Changes:\n${diff}`
+	}, {
+		role: "user",
+		content: instructions || "Generate a concise git commit message based on the above instructions and diff."
+	}]);
+}
+async function summarizeChanges(languageModel, toc, chunk$1) {
+	return ask(languageModel, summarizePrompt, [{
+		role: "user",
+		content: `All changed files:\n${toc}\n\nChanges (part of a larger diff):\n${chunk$1}`
+	}, {
+		role: "user",
+		content: "Summarize these changes."
+	}]);
+}
+async function generateCommitMessageFromSummaries(languageModel, instructions, toc, summaries) {
+	return ask(languageModel, system_prompt_default, [{
+		role: "user",
+		content: `All changed files:\n${toc}\n\nSummaries of all changes:\n${summaries.map((summary, index$2) => `Part ${index$2 + 1}:\n${summary}`).join("\n\n")}`
+	}, {
+		role: "user",
+		content: instructions || "Generate a concise git commit message covering ALL parts above."
+	}]);
+}
 function resolveLanguageModel(modelConfig) {
 	const envVarNames = Array.isArray(modelConfig.apiKeyEnv) ? modelConfig.apiKeyEnv : [modelConfig.apiKeyEnv];
 	let apiKey = null;
@@ -197959,6 +198128,10 @@ async function runWithLoading(label, task) {
 		readline.clearLine(process.stdout, 0);
 	}
 }
+const MAP_CONCURRENCY = 3;
+const PROMPT_RESERVE_TOKENS = 800;
+const MIN_CHUNK_BUDGET_TOKENS = 1e3;
+const MAX_TOC_LINES = 500;
 async function mainController(options = {}) {
 	const { git, liveGit } = await getGit();
 	const config$4 = await loadConfig((await git.revparse(["--show-toplevel"])).trim());
@@ -197992,7 +198165,15 @@ async function mainController(options = {}) {
 	}
 	if (finalCommitMessage.length === 0) while (true) {
 		if (forceLLMGenerate) console.log(source_default.yellow("• Using LLM to generate message"));
-		finalCommitMessage = (await runWithLoading("Generating commit message", () => generateCommitMessage(languageModel, config$4.instructions ?? null, diff))).trim();
+		finalCommitMessage = (await runWithLoading("Generating commit message", () => generateMessage({
+			git,
+			languageModel,
+			instructions: config$4.instructions ?? null,
+			diff,
+			files,
+			maxDiffTokens: config$4.maxDiffTokens ?? DEFAULT_MAX_DIFF_TOKENS,
+			perFileCap: config$4.perFileCap ?? DEFAULT_PER_FILE_CAP
+		}))).trim();
 		if (finalCommitMessage.length === 0) throw new Error("The selected model returned an empty commit message.");
 		console.log(source_default.cyan.dim(finalCommitMessage));
 		console.log("");
@@ -198012,6 +198193,34 @@ async function mainController(options = {}) {
 	await liveGit.push();
 	if (config$4.postCommand === "push-and-pull") await liveGit.pull();
 }
+async function generateMessage(options) {
+	const { git, languageModel, instructions, diff } = options;
+	if (estimateTokens(diff) <= options.maxDiffTokens) return generateCommitMessage(languageModel, instructions, diff);
+	console.log(source_default.yellow("• Large diff detected, minimizing"));
+	const { toc, body } = await minimizeDiff(git, {
+		perFileCap: options.perFileCap,
+		allFiles: options.files
+	});
+	if (estimateTokens(toc) + estimateTokens(body) <= options.maxDiffTokens) return generateCommitMessage(languageModel, instructions, `${toc}\n\n${body}`);
+	console.log(source_default.yellow("• Diff still too large, summarizing in parts"));
+	let effectiveToc = toc;
+	let chunkBudget = options.maxDiffTokens - estimateTokens(effectiveToc) - PROMPT_RESERVE_TOKENS;
+	if (chunkBudget < MIN_CHUNK_BUDGET_TOKENS) {
+		effectiveToc = toc.split("\n").slice(0, MAX_TOC_LINES).join("\n");
+		chunkBudget = Math.max(options.maxDiffTokens - estimateTokens(effectiveToc) - PROMPT_RESERVE_TOKENS, MIN_CHUNK_BUDGET_TOKENS);
+	}
+	const chunks = splitDiffIntoChunks(body, chunkBudget);
+	const summaries = new Array(chunks.length);
+	let nextChunk = 0;
+	async function worker() {
+		while (nextChunk < chunks.length) {
+			const index$2 = nextChunk++;
+			summaries[index$2] = await summarizeChanges(languageModel, effectiveToc, chunks[index$2]);
+		}
+	}
+	await Promise.all(Array.from({ length: Math.min(MAP_CONCURRENCY, chunks.length) }, worker));
+	return generateCommitMessageFromSummaries(languageModel, instructions, effectiveToc, summaries);
+}
 var AppUserCanceledError = class extends Error {};
 function handleError(fn$1) {
 	Promise.resolve().then(fn$1).catch((err) => {

package/package.json

@@ -1,6 +1,10 @@
 {
   "name": "gityo",
-  "version": "1.0.10",
+  "version": "1.0.11",
+  "repository": {
+    "type": "git",
+    "url": "git+https://github.com/NazmusSayad/gityo.git"
+  },
   "sideEffects": false,
   "type": "module",
   "scripts": {
@@ -13,7 +17,7 @@
     "json": "tsx ./src/json.ts"
   },
   "bin": {
-    "gityo": "./dist/index.js"
+    "gityo": "dist/index.js"
   },
   "devDependencies": {
     "@ai-sdk/amazon-bedrock": "^5.0.40",